# Some Thanks for Professor Olofsson

I’m halfway through mathematics Professor Peter Olofsson’s essay titled Probability, Statistics, Evolution, and Intelligent Design which originally appeared in the journal Chance.

The first thing I want to thank PO for is stating this early on:

Although [religion] is of interest in its own right, in fairness to ID proponents, it should be pointed out that many of them do not employ religious arguments against evolution and this article does not deal with issues of faith and religion.

The second thing I’d like to thank him for is describing ID as a valid scientific hypothesis in the discussion of the explanatory filter and the flagellum. PO brings up the same argument I’ve always pointed out when he talks about the rejection region for biological applications of the filter and the inability to bound it without doubt. I describe the same problem as that of, I think, a better term also used by Dembski called “probabilistic resources”. How can we ever possibly know some undiscovered pathway for law and chance to have assembled the flagellum doesn’t exist? The short answer is we cannot. We don’t know what we don’t know and we can’t calculate a probability for an unknown. This is in fact employing a valid “chance of the gaps” argument by PO.

The thing of it is there are lots of theories and hypotheses in science that are not provable because it is impossible in principle to rule out the unknown or the unobserved. Karl Popper famously stated how something cannot be provable but remain a valid scientific hypothesis. He illustrated it famously with a then mythical black swan. He gave the hypothesis “There are no black swans in nature.” He said this was a valid hypothesis because while it could never be proven true, that it was impossible to say all of nature was searched and no black swan possibly overlooked, the hypothesis can be disproven (falsified) by the observation of just one single black swan. In the meantime it was a valid hypothesis because it explained the known facts – millions of swans observed and none were black. A black swan in fact was eventually observed and science worked as it should – the black swan hypothesis was falsified.

So we have the ID hypothesis for the flagellum stated as “There are no unintelligent processes which can produce a complex machine (like a flagellum) in nature”. I have stated the ID hypothesis this way many times here. It is a perfectly valid scientific hypothesis which cannot be proven but can be falsified by observing just one unintelligent process producing such a machine. In the meantime we know that intelligence can produce complex machines. You’re reading this on one example of a complex machine which the explanatory filter would also predict is virtually impossible to have come about by law and chance alone.

I just glanced at the second half of the PO essay which addresses Behe’s “The Edge of Evolution”. I trust it rests on the same argument that the failure to observe any novel complex machines arising in the malaria parasite does not prove anything. I agree that it proves nothing. However, that doesn’t mean it doesn’t offer more evidence in favor of the ID hypothesis stated in the paragraph above. ID predicts that no complex machines would originate in even 10^30 reproductive opportunities for mutation and selection. Law and chance handily explains what was observed and what was observed fell far, far short of producing any significant machine-like complexity. What this represents is what Behe describes as probably the biggest real world test of evolution by chance and necessity ever witnessed. It validated microevolution by producing some small useful novelties but failed utterly to produce anything greater than what was predicted by known law and chance. No unknowns showed up. No black swan was observed.

The observations surrounding the malaria parasite could have, at least in principle, falsified the ID hypothesis. But it didn’t. So chalk up a successful prediction for ID.

## 347 Replies to “Some Thanks for Professor Olofsson”

1. 1
Atom says:

Good post. Hopefully PO will stop by to leave some comments as well.

2. 2
Jehu says:

Well done. Economist Mark Perry has a great quote from Lord Kelvin up on his blog today. Here it is

When you can measure what you are speaking about, and express it in numbers, you know something about it. But when you cannot measure it, when you cannot express it in numbers, your knowledge is of a meager and unsatisfactory kind: it may be the beginning of knowledge, but you have scarcely, in your thoughts, advanced to the stage of science, whatever the matter may be. -Lord Kelvin

It seems to me that it is the IDists who are the only ones attempting to lay hold of the mathematics of law and chance that would drive evolution and they are doing so in a fashion that is predictive of actual observable phenomena. The Darwinists on the other hand are still searching for the alchemists stone to make sense of it all. So who is it that is doing actual science?

3. 3
jerry says:

A couple things.

First – PO was always polite and listened and while I didn’t think he agreed with ID and thought naturalistic evolution was the answer to life, he approached his answers based on his expertise which is statistics. No condescension or arrogance. One problem is that we here are not in his league in terms of statistics.

Second – I am interested in what he has to say about Behe’s analysis and how valid it is. I believe Behe’s path is the more fruitful approach to build the case for the lack of statistical resources in the history of the world. I do not agree that if someone comes along and finds one black swan (by the way I just saw one a couple weeks ago in New Zealand) or even several that the jig is up for ID. What has to be done is to show that black swan events are a dime a dozen before ID is undermined. The life on the planet requires tens of thousands of black swan events to happen in order to have gotten to where it is today (probably much higher than tens of thousands of such events).

Thus, when we start to exhaust the examples of reproductive events in the history of life in particular areas we would start to enumerate just how many of these improbabilistic events must have taken place. For example, as genomes get mapped of various animal families or orders it will be possible to see which capabilities of the various species of the order were only possible through improbable combinations during reproductive events or which could have arisen easily through micro evolution given an initial gene pool for the order or class that contained these capabilities. In other words extending Behe’s ideas to multi-celled reproductive events to see if anything has arisen that is unique in a species and if this unique quality is outside the edge of evolution. Or it existed in some gene pool a long time ago and how it is manifested today is through normal micro evolutionary processes.

So it will be interesting to see what PO has to say about Behe as some critics of Behe say his statistical approach is a failure. Somehow I don’t think this is true.

4. 4
Jehu says:

Jerry,

What kind of statistical approach concludes with the argument that just because we have never seen a black swan doesn’t mean a black swan doesn’t exist? It doesn’t seem to me that POs argument, at least as described by DaveScott has anything to do with statistics.

5. 5
ribczynski says:

Jehu wrote:

It seems to me that it is the IDists who are the only ones attempting to lay hold of the mathematics of law and chance that would drive evolution and they are doing so in a fashion that is predictive of actual observable phenomena.

Jehu,

Can you give a few examples of testable predictions yielded by ID’s mathematical approach?

6. 6
Jehu says:

I just breezed through PO’s article and he commits the error of citing to Nick Matzke to counter Behe’s Edge of Evolution. I debated Matzke on the topic of Behe’s Edge and Matzke resorted to the claim that malaria causing parasites had never evolved the ability to reproduce below 68ºF because all of the mosquitoes freeze below that temperature. Unbelievable. Matzke is simply not a credible source -ever.

7. 7
Barry Arrington says:

ribczynski,
Did you not even read the post you are commenting on where DaveScot answered your question before you asked it? You are on probation (translation: keep it up and get the boot from this site).

8. 8
Barry Arrington says:

Jehu, re [6], could you please point me to that quote from Matzke. I am considering putting it on permanent display on the home page.

9. 9
ribczynski says:

Barry,

Yes, I did read the post, and I noted DaveScot’s comment.

However, the prediction he cites is compatible with both ID and NDE. I asked Jehu for examples because I am interested in seeing cases where ID and NDE differ in their predictions.

10. 10
ribczynski says:

DaveScot writes:

How can we ever possibly know some undiscovered pathway for law and chance to have assembled the flagellum doesn’t exist? The short answer is we cannot. We don’t know what we don’t know and we can’t calculate a probability for an unknown.

This is true, and it highlights a fatal flaw in Bill Dembski’s approach to design detection.

Dembski claims that his method is immune to false positives. As he puts it, “Only things that are designed had better end up in the net.”

But without an accurate probability distribution for the feature in question (in this case the flagellum), there is no way to judge whether its probability falls outside of Dembski’s “universal probability bound” — and thus no way to rule out law and chance as explanations for its existence.

11. 11
DaveScot says:

ribczynski

ID predicted that nothing of any significant novel complexity would emerge in 10^30 opportunities for mutation and selection in p.falciparum.

You said this agrees with the prediction of NDE. I disagree with that. NDE predicts something might emerge or it might not. That’s a prediction equally as useful as saying that a coin toss might be heads or it might be tails.

And no, it’s not a fatal flaw in the explanatory filter. There is NO requirement to prove a negative in science. That’s a good thing else the law of gravity would be fatally flawed too as it relies solely on a track record of a very large number of correct predictions without a single known exception. Similarly, although I didn’t mention it, is the law of biogenesis which states that life only comes from life. Again this law isn’t proven but it has been seen to work countless times without a single known exception.

12. 12
jerry says:

“What kind of statistical approach concludes with the argument that just because we have never seen a black swan doesn’t mean a black swan doesn’t exist?”

Inductive reasoning.

ID uses it all the time. We have never seen an example in nature where naturalistic processes have formed a novel functional complex capability, therefore naturalistic process are incapable of forming such capabilities. Behe looked at large numbers of events in single celled organisms to make his claims.

If we ever find such a capability then there will be a counting of just how many we find through naturalistic processes that will determine how viable ID is.

13. 13
jerry says:

“I am interested in seeing cases where ID and NDE differ in their predictions.”

For most of the science of NDE ID has no quibbles. NDE can be divided into macro evolution, for which there is no evidence, and micro evolution for which there is substantial evidence that it has taken place. So ID generally has no quibbles with this part of NDE. However, macro evolution, which requires the formation of novel, complex functional capabilities seems beyond the probabilistic resources of nature and for this case ID and NDE would predict different things.

14. 14
Jehu says:

Jerry,

Inductive reasoning? I don’t think so.

15. 15
Jehu says:

Barry,

I believe this is the link.

http://telicthoughts.com/the-edge/#comment-127937

Matzke’s quote isn’t that juicy because the real stupidity of it can only be appreciated from the context of the debate.

16. 16
gpuccio says:

Jehu:

Thank you for the precious Matzke reference. So now, when I have to discuss his “authority”, I will know what to say…

17. 17
jerry says:

“Inductive reasoning? I don’t think so.”

I think so. You are entitled to your opinion.

From Wikipedia

Inductive reasoning:

“Induction or inductive reasoning, sometimes called inductive logic, is the process of reasoning in which the premises of an argument are believed to support the conclusion but do not entail it; i.e. they do not ensure its truth. Induction is a form of reasoning that makes generalizations based on individual instances.[1] It is used to ascribe properties or relations to types based on an observation instance (i.e., on a number of observations or experiences); or to formulate laws based on limited observations of recurring phenomenal patterns. Induction is employed, for example, in using specific propositions such as ‘This ice is cold, All ice I have ever seen is cold, All ice is cold.’ ”

The Black Swan theory refers to a large-impact, hard-to-predict, and rare event beyond the realm of normal expectations. Unlike the philosophical “black swan problem”, the “Black Swan” theory (capitalized) refers only to events of large consequence and their dominant role in history.

The term black swan comes from the commonplace Western cultural assumption that ‘All swans are white’. In that context, a black swan was a metaphor for something that could not exist. The 17th Century discovery of black swans in Australia metamorphosed the term to connote that the perceived impossibility actually came to pass. Taleb notes that John Stuart Mill first used the black swan narrative to discuss falsification.

Sounds like what I would call inductive reasoning.

And in this vein, ID says naturalistic processes cannot produce novel, complex, functional capabilities. No one has seen a naturalistic processes produce such capabilities, therefore by induction such a process does not exist.

But then there may be a black swan example. However, as I point out above there would have to many, many black swan events before ID would be undermined.

18. 18
bFast says:

Matzke resorted to the claim that malaria causing parasites had never evolved the ability to reproduce below 68ºF because all of the mosquitoes freeze below that temperature.

Oooh, belly laugh!

I live in the sub-arctic — somewhere north of the 60th parallel. In these parts we celebrate summer days that get above 68ºF. However, we’ve got mosquitoes, believe me.

I saw a bumper sticker that read, “There’s not a single mosquito on the Alaska Highway. They’re all married and have lots of kids!”

One hunter reports seeing two mosquitoes packing off a moose. The one mosquito said to the other, “lets hide him quick before the big ones see him.”

I know, there’s bunches of species of mosquitoes, but some of those species are quite capable of living below 68ºF. It is fortunate indeed that malaria hasn’t settled into a colder mosquito species. It is fortunate indeed that the malaria carrying species have not evolved a more cold-loving strain. But mosquitoes certainly thrive below 68ºF. And I am highly entertained that anyone would think that they don’t.

19. 19
idnet.com.au says:

gpuccio excellent!

“Those in the Darwinian field cannot really counter Dembski’s arguments so they use mathematicians or statisticians to try to discredit them with technical and irrelevant objections, while ignoring the obvious hole which has been revealed in their position by the same arguments.”

20. 20
Jehu says:

BFast,

The CDC has diagrams of where the exact species of mosquito that carry malaria live, and the range of the mosquito greatly exceeds that of malaria. The reason for this seems to be that the plasmodium that causes malaria cannot reproduce below 68ºF. Matzke seemed to think that malaria’s failure to spread to cold climates was a result of mosquitoes freezing.

The punch line here is that the malaria parasite has more reproductive events every year than mammals have had in their entire evolutionary history and while mammals have supposedly evolved from tiny shrew like creatures into great ocean going whales, nimble flying bats, large elephants, and space ship building humans, and evolved such novel organs as hair, the cerebral cortex, the inner ear, mammary glands and placentas in fewer reproductive events than the malaria parasite has in a single year – the malaria parasite has failed to even evolve the ability to reproduce below 68ºF after over 100,000 years and exponentially more reproductive events than any mammal that ever existed.

So where we can observe the number of reproductive events that Darwinism claims is necessary to stochastically create the information needed for evolution – we observe no such information being created save those adaptations that can be reached by one or two mutations.

The idea that mammals or any other organism evolved by random chance and natural selection is not change you can believe in.

21. 21
Jehu says:

Jerry,

I don’t see where the black swan argument meets the wikipedia definition of inductive reasoning. The “all swans are white” on the other hand, prior to the discovery of the black swan, was inductive reasoning. PO is making the black swan argument where no black swan has been observed – that is not inductive reasoning it is imagination.

22. 22
Barry Arrington says:

Jehu, I understand that mosquitos live in very cold regions. My uncle was a smoke jumper in Alaska and he said he dreaded jumping near marshy areas (nearly everywhere in the summer I understand), because the mosquitos would torment him so. Out of curiosity, have you done any research on where mosquitos go when it gets really cold, I mean 40 below zero cold, in these regions?

23. 23
ribczynski says:

DaveScot wrote:

ID predicted that nothing of any significant novel complexity would emerge in 10^30 opportunities for mutation and selection in p.falciparum.

You said this agrees with the prediction of NDE. I disagree with that. NDE predicts something might emerge or it might not.

True, and the results did not contradict NDE.

In science, the most interesting experiments are the ones that distinguish between two theories by supporting one and contradicting the other. Eddington’s 1919 eclipse measurements are a classic example. According to both Newton’s and Einstein’s theories, starlight ought to be deflected as it passes through the sun’s gravitational field, but Einstein’s theory predicts a deflection twice as large as Newton’s. Eddington’s measurements confirmed that Einstein was correct, and the experiment generated headlines around the world. (Some questions remain about the margin of error in Eddington’s measurements, but in any case the experiment has been repeated and found to confirm Einstein’s prediction).

ID needs its own Eddington to come up with experiments that support ID while disconfirming NDE.

And no, it’s not a fatal flaw in the explanatory filter.

Actually, it is fatal. Allow me to elaborate.

Dembski claims to have a method — the Explanatory Filter — that allows us to positively identify instances of design.

Ideally, such a filter would be immune to both false negatives and false positives. However, Dembski realizes that false negatives are impossible to avoid, because a designer can always make something that appears to be undesigned. But this is not fatal. The question we are trying to answer is whether there is actual design in the biological world (apart from human genetic engineering, of course). If we miss some cases of actual design, it’s no big deal. In fact, catching even one incontrovertible case of design would be sufficient, though obviously we would prefer more.

False positives are much more of a problem than false negatives. If Dembski’s filter indicates design where there is none — if it cries wolf, in other words — then we can’t trust it, and our question about the existence of design in the biological world remains unanswered.

Dembski understands the importance of avoiding false positives. He writes:

And this brings us to the problem of false positives. Even though specified complexity is not a reliable criterion for eliminating design, it is, I shall argue, a reliable criterion for detecting design. The complexity-specification criterion is a net. Things that are designed will occasionally slip past the net. We would prefer that the net catch more than it does, omitting nothing due to design. But given the ability of design to mimic unintelligent causes and the possibility of ignorance causing us to pass over things that are designed, this problem cannot be remedied. Nevertheless, we want to be very sure that whatever the net does catch includes only what we intend it to catch — to wit, things that are designed. Only things that are designed had better end up in the net. If this is the case, we can have confidence that whatever the complexity-specification criterion attributes to design is indeed designed. On the other hand, if things end up in the net that are not designed, the criterion is vitiated.
— No Free Lunch, p. 24

Also from p. 24:

I want, then, to argue that specified complexity is a reliable criterion for detecting design. Alternatively, I want to argue that the complexity-specification criterion successfully avoids false positives — in other words, whenever it attributes design, it does so correctly. Let us now see why this is the case.

So Dembski clearly understands that false positives must be avoided, and he claims that his method manages to do this successfully.

The problem is that his method depends on having an accurate probability distribution for the feature in question. But as you pointed out, Dave:

How can we ever possibly know some undiscovered pathway for law and chance to have assembled the flagellum doesn’t exist? The short answer is we cannot. We don’t know what we don’t know and we can’t calculate a probability for an unknown.

To summarize:

1. Without accurate probability distributions, Dembski’s method cannot avoid false positives.

2. Dembski concedes that if his method were to generate false positives, it would be useless.

3. To form an accurate probability distribution, we most know all of the possible ways in which the structure in question could have arisen.

4. As DaveScot points out, this is impossible, and so we can never be sure we have an accurate probability distribution.

5. Therefore, according to Dembski’s own criteria, his method is useless.

24. 24
feebish says:

@ Barry #24: Insects are designed with four life stages: egg, larvae, pupae, adult. To exterminate an insect you have to kill all the life stages. That’s what makes it so hard. I did some googling and found that in some mosquito species the adults hibernate through winter, in others the larvae are the ones who survive, and in others it is the eggs/embryos.

@ ribcynski #25: Davescot needs you to lecture him on how science works like he needs a hole in the head. And before you criticize Dr Dembski’s ideas, you should read some of his books.

25. 25
Barry Arrington says:

ribczynski writes: “To form an accurate probability distribution, we most know all of the possible ways in which the structure in question could have arisen.”

Sigh.

ribczynksi says ID proponents must demonstrate perfect and absolute knowledge concerning all possible processes in the universe and then demonstrate that none of the processes that could possibly exist in the whole universe would lead to a Darwinian result before their conclusions are valid. This is – as it is intended to be – an obviously impossible burden, a way to stop the show before it starts.

This has been standard Darwinian fare from the beginning. Indeed, St. Charles started it all 150 years ago in Origin when he wrote: “If it could be demonstrated that any complex organ existed, which could not possibly have been formed by numerous, successive, slight modifications, my theory would absolutely break down.”

Trying to get your opponents to accept an impossible burden of proof is good rhetoric, if all you are trying to do is win the argument and you are not overly concerned about the truth of ther matter. But it is bad logic and even worse science.

26. 26
Barry Arrington says:

Another problem with ribczynski’s formulation is that it proves too much. It leads to obviously false negatives in cases of known design.

Consider once again Mt. Rushmore. A hypothetical investigator who has no knowledge of the mountain other than its bare existence, is trying to decide whether the four faces are the result of design or wind and rain erosion and other natural forces. It seems fairly obvious to the investigator that the faces are designed. But under ribczynski’s formulation the investigator would have reject the design hypothesis unless he were able to exclude with absolute certainty all known AND UNKNOWN natural causes. If these are the decision parameters, the investigator would be compelled to reject the commonsense design conclusion, and he would be wrong.

27. 27
ribczynski says:

Barry Arrington wrote:

ribczynksi says ID proponents must demonstrate perfect and absolute knowledge concerning all possible processes in the universe and then demonstrate that none of the processes that could possibly exist in the whole universe would lead to a Darwinian result before their conclusions are valid.

Barry,

I’m not the one imposing this requirement. Dembski imposes it by requiring an accurate probability distribution as a prerequisite for using his method.

I’m not trying to force people to use Dembski’s method. Indeed, I advise against it, for the very reason you point out: it demands perfect knowledge of all possible means by which the structure could have come about. Without this knowledge, we can’t get an accurate probability distribution. Without an accurate probability distribution, we can’t avoid false positives, contra Dembski.

Barry again:

Another problem with ribczynski’s formulation is that it proves too much. It leads to obviously false negatives in cases of known design.

Barry, did you even read the thread? Dembski himself concedes that false negatives are unavoidable, as I already explained. Here are Dembski’s words again:

The complexity-specification criterion is a net. Things that are designed will occasionally slip past the net. We would prefer that the net catch more than it does, omitting nothing due to design. But given the ability of design to mimic unintelligent causes and the possibility of ignorance causing us to pass over things that are designed, this problem cannot be remedied.

28. 28
ribczynski says:

Jehu,

You owe Nick Matzke an apology.

You wrote:

I debated Matzke on the topic of Behe’s Edge and Matzke resorted to the claim that malaria causing parasites had never evolved the ability to reproduce below 68ºF because all of the mosquitoes freeze below that temperature. Unbelievable. Matzke is simply not a credible source -ever.

I read every comment by Matzke in that thread and he says nothing of the sort.

Um — is someone going to point out that the malaria parasite lives in adult mosquitos, but that in cold regions all the mosquitos (and all other flying insects) die when the temperature hits freezing, and that this provides a perfectly obvious explanation for the distribution of malaria which Behe and all his fans somehow, incredibly, shockingly, astoundingly missed?

Looks like I just did.

Comment by Nick Matzke — July 27, 2007 @ 8:21 pm

He says that the malaria-carrying mosquitos die when the temperature hits freezing.

The malaria parasite lives in the saliva of the adult mosquito. When the mosquito dies, the parasite dies.

29. 29
tribune7 says:

Indeed, I advise against it, for the very reason you point out: it demands perfect knowledge of all possible means by which the structure could have come about. Without this knowledge, we can’t get an accurate probability distribution.

So how do we prove evolution?

30. 30
DaveScot says:

rib

True, and the results did not contradict NDE.

Of course not. How can an outcome contradict NDE when NDE predicts all outcomes? NDE predicts major evolutionary changes happen except for when they don’t happen. I guess you didn’t get the subtle point that NDE made no prediction at all – a theory that predicts everything predicts nothing.

gpuccio & kairosfocus

I’m sorry but I deleted your very long comments. When I have to scroll through pages and pages and pages of a single comment to get to the next one that comment is not going to survive. Next time you need that much space figure out a way to link to the bulk of it offsite. Thanks.

31. 31
DaveScot says:

rib

The explanatory filter is an algorithm not a formula. It is therefore subject to the GIGO rule of algorithms – Garbage In, Garbage Out. The algorithm is as robust as the quality of the input data. It works quite well in tractible situations but I’d tend to agree that something as complex as the flagellum purportedly coming about through a process of mutation & selection over possibly billions of years and trillions of trillions of opportunities for chance to make changes might fairly be called intractible. Problems in quantum mechanics quickly become intractible too when more than a few particles are involved. That doesn’t make quantum mechanics uselss it simply limits the practical application of it.

That said, the scope and reliability of the explantory filter has nothing to do with the scientific validity of the design hypothesis stated congruently with Popper’s black swan hypothesis. To wit, a means devoid of intelligent design must be demonstrated to be capable of forming a complex machine like the flagellum. That is the black swan to be found. Good luck in your search for it.

On Matzke – he was quite wrong:

Temperature

Temperature has a profound influence on the developmental cycle of the malaria parasites. The body temperature of the mosquitoes is directly related to the environmental temperature. Malaria parasites cease to develop in the mosquito when the temperature is below 16ºC. P. falciparum sporozoites can only develop at temperatures above 18ºC. The best conditions for the development of Plasmodia in the Anopheles and the transmission of the infection are when the mean temperature is within the range 20-30ºC.

18ºC is well above freezing and adult Anopheles mosquitos are active at that temperature. The problem is not that the malaria parasite requires adult mosquitos to complete its reproductive cycle. The problem is that the malaria parasite requires adult mosquitos with a body temperature above 18ºC to complete its reproductive cycle.

32. 32
Sal Gal says:

There is no need for a precise probability in design inference. All that is required is an upper bound on the probability of an event. Dembski has stated this a number of times.

Has everyone actually forgotten that Dembski reported that he was working on an upper bound for the probability of emergence of the bacterial flagellum by chance-and-necessity?

33. 33
DaveScot says:

Sal Gal

It has always been my contention that establishing an upper bound for the emergence of a bacterial flagellum by chance & necessity is a problem as intractible as quantum mechanics when more than a few particles are involved. That is the reason I liked Behe’s “Edge of Evolution” which, rather than trying to prove something virtually impossible for chance & necessity to accomplish it rather examined empirical evidence of what chance & necessity was able and not able to accomplish in the real world with a number of opportunities so great as to be far beyond the reach of practical duplication in a laboratory. The result was that the limits predicted by ID, which acknowledges and accounts for microevolution, were the same limits observed in nature. No black swan was observed. PO’s attack on the average mutation rate figure used to predict what chance & selection could accomplish might not be absolutely correct but it nonetheless correctly predicted the observed outcomes.

34. 34
gpuccio says:

Dave,

I hope you have not deleted my review of that paper permanently, because I did not have a copy. If it still exists on the serve, could you please send me a copy of the text by e-mail?

Please notice that I have never been aware of any guideline against long comments. I would have appreciated if you had informed me before deleting the text.

35. 35
allanius says:

The significance of the black swan to Popper was that inductive reasoning is not false as long as it can be falsified. As he himself states, this is actually a new way of applying Kant’s method. Popper envisages knowledge as an organic thing that moves upward to Truth through a process of induction and falsification, leading to a new induction.

Popper wanted to discount Hume’s “pessimism” about scientific method and its reliability as a source of knowledge. Like Kant, Popper only appears to be arguing against induction; in fact, he is arguing that what we call “induction” is synthetic knowledge, capable of further synthesis and change.

The black swan is the negative that makes the positive possible. Hume’s pessimism is not justified because scientific inductions continue to evolve beyond their own limitations through the resistance of the black swan. Theories that have no black swan are incapable of evolving and hence are not (in his view) scientific—including those of Darwin and Freud.

The invocation of the black swan in defense of ID is therefore perfectly reasonable. If unintelligent processes were shown to produce complex machines in the lab, then the induction that is irreducible complexity would be falsified. It is ID that is falsifiable, not NDE, which has no such negative bound. Virtually every observable fact can be made to fit Darwin’s hypothesis. The irony of the theory of evolution is that it cannot organically evolve.

36. 36
crater says:

tribune7:

So how do we prove evolution?

By providing absolute proof of every step along the way. In the case of organisms, showing fossils for each and every transitional along the way. And, in the case of smaller stuff, like the flagellum, prove each and every mutation along the way.

37. 37
Patrick says:

3. To form an accurate probability distribution, we most know all of the possible ways in which the structure in question could have arisen.

Already had this argument and I’m personally not interested in rehashing whether an object’s entire causal history must be known:

Ending of Argument on Casual History

I’d suggest reading the conversation at length but if you do not I will say that Bill would disagree with your point #3, since I asked him directly.

gpuccio,

jerry,

I do not agree that if someone comes along and finds one black swan (by the way I just saw one a couple weeks ago in New Zealand) or even several that the jig is up for ID.

I agree. I’ve said this before in the past but given an already complex system it is possible that non-foresighted mechanisms “might” be able to produce SOME systems comprised of 500 informational bits with a new function. No examples are known yet but I would expect that such cases be limited to a specific interplay of laws–NOT a generalized capability which could produce any type of system. As such, ID theory would then be modified to incorporate known exemptions. Only if a generalized capability was found would ID be completely demolished.

38. 38
ribczynski says:

SalGal writes:

There is no need for a precise probability in design inference. All that is required is an upper bound on the probability of an event. Dembski has stated this a number of times.

SalGal,

Yes and no. In Specification: The Pattern That Signifies Intelligence, Dembski writes:

The fundamental claim of this paper is that for a chance hypothesis H, if the specified complexity ? = –log2[ 120 10 · ?S(T)·P(T|H)] is greater than 1, then T is a specification and the semiotic agent S is entitled to eliminate H as the explanation for the occurrence of any event E that conforms to the pattern T.

This is repeated for every chance hypothesis H. If all of them are eliminated as explanations for E, then Dembski infers design.

Elsewhere in the paper he makes it clear that P(T|H) is a probability, not an upper bound on a probability.

However, I agree with you that it is safe to substitute an upper bound for P(T|H), provided that we are talking about a true upper bound. The worse that can happen is that we will create a false negative, but as we saw earlier in the thread, this is unavoidable anyway.

The more serious problem is that to get a true upper bound on P(T|H), you have to know enough about P(T|H) itself. This can be difficult enough when H is known; if H is unknown, it is impossible, as DaveScot pointed out earlier.

There’s an interesting passage in the paper where Dembski acknowledges the problem, then handwaves it away:

Now, in many statistical applications, the elimination of H (whether on Fisherian or Bayesian or other grounds), does not rule out chance as such but merely invites alternative chance hypotheses H´… Thus, if specified complexity is to rule out chance überhaupt, we must have a good grasp of what chance hypotheses would have been operating to produce the observed event E whose chance status is in question… Probabilistic arguments are inherently fallible in the sense that our assumptions about relevant probability distributions might always be in error. Thus, it is always a possibility that {Hi}i?I omits some crucial chance hypothesis that might be operating in the world and account for the event E in question. But are we to take this possibility seriously in the absence of good evidence for the operation of such a chance hypothesis in the production of E? Indeed, the mere possibility that we might have missed some chance hypothesis is hardly reason to think that such a hypothesis was operating. Nor is it reason to be skeptical of a design inference based on specified complexity. Appealing to the unknown to undercut what we do know is never sound epistemological practice. Sure, we may be wrong. But unknown chance hypotheses (and the unknown material mechanisms that supposedly induce them) have no epistemic force in showing that we are wrong.

This is a remarkable passage. First, Dembski is admitting that there are two ways the EF can yield false positives — erroneous probability distributions and overlooked chance hypotheses — though he denies this elsewhere. Second, he is saying that we can assume that the probability of all unknown chance hypotheses is zero.

In other words, Dembski is saying “It’s okay to assume that nothing you haven’t thought of already will ever happen.”

Not the wisest counsel, is it?

39. 39
tribune7 says:

Probabilistic arguments are inherently fallible in the sense that our assumptions about relevant probability distributions might always be in error.

IOW, ID is falsifiable. How would you falsify evolution?

Appealing to the unknown to undercut what we do know is never sound epistemological practice.

You don’t agree?

40. 40
tribune7 says:

Crater —

So how do we prove evolution? . . . .By providing absolute proof of every step along the way. In the case of organisms, showing fossils for each and every transitional along the way. And, in the case of smaller stuff, like the flagellum, prove each and every mutation along the way.

I could come to accept it with that kind of evidence.

41. 41
Patrick says:

Dembski is admitting that there are two ways the EF can yield false positives — erroneous probability distributions

As Dave already pointed out, GIGO.

and overlooked chance hypotheses

Otherwise known as unknown laws. EDIT: Or at least indirect stepwise pathways with intermediate probabilities that scientists are “somehow” missing.

Second, he is saying that we can assume that the probability of all unknown chance hypotheses is zero.

Not the wisest counsel, is it?

You think it wiser to presume an unknown law is actively at work? That we should somehow incorporate into calculations something we know nothing about? If that’s your personal standard or preference, then why could someone not presume all sorts of supernatural intelligences and boogiemen? 😛

Everyone,

Also, when it comes to the flagellum–or any other biological system–I think there needs to be a distinction made between two related BUT very different things:

1. Calculating the probability of ANY indirect stepwise Darwinian pathway, which presumes a “potential” causal history we can attempt to evaluate. Unfortunately, in practice this has been difficult to do since to date no one can produce even a complete hypothetical pathway.

Here is a recent conversation on UD where the hypothetical indirect pathway of the bacterial flagellum was discussed:

http://www.uncommondescent.com.....ent-289741

The end of this conversation puts the problem in perspective:

http://www.uncommondescent.com.....ent-290187

2. Node 3 of the EF. Calculating the informational bits in a biological system. This deals with a real specification (or function) so it cannot rely on a hypothetical causal history, which–depending on the hypothesis–can have many different associated results. I would think these would be the steps involved:

a) Identify the IC core set of components.

b) We “know” the probability associated with each bit of information, since we can measure the genome’s capacity in informational bits via its quaternary code. The unit of information can be T, A, C or G so it takes 2 informational bits to represent it.

(I’ll explain why I put “know” in quotes later.)

c) Presuming we’ve sequenced the entire genome, and know what information corresponds to producing the components composing the IC core of the biological system, we can then derive the associated informational bits.

Unfortunately, at this point I believe this would produce only a rough estimate. I can look at the code for an IC system in a computer program and since I understand the rules of the system completely I can derive an exact count for informational bits. But scientists as a whole currently do not understand the rules of the biological system. In fact, my personal guess is that any current estimate will likely be lower than reality.

And of course there’s the issue of a designer trying to disguise a signal as noise or us not comprehending how a signal is encoded (which I HAVE done in previous examples, although it was a very simple cipher). So step C might be very difficult in practice (know what information corresponds).

42. 42
crater says:

Trib responds to me:

So how do we prove evolution? . . . .By providing absolute proof of every step along the way.

I could come to accept it with that kind of evidence.

So then why not challenge Barry (in comment 25) when he suggests that perfect information is an unreasonable burden of proof for ID?

43. 43
tribune7 says:

KF, had a good answer too. I wish someone managed to save it.

44. 44
ribczynski says:

DaveScot wrote:

How can an outcome contradict NDE when NDE predicts all outcomes? NDE predicts major evolutionary changes happen except for when they don’t happen. I guess you didn’t get the subtle point that NDE made no prediction at all – a theory that predicts everything predicts nothing.

A contingent science cannot make predictions in every instance. And by the way, that applies to ID (which is contingent on the idiosyncrasies of an unknown designer) as well as every other theory. Even the “Theory of Everything” being sought by physicists won’t predict the weather on August 13, 2045.

As for NDE predicting “everything”, that is obviously false.

NDE does not predict that rabbit fossils will be found in Precambrian strata, to use Haldane’s classic example.

NDE does not predict that sheep are more closely related to paramecia than they are to goats.

NDE does not predict that the Cubs will win the Series this year.

I could go on.

The explanatory filter is an algorithm not a formula. It is therefore subject to the GIGO rule of algorithms – Garbage In, Garbage Out. The algorithm is as robust as the quality of the input data.

True. And if one of the parameters is inherently garbage — like the probability distribution for an unknown process — then the algorithm cannot work.

…I’d tend to agree that something as complex as the flagellum purportedly coming about through a process of mutation & selection over possibly billions of years and trillions of trillions of opportunities for chance to make changes might fairly be called intractible.

As you probably know, Dembski attempts to estimate the probability of that in No Free Lunch. The problem is that he treats the flagellum as a combinatorial object and tries to compute the probability of it assembling all at once — despite the fact that no evolutionary biologist believes that it happened this way.

45. 45
Patrick says:

The problem is that he treats the flagellum as a combinatorial object and tries to compute the probability of it assembling all at once

Again, that objection related to the causal history has been dealt with at length…

46. 46
Patrick says:

So how do we prove evolution? . . . .By providing absolute proof of every step along the way. In the case of organisms, showing fossils for each and every transitional along the way. And, in the case of smaller stuff, like the flagellum, prove each and every mutation along the way.

I would argue that is unreasonable. Why I’ve already discussed in the past (google), and I’m out of time at the moment. But I will say in short that providing evidence for a generalized non-foresighted mechanism would be good enough for me. (BTW, I’m equating your “evolution” to macro-evolution, not common descent or any other distinct but related topic.)

47. 47
DaveScot says:

rib

NDE doesn’t predict when or if significant changes will happen. Since you cannot seem to acknowledge that simple fact you need to move along. Don’t post any more in this thread. Other authors here may continue to entertain your obstinance but I will not.

48. 48
hazel says:

I think ribczynski’s objection is valid. Just because we don’t know – can’t know – all possible causal histories by which something might have come about doesn’t justify assuming that causal history is irrelevant to the question of the probability of something coming to be. Assuming that something is merely a combinatorial object and then computing the probability of it assembling all at once is only justifiable if we have reason to believe that is how it was assembled. If you have reason to believe that something came about through some causal chain, the combinatorial approach is invalid, and it’s not made valid just because you don’t know the causal chain in question.

Saying “we don’t know” how this came about is entirely reasonable; saying since we don’t know we will assume pure chance is not reasonable.

49. 49
Atom says:

hazel and ribczynski:

This is ground we’ve covered before: Jack Krebs and Congregate raised objections about calculating probabilities of objects based on the chance hypothesis. Here is my summary of why the criticism can’t be made relevant:

==================
To sum up for the day,

congregate wrote:

…but you are still assuming that the causal history of the situation is that all six dice were thrown in one roll, and the result determined by chance. Another causal history would result in a different probability.

[A]n adjusted set of the “rules of Yahtzee” are standing in for “laws of nature”. In that scenario, random chance determines how the first roll of the dice goes. After that the rules interact with chance. If two ones are rolled, those are retained, and only the other three dice are rolled. If another one is rolled from among those three, it is retained for the next roll, and so on.

First off, I am not assuming anything about the causal history of how the dice got to be that way; I am simply calculating what the probability of unguided (by intelligence) mechanisms (law, chance and their interaction) hitting a functional configuration. Again, our knowledge of the known laws plugs into this formula.

Using this information, I can then possibly make a tentative inference to design, if my assumptions hold. If they do not hold (for example, maybe I missed a law that aligns dice when thrown) then my inference will be overturned. But the mere possibility of such a future discovery cannot be used as current evidence against my inference.

So you are free to claim that our assumptions do not hold (we are missing a relevant natural law, the search space has more functional configurations than we’re estimating, the search space is smaller than assumed, etc), but simply stating that this tentative nature of the EF is somehow grounds to dismiss it is illogical. All empirical science is tentative.

As for the causal history, we sometimes don’t know it. When we do, we can test the validity of the EF. When we don’t, we can use the EF to give us insight to whether or not that history required Intelligence for a given set of data. In my first example, Jack admitted he didn’t know it (since he guessed the wrong history accidentally) but was still able to conclude design. So notice, the causal history of an object/event does not affect the probability of such an event coming about by unguided mechanisms. We can estimate that based soley on the known laws of nature plus our knowledge of the search space / configuration space.

Get this point, as it is vital: the probability of throwing a die and getting a snake eye is independent of whether or not I actually placed a die on the table in a snake eye position. I did not leave it to chance in the latter case; but that does not affect in any way what the chance probabilities are of it occurring naturally.

If my actions are taken as a part of nature, then you’d have to include it in your probability calculation. But not until you know about my actions and their effects. (Again, no appeals to magical, unknown ordering principles.)

Hopefully that clarifies everything for you guys. If not, please let me know what is still an issue.

================

The posts continue; I think it would benefit those who wish to read the entire exchange, as the object is discussed in great detail.

50. 50
Atom says:

Another post in that thread that sums the issue up:

==================

Jack wrote:

So here we have the important point. If the dice “behaved normally” (1/6 of the time 6 comes up), we can determine the probability of 800 6’s out of a 1000 dice – it comes to about 10 ^ -425, which would classify that result as designed.
But since the dice don’t behave “normally” in that sense, that result really tells us nothing. Until we can figure out the probability of 800 out of 1000 6’s in this particular situation, we have no idea whether the result meets the EF criteria for being declared designed.

Yes, I agree. I emphasize “Until we can figure out the probability of 800 out of 1000 6’s in this particular situation“, because once we do, we are able to figure out the probability.
ID is based on what we know. We know there are no natural laws, chance “forces” or interactions between the two that will cause snake eyes on unloaded dice. Therefore, we can infer design (cheating) on dice rolls. The same is true for your example: once we know how your machine affects outcomes over many trials (as good scientists we do the proper runs and stats on them), we can also infer design on your set-up.
Cutting to the heart of the matter (since I’m guessing this is where this is all headed), we know that DNA is contingent and there is no ordering principle forcing particular base pair sequences. If there were, the information carrying capacity of DNA would be reduced proportional to the loss of contingency. (Information Carrying capacity in a channel directly related to the contingency of the possible sequences.)
Again, you are free to argue against contingency in biology, but to do so you must show what the ordering principle is and how it changes probabilities quantitatively. Until then, it is an unknown law and we cannot evoke it in defense of a position. If you can elucidate it in detail, then we can include it in our probability estimation.
Either way your position cannot be defended.
================

And Jack Krebs response:

================

I know the rejoinder I will get here is “how”? I don’t know how. I am just saying that the argument that the absence of an ordering principle that explains the orders of the base pairs does not mean there is no possible explanation whatsoever involving the interplay of law and chance. As I have pointed out, we explain many things quite successfully even though there is no one principle responsible for the things in question. One can argue that no possible interplay of laws and chance could have created the ordered base pairs, but one should be clear that that is quite different than arguing that the absence of an underlying principle makes an explanation for that order impossible.

===============

I’ll let the two posts speak for themselves. – Atom

51. 51
Jehu says:

ribcynskichev, you wrote Jehu,

You owe Nick Matzke an apology.

You wrote:

I debated Matzke on the topic of Behe’s Edge and Matzke resorted to the claim that malaria causing parasites had never evolved the ability to reproduce below 68ºF because all of the mosquitoes freeze below that temperature. Unbelievable. Matzke is simply not a credible source -ever.

I read every comment by Matzke in that thread and he says nothing of the sort. …

He says that the malaria-carrying mosquitos die when the temperature hits freezing.

The malaria parasite lives in the saliva of the adult mosquito. When the mosquito dies, the parasite dies. Thank you for the laugh. Are you Nick Matzke using a nom de plume or are all Darwinists that stupid?

ROFLMAO

52. 52
Jehu says:

ribcynskichev,

Sorry, I got so giddy trying to deliver a clever coup de gra to your unbelievably stupid post I forgot to check my formatting.

But let me explain it real simple so you can understand. The malaria parasite does not exist in cool climates because it cannot reproduce below 68ºF. This has nothing to do with freezing mosqiutoes. Matzke’s stupidity is only exceeded by the arrogance with which he made his argument.

The context of Matzke’s comment was the issue of why malaria has not evolved the ability to reproduce below 68ºF. After all, the malaria parasite has more reproductive events in one year than mammals have had in their entire existence. Why can’t malaria, in the same amount of reproductions that mammals were able to evolve hair, the placenta, the cerebral cortex, the inner ear, and mammary glands and turn into bats, whales, tigers, elephants, and humans – even evolve the ability to reproduce below 68ºF. Why is it that where we cannot observe the reproductive events we are told that dramatic innovation has occurred but where we can observe the same number of reproductive events we see only trivial change?

Matzke’s answer is that it is because mosquitoes freeze. Which is manifestly false.

53. 53
tribune7 says:

Crater–So then why not challenge Barry (in comment 25) when he suggests that perfect information is an unreasonable burden of proof for ID?

I wasn’t demanding perfect information for evolution but if provided I would accept it. Would you accept ID with perfect information?

More relevantly, would you agree that with existing information — what we know now — ID is a much better explanation than NDE?

54. 54
gpuccio says:

Atom (#49 and 50):

Wonderful clarifications! You say:

“Cutting to the heart of the matter (since I’m guessing this is where this is all headed), we know that DNA is contingent and there is no ordering principle forcing particular base pair sequences. If there were, the information carrying capacity of DNA would be reduced proportional to the loss of contingency.”

That’s the reason why, as I had argued in my deleted post, it is absolutely correct, methodologically, and from our knowledge of the biological context, to assume a uniform distribution for the search space of protein sequences. Even if the distribution were not strictly uniform, it could never be significantly different, and that would not in any way empirically affect the results, least of all in a matter where we are discussing search spaces whic are usually of the order of 10^300 or more for most proteins.

PO’s arguments are at best specious technicalities without any relevance, at worst (see the part regarding Behe) evident examples of incorrect methodological approach to the biological context.

55. 55
gpuccio says:

Jehu:

“Are you Nick Matzke using a nom de plume or are all Darwinists that stupid?”

Well, don’t be so extreme… Maybe the distribution of the phenomenon is not completely uniform 🙂

56. 56
kairosfocus says:

On receiving an offline request for a copy of the — deleted on grounds of “length” — post that originally and briefly appeared at 30 from a fellow commenter, and on seeing remarks addressing it overnight, I have decided to put the post up in my own blog with some prefatory remarks.

PPS: R, you will find some remarks on Bayesian vs Fisherian inferential statistics here, and on the chance- necessity- agency causal factors here, that I believe will prove helpful in clarifying the balance of the matter on the merits.

57. 57
DaveScot says:

Use a bit of common courtesy in the length of comments. If someone has to scroll through more than two screens to get past it to the next comment it’s probably too long.

58. 58
PaV says:

It seems no one is addressing Prof Olafsson’s paper.

Here is where I see fault in his paper.

Olafsson appears to find it a very problematic that Dembski’s “filter” doesn’t “rule out all chance explanations; that is, we need to rule out both HsubO and HsubA.”

Well, let me point this out:
(1) Does Olafsson ever tell us what HsubA is? Does he ever specify it? No. He only gives us a probabilistic description of what HsubA is: namely, “p>1/2″. Does this mean p=.55, or that p=.6? IOW, p>1/2, without an actual hypothesis attached and stipulated, is altogether meaningless.

2.) Olafsson says we must “rule out both HsubO and HsubA”. If we rule out a chance hypothesis that the probability of something occurring is less than one half, and then we go on to rule out the alternative chance hypothesis which carries with it the the probability of said event is greater than one half, then I can only conclude the event never happened. (Let’s note that the probability of less Democrats than Republicans doesn’t exist in the situation we’re considering.) But, indeed, the Caputo case did happen. So, given what we know was supposed to be the method employed by Caputo, there are only two hypothesis that need to be considered: it happened by chance, or it didn’t

While more or less conceding Dembski’s argument about Bayesian methods and their inherent problem in trying to assign prior probabilities, he bristles when Dembski says that Bayesian inference is “parasitic on the Fisherian approach.” He then goes on to cite the Caputo case. In Olafsson’s defense, he is citing Dembski’s argument from the book Design Inference (I don’t own a copy), but in No Free Lunch I think Dembski is perfectly clear in what he means by Bayesian’s being “parasitic” and he does so specifically using the Caputo case. On p. 107 of NFL he writes: “Such [likelihood=Bayesian] analyses do not go far enough. To see this, ask yourself how many times Caputo had to give the Democrats the top ballot line before it became evident that the was cheating. Two for the Democrats, one for the Rebpulicans? Three for the Democrats, one for the Republicans? Four for the Democrats, one for the Republicans? Etc. On a likelihood [Bayesian] analysis, any disparity favoring the Democrats provides positive evidence for Caputo cheating. But where is the cutoff? There is a point up to which giving his own Democratic party the top ballot line could be regarded as entirely innocent. There is point after which it degenarates into obvious cheating and manipulation. . . . Specified complexity—and not a likelihood analysis—determines the cutoff. Indeed, specified complexity determines when advantages appparently accruing from chance can no longer legitimately be attributed to chance.”

IOW, the “cutoff” is a “rejection region”, and without resort to such “cutoffs”, Bayesians cannot make confident enough conclusions. (Dembski points out that the court did not find Caputo guilty—presumably because the improbability was not sufficiently high). Thus, the “rejection regions” are implicit in Bayesian analysis, and such analysis is, therefore, “parasitic” on the Fisherian approach” (which provides such needed ‘rejection regions’)

59. 59
kairosfocus says:

GP has given me permission to post his former comment, originally no 16, which may be seen at the following

60. 60
Atom says:

Professor Olofsson tried to post here in response, but he is still blacklisted.

Mods?

(I think he should be given a chance to respond to criticisms. Plus, he’s one of the most respectful dissenters I know.)

61. 61
Patrick says:

Not sure why Olofsson was blacklisted in the first place so while he now should be able to respond he is still in moderation.

62. 62
kairos says:

jerry #52

Why is it that where we cannot observe the reproductive events we are told that dramatic innovation has occurred but where we can observe the same number of reproductive events we see only trivial change?

In your post you have just stated what I would want to write about. I also agree with Dave’s consideration about the EoE book. While NDE supporters have objected to Dembski’s argument on the base of “what is unknown”, they cannot object to “what is actually known” but using non pertinent arguments (such as A. Smith’s argument about HIB evolution) or the usual ad hominem attacks.
That’s very good news for the future of ID

Matzke’s answer is that it is because mosquitoes freeze. Which is manifestly false.

63. 63
kairos says:

#54 gpuccio

assume a uniform distribution for the search space of protein sequences. Even if the distribution were not strictly uniform, it could never be significantly different, and that would not in any way empirically affect the results, least of all in a matter where we are discussing search spaces whic are usually of the order of 10^300 or more for most proteins.

I completely agree on this. The fact is that the chemical support of nucleotides bases is pretty independent on which pair (AT,TA,CG,GC) colud actually stay in each given position. This means that really the genetici information in DNA is at an *upper and independent level that the chemical level* exactly as the bit information whitin a computer memory is at a upper level and independent on the kind of memory that does store it (ROM, RAM, Hard disk, DVD and so on.
IMHO this fact provides scientific evidence about a nearly uniform probability distribution.

64. 64
Prof_P.Olofsson says:

I’d like to respond to the criticism offered by gpuccio and others but first I need to make sure that I can post. If this comment is posted, I’ll go ahead.

P.Olofsson

65. 65
kairos says:

66. 66
Prof_P.Olofsson says:

gpuccio writes:

So, again, PO’s objections have some theoretical grounds, but are completely
irrelevant empirically, when applied to the biological systems we are
considering. That is a common tactic of he darwinian field: as they cannot
really counter Dembski’s arguments, they use mathematicians or statisticians
to try to discredit them with technical and irrelevant objections, while
ignoring the evident hole which has been revealed in their position by the
same arguments. PO should be more aware that here we are discussing
empirical science, and, what is more important, empirical biological
science, which is in itself very different from more exact sciences, like
physics, in the application of statistical procedures.

Dembski’s arguments are mathematical and statistical so it makes perfect sense that somebody with expertise in these fields examines them. In fact, he has claimed on this blog that the record of mathematical criticism of his work is unconvincing, that his critic Jeff Shallit has no expertise in probability theory, and that another critic, Richard Wein, has only a bachelor’s degree in statistics. My criticism is about the applicability of the filter in biology, not about its theoretical foundation.

67. 67
Prof_P.Olofsson says:

I’ll submit comments bit by bit, so they won’t run too long. The first was just posted.

68. 68
Prof_P.Olofsson says:

gpuccio writes:

In general, he affirms that Dembski does not explicitly state how to define the rejection region. Let’s begin with the case of a single functional protein. Here, the search space…

Your discussion about “search space” is not relevant to Dembski’s “shopping cart” calculations in NFL where he assumes that all the needed proteins exist and calculates the probability that they fall randomly into place to form the flagellum. I may not know much about biology but I dare say that such a mechanism has never been proposed by any biologist. In other words, he successfully rules out a chance explanation but one that is irrelevant to evolutionary biology which, after all, is his target.

69. 69
tribune7 says:

I may not know much about biology but I dare say that such a mechanism has never been proposed by any biologist.

If something is IC what other mechanism is there? I mean other than (cough cough) you know, the “d” word?

Granted, evolutionary biologists claim the flagellum is not IC which leads us to consider the reasonableness of the evolutionary pathways they claim are possible albeit they insist that demonstrations of such pathways are not necessary.

70. 70
Prof_P.Olofsson says:

gpuccio writes:

… practically in all biological and medical sciences, the statistical approach is Fisherian, and is based on the rejection of the null hypothesis. So, Dembski is right for all practical applicatons.

It is true that hypothesis testing is still the most frequently used methodology, but Bayesian methods have gained ground with the development of faster computing. Pharmaceutical companies and medical centers increasingly use Bayesian methods in clinical trials, microarray analysis, etc. At any rate, these are two different approaches with their own strengths and weaknesses, and most statisticians take a pragmatic view on which one to use in any particular application.

That still doesn’t make Dembski’s characterization of Bayesian methods correct. In my article, I show how a typical Bayesian analysis of the Caputo case would be done and it directly refutes Dembski’s incorrect claim about the event E*.

71. 71
Prof_P.Olofsson says:

gpuccio wrote:

PO makes a rather strange affirmation: “A null hypothesis H0 is not merely rejected; it is rejected in favor of an alternative hypothesis HA”. That is simply not true,…

Yes, it is true. The very setup of (parametric) hypothesis testing is that the null hypothesis specifies a subset of the parameter space and the alternative hypothesis is a subset of the complement thereof. Setting such technicalities aside, if you think I’m wrong, you should be able to refer to at least one medical or biological research article to back up your claim.

72. 72
Prof_P.Olofsson says:

tribune7, [69].

That’s for the evonutionary biologists to answer. I’m merely pointing out that Dembski is testing one chance hypothesis of his own choice, thereby concluding that every other chance hypothesis is ruled out as well.

73. 73
Patrick says:

Prof Olfsson,

I’ll leave the debate to others like gpuccio since I’m short on time. But I’ll ask one question.

In your opinion, what should be done if other chance hypotheses (indirect stepwise pathways in biological terms) are unknown? As in, when would you personally consider a design inference to be validated by the available data?

74. 74
tribune7 says:

I’m merely pointing out that Dembski is testing one chance hypothesis of his own choice,

He tests the only chance hypothesis and falsifies it.

The only way to show that the flagellum is something that evolved is to mix necessity in.

The only necessity offered by evolutionary biologists is natural selection. No details are provided as to the pathway using this necessity to allow us to consider the probability of it happening.

75. 75
gpuccio says:

Prof_P.Olofsson:

First of all thank you for considering my objections and for answering. I appreciate it very much.

It is my duty to comment on your answers. I will try too to stay brief and answer in parts.

Here I would like to give a preliminary impression which can perhaps expalin in a more general way my reaction to what you are saying now.

I want to state again that I don’t discuss your competence in your field. I am a medical doctor with some experience in medical statistics, but you certainly are more qualified in discussing statistical issue in themselves. That’s why I have focused more on the empirical aspects of the application of your discussions to biology, and not on technical issues in themselves.

In your post #66 you say:

“Dembski’s arguments are mathematical and statistical so it makes perfect sense that somebody with expertise in these fields examines them.”

I agree with you, and it is not my role to enter in that kind of confrontation. I leave that to Dembski and other mathematicians. But, as you say:

“My criticism is about the applicability of the filter in biology, not about its theoretical foundation.”

On that, I feel that we can discuss.

76. 76
Prof_P.Olofsson says:

Patrick[73]:

In your opinion, what should be done if other chance hypotheses (indirect stepwise pathways in biological terms) are unknown?

I can’t see how we could do anything along the lines of the explanatory filter unless we have a reasonably clearly specificed chance hypothesis.

77. 77
gpuccio says:

Prof_P.Olofsson:

Still another general remark, before going into the details. And, I think, an important one.

I have read your answers up to now, and I will discuss them, but I must say that I feel a little bit sidetracked by them. My general impression (maybe I am wrong) is that you are answering as though in you article you were only criticizing some details in specific parts of Dembski’s books, and not his general thought. My impression, instead, in reading your original article, was that you were trying to criticize the foundations of ID theory, at least for what concerns its statistical and mathematical arguments. You final conclusion:

“Probability and statistics are well developed disciplines with wide applicability to many branches of science, and it is not surprising that elaborate probabilistic arguments against evolution have been attempted. Careful evaluation of these arguments, however, reveals their inadequacies.”

seems to confitm that such was the purpose of your essay, and I am sure that most, if not all, of the readers have thought the same thing.

I am making this premise because, in answering your comments, I will always make reference to the general theory of ID, as expressed in the works of Dembski and Behe especially, and not only to the specific examples you make. That’s important, becasue you are using those examples to “falsify”, or at least strongly criticize, the whole theory.

In other words, your attitude is not simply that the general perspective of Dembski and others could be acceptable, but you are criticizing some particular issues. It seems that you are criticizing the whole approach. So, I think we should both keep that in mind in the following discussion.

78. 78
gpuccio says:

Prof_P.Olofsson (#68):

“Your discussion about “search space” is not relevant to Dembski’s “shopping cart” calculations in NFL where he assumes that all the needed proteins exist and calculates the probability that they fall randomly into place to form the flagellum.”

Here is an immediate application of what I was saying in my previous post. I was not criticizing specificall your objection to that passage in Dembski (which, at present, I don’t remember in detail). I was criticizing that objection in the light of the general theory of ID. And my discussion about search space is absolutely relevant. I have made the example of a single protein because it is the simplest to deal with. The concept of search space is fundamental in all ID reasoning, up to the last papers by Dembski amd Marks. Are you saying that my discussion about search space is correct, but that you were only objecting that Dembski is wrong in calculating the search space of the flagellum once all the proteins are already there?

If I should calculate the search space of the flagellum (but why everybody concentrates only on the flagellum? We have tons of CSI and IC in biological systems) I would use one of two methods:

a) I would calculate a minimum limit for search space, starting form scratch (the search space of each protein plus a reasoble esteem of the search space of the higher organization of the system)

or

b) If I should decide to consider the rather imaginative hypothesis by Matzke et al. that the flagellum may have arisen by “cooption” from the proteins of TTSS, I would reason this way: let’s suppose the TTSS is there, and we have to arrive to the flagellum; then we start computing the search space for the necessary transitions (the whole search space for new proteins, which are not in TTSS, plus the search space for the mutations necessary to transform the suppsoed precursors in the final proteins; let’s remember that the proteins in TTSS are not the same as those in the flagellum: some of them are more or less homologous; others are not); then we should consider the variations in organization, including different site in the genome, different regulation, different timelines in transcription, control of assemblage, and so on. All that is certainly hugely increasing the search space.

You may say that it is impossible to compute all that. I would agree, but it sould be done as our knowledge increases. And, anyway, we are not dealing with the necessity of a precise computation: we can be satisfied with a minimum limit.

A minimum limit is easy to find: even a single new protein is enough for it.

And, finally, it’s not me who am saying that I have a model for the unguided evolution of the flagellum; it’s the darwinist who say that. They say more: they say that such a model has repeatedly been shown, and that it has falsified Behe’s concept of IC.

That’s not true. Cooption is not a falsification of IC, unless it is supported by a credible model. Otherwise, it is only propaganda.

That credible model does not exist. Even if we admit that the evolution of the flagellum from the TTSS is admissable in a biological context (and, as you know, there are serious objections to that, and not all by IDists), still the reality is: TTSS is not a flagellum. Those who affirm that the transition is possible, has the duty to build a credible model of how the transition may have happened. Then we can see if it is possible to compute the serach space and target space of he transition, to verify if the model is statistcally acceptable. All these things should be done by darwinists. They are proposing the supposed model. They are declaring that the model exists, and is already demonstrated. Until and unless they support their affirmations with something more then void propaganda, the IC of the flagellum remains an absolutely sound scientific concept.

79. 79
gpuccio says:

Prof_P.Olofsson (#68):

Again, it’s a problem of perspective and of what you are trying to say. If you are just saying that you don’t approve of Dembski’s criticism against Bayesian methods, I will leave that to you and Dembski. And you know well that you are not the only ones who are debating that. I have nothing against Bayesian methods, and I am not interested in a discussion about Fisherians against Bayesians. The same is probably true for most readers here (and almost everywhere).

But if you are using those techinical discussions to argue against ID, that’s certainly not correct. You admit that “It is true that hypothesis testing is still the most frequently used methodology”, which was exactly my point. In other words, for those who are not familiar with the problem, Dembski and all of us in ID are using for our scientific reasoning exactly the same approach which is by far the most used in all sciences.

So, why discuss Bayesian methods as though they were in some way a criticism to ID? They are not. Or if they are, they are a criticism to all modern science.

80. 80
Prof_P.Olofsson says:

gpuccio,

On your very last question, the context is that Dembski has been criticized from a Bayesian point of view (arguing that we need to ask how likely a hypothesis is in the light of data instead of the converse). In his “Elimination vs comparison” chapter, he counters the criticism and whereas some of it is valid, some is not. I brought it up because it might be of interest to readers of Chance which is where my article was published. It has no particular relevance to the UD bloggers other than to point out that Dembski isn’t always right.

81. 81
Prof_P.Olofsson says:

gpuccio,

I forgot to say, I will get to your other points later but right now I don’t have time.

82. 82
gpuccio says:

Prof_P.Olofsson (#71):

I will try to express more in detail what I mean. H0 is the hypothesis that our data can be explained by randomness, and need not a special explanatory cause. Being a randomness hypothesis, it can be tested in respect to its probability, in our frequentist scenario. The mathemathical analysis gives us the p, the probability of our data assuming the null hypothesis as true. According to the value of p, and to a conventionally defined rejection region, we decide to reject or not reject H0 (and accept the inherent alfa or beta error).

My point is that, if you define HA as simply non-H0, that is the logical negation of H0, in other words the simple affirmation of non randomness, then your statement is true: rejecting H0 logically implies affirming HA.

But that’s not what ususally happens, at least not in medicine or biology. Rejecting H0 is accomplished to affirm HA, which is not only an affirmation of “non-H0”, of non randomness, but of a specific causal model. The proposed causal model is not necessarily the only possible causal model. Indeed, very often it is only the causal model that is supposed to be “the best explanation”. Affirming HA, then, implies another level of possible error: that the cause of the inferred non randomness may be different form what we suppose. In other words, our methodological approach can be biased, or simply incorrect, and our explanation is not the correct explanation. In other words, affirming non-randomness is one thing, affirming a specific causal model is another thing. The affirmation of a causal model is a methodological problem: our model is supported by statistics at one level, and by its methodological credibility at another level.

You are right that ID uses hypothesis testing in a diffrent context: ID uses statistics to reject the hypothesis of randomness because that hypothesis is an integral part of the darwinian model (not the only part, obsviously: but still the part of the model which assumes randomness as a generative engine of variation must respect statistical restraints). In rejecting the random hypothesis for the pertinent contextx, ID falsifies the current theory.

But the proposal of design as an alternative theory does not “logically” derive from that. It “empirically” derives from that. In other words, the reasoning is simple:

a) the current hypotheses for the generation of CSI in biological information, which incorporate random models, are inconsistent, because those random models (if and when they are detailed) are incompatible with statistical laws.

b) design (an empirical, observed event) can explain CSI and therefore biological information

c) we have at present no other reasonable explanation (no other credible HA)

d) therefore, design is at present the “best empirical explanation” for biological information.

That is a methodologically correct affirmation of the design hypothesis, supported by very strong (I would say extremely strong) statistical arguments.

83. 83
gpuccio says:

Prof_P.Olofsson (#80 and 81):

First of all, please take all the time you want to answer. I will take a rest too now.

I appreciate that you have clarified the original context of your article, and therefore the context for the discussion about Bayesian methods. I am happy that you agree that that is not a significant criticism to ID.

Finally, I can certainly agree that “Dembski isn’t always right”. I have great respect, admiration, and even affection for Dembski, but infallibility is usually too heavy a burden to impose on anybody.

And without subtracting a comma to Dembski’s importance for ID, ID is certainly much more than the thought of a single man.

84. 84
Patrick says:

Professor Olofsson,

I can’t see how we could do anything along the lines of the explanatory filter unless we have a reasonably clearly specificed chance hypothesis.

Just so I understand you, what you are claiming is that if evolutionary biologists cannot conceive any Indirect Stepwise Pathways that would invalidate Dembski’s design detection method (at least in regards to biology)?

85. 85
Prof_P.Olofsson says:

Patrick,

It would not invalidate the method, it would just mean there is nothing to test.

86. 86
kairos says:

#78 gpuccio

Prof_P.Olofsson (#68):

If I should calculate the search space of the flagellum (but why everybody concentrates only on the flagellum? We have tons of CSI and IC in biological systems) I would use one of two methods:

Precisely. Moreover I strongly think that, although th flagellum is the iconic example for ID, the real battle for disproving NDE will be on very very basic examples, possibly with IC bit non necessarily. After alla let us see the very very simple example of malaria plasmodium. I think that NDE will be defeated empirically “on the field”.

b) If I should decide to consider the rather imaginative hypothesis by Matzke et al.

A minimum limit is easy to find: even a single new protein is enough for it.

I completely agree. Darwinists such as NM can speculate freely and without any constaint just because the example yo are speculating to is too much complex for any quantitative evaluation.

BUT

We can wait them on the single details. If NM, or whiever else, has argued a hypotetical indirect pathway, we can ask him:

“OK Nick, let’s just for a moment suppose that this could be true and that the IC structure B did evolve from the structur A through the large number of evolution steps that you have speculated:
A->s1->s2->…->b
But now let’s see the details and let’s consider only one of these little steps, let’s say s23->s24. Let’s empirtically study what’s the chance that this transition could have occurred”

If the overall chance for s23->s24 transition (taking into account an upper bound for the maximal number of organisms involved in the transition) would be estimated as extremely low, let’say 10^-100, the whole indirect pathway would be disproven scientifically, for its cumulative chance would be
P=P1*P2*…*P23*…*Pn

Do you agree Prf. Olofsson?

And, finally, it’s not me who am saying that I have a model for the unguided evolution of the flagellum; it’s the darwinist who say that. They say more: they say that such a model has repeatedly been shown, and that it has falsified Behe’s concept of IC.
That’s not true. Cooption is not a falsification of IC, unless it is supported by a credible model. Otherwise, it is only propaganda.

Completely agree. It’s incredible that evolution is the *only* scientific field in which the burden of the proof is reversed in this way

87. 87
tribune7 says:

Dembski claims that his method is immune to false positives. As he puts it, “Only things that are designed had better end up in the net.”

Again, he doesn’t claim his method is “immune” to false positives.

He says that if his method yields a false positive it fails.

IOW, it’s FALSIFIABLE!

IOW it’s SCIENCE!!

88. 88
Prof_P.Olofsson says:

tribune[88],
You are confusing concepts here. The filter is a method, you can’t say that a method is “falsifiable” or “science.” Or would you claim that ID has been falsified if the filter fails?

89. 89
Prof_P.Olofsson says:

gpuccio[82],
First, I don’t know what you mean by design being an “empirical observed event.” Second, when you start talking about “reasonable” and “credible” explanations, you are leaving the realm of the eliminative filter and, ironically, “thinking Bayesian.” Note how you rule out one chance hypothesis, the uniform one, by calculating a probability and others by claiming they are unreasonable. I might just as easily claim that design hypotheses are unreasonable and infer an alternative, unspecified chance hypothesis as the best explanation.

90. 90
tribune7 says:

The filter is a method, you can’t say that a method is “falsifiable” or “science.” Or would you claim that ID has been falsified if the filter fails?

The claim that design could be detected using this method would be falsified. Why wouldn’t it be?

Now, faith of course, remains but faith and ID are not related.

91. 91
Prof_P.Olofsson says:

tribune7,
I’m just pointing out that you need to distinguish between “science” and “scientific method.”

92. 92
tribune7 says:

Rib, Dembski says his method won’t give false positives. You seem to be saying that this method is impossible to test because we don’t have perfect knowledge, even though we can put this method to practical application since designed and non-designed events occur and we use this method now with regard to these events.

If perfect knowledge is a requirement for a scientific proposal to be taken seriously, Newton’s laws should have been ignored.

And would you apply that standard to Darwinism?

93. 93
tribune7 says:

I’m just pointing out that you need to distinguish between “science” and “scientific method.”

Professor, are you in agreement that Dembski’s method is a scientific method?

94. 94
IDskeptic says:

Newtown’s laws are not advanced by ‘not this, therefore that’, though, like ID is.

95. 95
Prof_P.Olofsson says:

tribune[95],
Sure, insofar as it’s hypothesis testing.

96. 96
tribune7 says:

Professor, so how is ID not science?

97. 97
Atom says:

PO and gpuccio, do you mind if I enter the conversation?

98. 98
tribune7 says:

I’m saying that it’s unreliable unless the probability of the non-design hypotheses is known with sufficient accuracy.

It’s very reliable in practice, further it rules out chance entirely. To reject it requires an appeal to undiscovered law.

When you attempt to demonstrate design via Dembski’s method, you’re actually trying to validate a negative hypothesis: “This object could not have been formed through chance and necessity.”

While, the filters test for chance and necessity, that only serves to increase the burden on the determination for design.

The positive hypothesis considers the complexity and specificity of the object.

Something to ponder, how would ID fair without the filters? Apply it to computer code or a written language. It works.

99. 99
tribune7 says:

If posing testable hypotheses makes a field a science, then ID is a science. But in that case so are astrology and phrenology.

And Darwinism 🙂

Astrology and phrenology were treated in a scientific manner — and considered seriously by smart people, and for awhile considered as sciences– until they were falsified.

Has ID? Darwinism?

Something to remember too is that taking phrenology, and especially astrology, and even Darwinism seriously has contributed to real advances in understanding nature, although in the case of Darwinism quite a bit of pain and suffering went along with it.

Galileo and Tyco Brahe both put stock in astrology.

100. 100
Prof_P.Olofsson says:

Atom,
Stay away, you’re trouble! 🙂

PO

101. 101
Atom says:

PO,
🙂 True enough.

102. 102
gpuccio says:

Atom:

You’re welcome! We need some more trouble!

103. 103
tribune7 says:

In the vernacular, when someone says that “X is science” or “X is scientific”, they typically mean that it is well-supported by evidence and accepted in the scientific community — in other words, that it’s good science.

Again, no. When someone thinks of science they think of observation, quantification and testing to falsify (experimentation). Consensus has — or should have — absolutely nothing to do with it. In fact, science by consensus is recognized as very bad science.

Now ID has quantified observations — the integrity of which really should not be questions — and its conclusions have not been falsified.

104. 104
tribune7 says:

When someone thinks of science they think of observation, quantification and testing to falsify (experimentation). . .Give me a break, tribune.

You don’t agree with this?

We’ve all heard someone say something like “I don’t believe in astrology. It’s not science.”

I don’t believe in astrology but that’s because its claims have been falsified and my religion teaches me to shun superstition.

Now, are you saying you don’t believe in astrology simply because someone told you it wasn’t science? If this person told you jumping off bridges was science would you go ahead and do it? 🙂

Suppose he said there was a scientific consensus that you should jump off the bridge? 🙂

105. 105
tribune7 says:

For example, Dembski’s attempt to demonstrate the design of the flagellum fails because he treats it as a combinatorial object without considering any stepwise pathways

And what are these stepwise pathways?

It’s absolutely clear: once chance and necessity have been ruled out, design is assumed.

Rib, I know it’s not the point you are trying to make and it doesn’t address the point I’m going to make with regard to the positive aspects of CSI, but once you rule out chance and necessity, um, what should you assume?

So the hypothesis of design is still a negative hypothesis, but the negativity is hidden in the calculation of CSI.

No. Design is real — hopefully you aren’t going to say it isn’t — and has certain characteristics. These characteristics include a particular amount of complexity along with specificity of function. It’s a very positive hypothesis.

Now, you can say it will fail because that there are things that are not designed that have these characteristics. And then of course, you will bring these things forward and falsify ID and become famous.

But it is a very positive hypothesis.

106. 106
gpuccio says:

Prof_P.Olofsson (#90):

You say:

“First, I don’t know what you mean by design being an “empirical observed event.” ”

I mean that we observe design every day in human artifacts. We observe the designer, the process of design, and the product of design exhibiting CSI. We observe that every time someone speaks, or writes, or makes machines, or makes computer programs, and so on. We see the CSI while it is being outputted. More than that, we have a subjective perception of the act of design when we ourselves design, and of the faculties involved, such as intelligence and will and purpose. You are reading, just now, the product of my design, with all its inherent CSI.

Isn’t that empirical? Isn’t that observation?

We have, indeed, created the word “design” exactly to describe that process, both subjectively and objectively.

You say:

“Second, when you start talking about “reasonable” and “credible” explanations, you are leaving the realm of the eliminative filter and, ironically, “thinking Bayesian.” ”

I take that as a compliment! Me, a Bayesian!

Anyway, I think that what I say is very simple. Our statistical reasoning is applied, as requested by the Fisherian approach, to the null hypothesis H0 because that hypothesis is based on the affirmation of randomness, and is therefore perfectly appropriate to compute the probability of the outcome we are observing under that hypothesis. But if we reject H0, then we are left with the problem: our data are probably not random, but what is the causal model which best explains them? Therefore, we compare the possible causal model available (if more than one is possible), and give our “methodological” and epistemological judgment on what is, for us, the “best explanation”. That judgment is not based on statistics (if not for the fact that it is supported by the first judgment of rejection of H0), but may be based on various considerations: internal consistency of the model, simplicity, even elegance, and any consideration which can derive from ou general knowledge of the subject.

You say:

“Note how you rule out one chance hypothesis, the uniform one, by calculating a probability and others by claiming they are unreasonable.”

See previous point. I can rule out the chance hypothesis for statistical reasons, and other hypotheses for internal inconsistencies. They are two different things, as you say, both perfectly valid. For instance, if a darwinist tells me that RV + NS can do specific things in his model, I can rule out the role of RV on a statistical basis, and the role of NS on methodological considerations (such as: NS can act only on a function which is already phenotypically detectable and relevant enough to be fixed, and if the model does not account for that, it is internally inconsistent).

You say:

“I might just as easily claim that design hypotheses are unreasonable and infer an alternative, unspecified chance hypothesis as the best explanation.”

You can do what you want. But you must motivate it, as I have motivated my reasoning.

If you affirm that design hypotheses are unreasonable, you have to show why. Are they internally inconsistent? Is there anything in biological CSI which cannot be ascribed to design? Is there anything which makes CSI in biological information essentially different from CSI in human artifacts, which is certainly designed?

And, if something is unspecified, how can it be a “best explanation”? The hypothesis of design is not unspecified: it is based on a well known process (design) which has the causal power to explain the data. And, see first point, which is empirical abd certainly effective.

107. 107
kairos says:

#90 P.Olof

First, I don’t know what you mean by design being an “empirical observed event.”

Let’s reason by step.

First, design requires intelligence (in a certain sense in the term Intelligent Design the first word does only enforce the concept that design is due to an itelligent entity).

Second, the ethimology of intelligence tell us that an intelligwnt act is something that involves some sort of proper choice and/or selection of real or abstract entities.

Third, in the natural world we can see that what is typical is loss of order. Instead, during the evolution there has been a huge growth i order; something that we only know to be done directly by intelligent entities or indirectly by complex algorithms, i.e. artificial entities which embed and simulate intelligence in the “selection” and “choicees” (if’s and assignment statements) they perform. So, what would be clearer?

108. 108
Atom says:

Ok, I’ll try to keep this focused.

PO (and others) claim that Dembski’s filter is unusable because it only rules out one chance hypothesis. It doesn’t rule out all the others.

True, it doesn’t rule out the hypothesis that there are some forces or processes at work (like “natural selection”) that could possibly make the formation of a flagellum more likely. It also doesn’t rule out the “solar rays make bacterial flagella more likely” hypothesis, the “quantum fluctuations facilitate the formation of nano-machinery” hypothesis, or any number of hypotheses we want to invent. We can always invent new hypotheses that supposedly differ from the uniform hypothesis.

But we would still need to define exactly how (in quantitative, numerical terms) these hypotheses differ from the uniform (unguided) chance hypothesis, since we want to claim that their probabilities are quantitatively different. If not, we’re appealing to unspecified, magical ordering principles. It simply isn’t enough to say “They are different, I can feel it in my gut.” If you can’t give at least a rough estimate of how the probabilities differ, then how can you be certain that they differ at all?

If, however, you can quantify how the probailities differ, then please do so, so that we can use them in our filter. We will then account for them as one of the “natural forces/regularities” that we already mention. (It then becomes a question of whether or not your proposed forces/processes actually exist in nature and if they in fact quantitatively alter the probabilities in the way you claim.)

As an illustration, let’s use the flagellum. It is made of several parts and somehow modern day bacteria aquired traits (genes, etc) that code for these parts. Now we assume there was a time when none of these traits existed (unless the first lifeform was equipped with a fully functional flagella – unlikely, so our assumption is pretty safe.) The traits each had to be aquired, either all at once or stepwise, a few at a time. Now these traits each exist somewhere in genome space – they may be clustered (the genes for each part very similar to the genes of the other parts), or they may be separate (the genes for each part are dissimilar.) The important thing is that they are distinct. Now, as the organisms stumbled upon these traits (by mutation, gene duplication, inversion, etc) either something was guiding them towards the exact traits they needed or something was not.

If something was, then yes, the causal history differs from the random hypothesis, but you’ve just introduced teleology (ID) into biology. Mainstream biology holds that organisms mutuate in a way that is random with respect to fitness. ID claims that this is not always the case. So if you suggest a guiding mechanism for genomic changes/mutations/variation, you’ve entered into the ID camp and left the Darwinist camp.*

If something was not guiding the mutational changes, so that the organism randomly stumbled upon the traits in a truly random way, then how is this any different than the chance (combinatorial) hypothesis? We would have to know the size of the genome space (which we can estimate), the number of functional states relative to that genome space (which we can estimate – see Douglas Axe’s work on protein folding), and whether or not some mutations are more likely than others (which we can test for empirically – do some types of inversion or mutation show up more regularly in large scale experiments?) But once we are armed with that knowledge, we again are still in a place to use the filter and are again practicing ID science.

But appeals to unknown ordering principles that may somehow make the formation of a flagellum more likely, in an unspecified, unquantified way cannot be considered science. That’s wishful thinking.

Atom

*Note: Do not say that “natural selection” guides the mutations/variations themselves, because it does no such thing; selection can only act on traits that already exist, reducing diversity, not guiding the initial changes themselves before they occur.

109. 109
Prof_P.Olofsson says:

gpuccio[111],

You say:

“First, I don’t know what you mean by design being an “empirical observed event.” ”

I mean that we observe design every day in human artifacts.

OK, I thought you were talking explicitly about intelligent design of mice and men. I just don’t see how humans designing objects have anything to do with intelligent designers designing humans.

110. 110
IDskeptic says:

But don’t beneficial traits have a great chance of being selected and persevering?

111. 111
Prof_P.Olofsson says:

Atom[113],

Minor point. You say:

PO (and others) claim that Dembski’s filter is unusable because it only rules out one chance hypothesis. It doesn’t rule out all the others.

I said that in Dembski’s application to the flagellum, he only rules out one chance hypothesis. The filter (which, again, is essentially hypothesis testing) can of course rule out any chance hypothesis once it is specified.

112. 112
Atom says:

Sorry about that PO. You’re right, your clarification is what I meant by those sentences.

113. 113
tribune7 says:

Atom, great post (113)

114. 114
tribune7 says:

I just don’t see how humans designing objects have anything to do with intelligent designers designing humans.

Professor, we know design is real because as you note we humans do it. We know designed objects have certain characteristics that are exclusive to designed objects. We attempt to quantify those characteristics, which we then apply to a methodology used to determine design.

We find them successful.

We then apply such a methodology to the universe and DNA and life in general and it registers a positive.

What does that tell you?

115. 115
Atom says:

ribczynski,

The odds improve greatly when you consider pathways that spread the changes over many generations.

By how many more percentage points do these “pathways” make the formation of a flagellum more likely? Either you know the “greatly improved odds” or you don’t. If you do, share them and let’s run the filter again. If not, you’re appealing to magical forces.

116. 116
tribune7 says:

And what are these stepwise pathways?

We don’t know them,

IOW, they are imaginary 🙂

What you seem to be missing is that computing a CSI number and comparing it to a threshold (500 bits, for example) is equivalent to running through Dembski’s flowchart.

What you are missing is that design has real characteristics and quantifying and applying them is a positive argument.

117. 117
Atom says:

ribczynski,

Lay off tribune7. The probabilities involved in the search space act as a quantifiable measure of Complexity in CSI. Adding the notion of Specification, you get Complex Specified Information, or a positive feature of designed objects. Just because probabilities are used as a proxy measure for information content (as they are in Information Theory) doesn’t make CSI a “negative” argument any more than using “uncertainty” and probability in Information Theory makes cryptographic analysis a “negative” science.

118. 118
Atom says:

ribczynski,

Please have the respect to not misquote me. I never said the pathways didn’t exist…I just said you either have evidence for them or you don’t.

Magical forces may in fact exist; I just don’t know how they quantitatively alter probabilities. And in this very important sense, I cannot distinguish your appeals to unknown forces form appeals to magic.

119. 119
Atom says:

form* = from

120. 120
Atom says:

ribczynski,

Again, classical Information Theory uses probability measures in much the same way Dembski would use them. (How likely is a given sequence of characters?) Does this make Cryptography (which uses Information Theory and infers message sequences) a negative science?

121. 121
tribune7 says:

rib

Then you should be able to show us how you would determine whether the flagellum was designed without looking at the probabilities of the non-design hypotheses.

You look at the characteristics of objects of designed objects and apply them to the object in question. Do you believe that design exists? Do you believe it has discrete characteristics?

If you think that something unknown is therefore imaginary,

Do you believe in leprechauns? Scary leprechauns with funny green hats? Would you believe in them if someone said believing in them is science?

And how, specifically, do you know leprechauns don’t exist?

122. 122
tribune7 says:

rib– He claimed that if the pathways were unknown, they must be imaginary.

OK, rib. How do you know they are not imaginary?

123. 123
tribune7 says:

Calling the EF a “negative argument” is not a pejorative.

I don’t think there is much dispute about the EF being a negative argument. The problem comes when you call the application of CSI a negative argument without basis and apparently simply to argue.

124. 124
PaV says:

PO:

I’ve already commented above at #58.

In reference to this Chapter you write in your paper:

“For example, if we choose a uniform prior distribution for p, the posterior distribution turns out to be a so-called Beta distribution with mean 41/43. In this posterior distribution, the probability that pis not above 1/2 turns out to be only about 10^-11, which gives clear evidence against fair drawing. The Bayesian analysis does not involve the set E* or any other rejection regions.”

You have here made Dembski’s point while trying to deny it. What I wrote at #58 still applies.

Now, let me try to demonstrate why I say you’re making Dembski’s point here.

Here’s your statement: “In this posterior distribution, the probability that pis not above 1/2 turns out to be only about 10^-11, which gives clear evidence against fair drawing.”

Well, what if the posterior distribution turned out this way: “The probability of p is not above 1/2 turns out be only about 0.01, i.e., one in a hundred.” Would you say , on this basis, that this probability gives “clear evidence” against fair drawing? I suspect not. Well, what if it turned out to be 1 in a thousand? Would you then say that was “clear evidence”? IOW, at what point does an “improbability” become “clear evidence”? Dembski would say that it would happen precisely at the point that you would consider it specified complexity, that is, when the “improbability” of the class of events “E*” exceeds the UPB. (Yes, the “class” of events, since any permutation of D’s and R’s results in the same “specification”/pattern) In this case, you’ve decided that the rejection region is certainly in place once the improbability of p not exceeding 1/2 “turns out to be about 10^-11. (which it would be for all 42 permutations of the pattern in the Caputo case. Divide 10^-11 by 42, and you come up with roughly the 1 in 50 billion probability of the rejection region Dembski comes up with.)

I think it’s very clear you’re making Dembski’s point for him.

125. 125
tribune7 says:

rib–You look at the characteristics of objects of designed objects and apply them to the object in question. Um, could you be a little more specific?

You are confused about the difference between a negative argument i.e. is it by chance, no; is it by necessity, no, then it is designed vs. a positive argument i.e. design has complexity and specification. Does the object have complexity? Yes.Does it have a specific purpose? Yes, is it designed. Yes.

The former is a negative argument. The latter is a positive one.

The fact that some unknown things are real doesn’t mean that they’re all real.

Some people think leprechauns are real. Some people think stepwise pathways are real. What reason do you have for believing in stepwise pathways? What evidence is for them? Seriously.

126. 126
tribune7 says:

ribczynski, you are whistling past the graveyard.

You are offer semantic arguments about which you have convinced, well, at least yourself of its merits, in an effort to ignore the evidence of ID.

And there are several things you left on the table.

You failed to address your “unknown, unquantified” forces as per Atom’s request. Nor did you answer my question as to what reason do you have for believing in stepwise pathways?

127. 127
Prof_P.Olofsson says:

PaV[137].

In your post [58], you are attacking statistical hypothesis testing. While you are of course free to do so, you might want to check any textbook to see that hypotheses are typically set up in that way: HA specifies an entire set of parameter values.

As for my comments on Dembski, I am afraid that you don’t understand what I criticize. You quote me as writing

The Bayesian analysis does not involve the set E* or any other rejection regions.”

which is precisely my point, and which is where I refute Dembski’s claim. The rest of your comment has nothing to do with the distinction hypothesis testing/Bayeisan inference. As gpuccio pointed out eariler, this discussion is technical and few here has the expertise to follow it.

128. 128
Prof_P.Olofsson says:

tribune[120],
Nothing.

129. 129
Prof_P.Olofsson says:

PaV[137],
What constitutes “clear evidence” is subjective and arbitrary. In much statistical hypothesis testing, we are satisfied with 1% or 5% or so, depending on the application and the severity of an erroneous conclusion.

Also, you bring up the UPB but in the Caputo example, you can’t get anywhere near it.

130. 130
gpuccio says:

ribczynski:

1) First of all, the only thing where you are obviously “partly” right. The argument for design is “partly” a negative argument, as I have always said. And, as you say, that “is not a pejorative” at all. But there is also a “positive” aspect in attributing to design the information present in biological information: that positive aspect is the formal similarity (presence of CSI) between information in human designed artifacts and biological information. Therefore, the whole ID argument is made up of a negative part and a positive part, as I have always told here.

2) You are insisting ad nauseam with the problem of statistical distributions, which means practically nothing as you posit it. I wonder if you really understand what a probability distribution is.
I am answering here, for brevity, also to your other post in the other thread, where you say:

“You seem to be confused here. Dembski’s method does not compute the probability distributions of the non-design hypotheses. It takes them as parameters.”

What do you mean? A probability distribution is a mathematical object, characterized by a probability function, either discrete or continuous, which has to satisfy definite mathematical axioms. What do you mean with “Dembski does not compute the probability distributions of the non-design hypotheses”? In empirical science, one has to solve the methodological problem of deciding which statistical distribution is appropriate for his theoretical model, given the empirical knowledge that he has of the data. Darwinism always refuses to give an explicit and detailed model, and hides behind that vague indetermination.

You say that you don’t know “any attempts to go through the motions and compute the specified complexity of a biological structure”. Well take any functional protein of, say, just to be comfortable, at least 200 aminoacids, for which you, or any darwinist, cannot give any explicit theoretical model of how it came into existence through explicit selectable steps. Take, if you want an extreme example, any functional protein for which no homologue is known (orphans, or TRGs): there are many of them, and the reasoning is simpler with them.
For a single protein or gene space, as I have already discussed, you can only assume a uniform distribution or a quasi uniform distribution, given what we know of proteins and genes. What would you pretend: that we test all known mathematical distributions for each experimental context, and see how they fare? Are you kidding, or what?

3) Let’s go to Dembski. First of all I must commend you for your untiring and remarkable exegesis of Dembski’s work: you must really be a fan of him, have you signed first editions and similar? Someone could be interested here… 🙂

A little more seriously, you cite the following passage from Dembski (I take it for good, I have not the time to check):

“If something genuinely exhibits specified complexity, then one can’t explain it in terms of all material mechanisms (not only those that are known but all of them, thanks to the universal probability bound of 1 in 10150; see chapter ten).”

Well, as a premise, in my post #83 here you can find this statement from me to Prof. Olofsson:

“Finally, I can certainly agree that “Dembski isn’t always right”. I have great respect, admiration, and even affection for Dembski, but infallibility is usually too heavy a burden to impose on anybody.”

If we agree that Dembski is not necessarily an authority in everything for me, I will add that the passage you cite is probably not very clear. Maybe Dembski, if he likes, could explain better what he meant.

But I am sure that he did not mean what you think he meant. First of all, here he is speaking, apparently, of the implications of having attributed correctly the CSI status to an information. Indeed, he says: “If something genuinely exhibits specified complexity, then…” In other words, the attribution of CSI has already been accomplished, and Dembski is just stating his certainty that no possible mechanism will ever explain that result in other ways than through design. That is a way of saying that the filter has no false positives.
Personally, I would not have expressed myself in that way. So, if you want to say that I don’t agree with Dembski on the opportunity of that specific sentence, be my guest, I will live with it.

My thought is very clear about that: the filter has “logically” false positives, like any other similar tool, but has “empirically” no possible one.
In the same way, I would confidently affirm that CSI cannot be explained, out of design, by any of all known or presently conceivable material mechanisms, but I would refrain from speaking of “not only those that are known but all of them”.

For me ID, like all good scientific theories, remains a “best explanation”, and not absolute truth. I have always said that, and I am not afraid to repeat it.

Anyway, please notice that Dembski is speaking of “material mechanisms”, and not of your beloved distribution computations. Dembski is very clear that he assumes the uniform distribution.

Anyone is free to assume any other reasonable distribution: the results will not change.

Unless you create an unreasonable, “ad hoc” distribution. For instance, to explain the sequence of myoglobin, you (and I mean you, rib, because nobody else would ever propose such a thing) could suggest a discrete distribution where the specific sequence of myoglobin has probability 0.99, and all the remaining possible sequences account for the remaining 0.01. That would be a perfectly valid probability distribution (mathematically), and would certainly explain away myoglobin. Unfortunately, it would have no empirical credibility.

Shall we have to consider it, so that you are satisfied? And then will you propose a different, specific, ribczynski made distribution for each protein?

Ah, that’s science!

131. 131
gpuccio says:

Prof_P.Olofsson:

You say:

“OK, I thought you were talking explicitly about intelligent design of mice and men. I just don’t see how humans designing objects have anything to do with intelligent designers designing humans.”

Maybe you are not familiar with the ID theory, then. Obviously, that has nothing to do with the “negative” part of ID, which falsifies the existing theories about unguided evolution. That is the only part which has a statistical content, and so it’s natural that it has drawn more specifically your attention.

But the “positive” part of ID proposes an alternative hypothesis, and that hypothesis is base on the empirical observation of the process of design in humans. The important fact that humans can very easily generate CSI through design, and nothing else can, is a compelling clue that cannot be overlooked when one tryis to explain the presence of CSI in biological information. The hypothesis that a process equivalent to human design can explain biological information is absolutely due. ID makes that hypothesis, and builds its theories on it, present and future, on it. Everyone is free not to share the design hypothesis. But, if the “negative” part of ID is true, then the presence of CSI in biological information remains a completely unexplained mystery.

132. 132
Prof_P.Olofsson says:

gpuccio:

…humans can very easily generate CSI through design, and nothing else can…

Conclusion: we have designed ourselves! 🙂

133. 133
StephenB says:

PO: All attempts at humor are appreciated. In that spirit, I offer you a second chance to rethink your comment to gpuccio @146. Please reassure us that it was a typo.

134. 134
PaV says:

PO (#143):

What constitutes “clear evidence” is subjective and arbitrary. In much statistical hypothesis testing, we are satisfied with 1% or 5% or so, depending on the application and the severity of an erroneous conclusion.

Also, you bring up the UPB but in the Caputo example, you can’t get anywhere near it.

If you say that you are satisfied with “1% or 5%”, then you’ve set that up as your “rejection region”. This is being “parasitic” on the Fisherian approach, which explcitly, and not just tacitly, establishes a rejection region. This is preciesely Dembski’s criticism.

As to the UPB not even being close to having been reached, Dembski, on p.103 of NFL, notes this very fact, but with this additional insight: Even though, according to your “Bayesian” analysis, you’re very content in saying that what happened in the Caputo case certainly didn’t happen by chance,Dembski points out to his readers—assuming they are clearly aware that the improbability of the ‘chance hypthesis’ has not come anywhere near attaining the UPB—that the New Jersey (York?) court did not find Caputo guilty of any crime. Apparently the “rejection region” the court was “satisfied with” was not 1%, or 5%, or even a millionth of a percent.

135. 135
Prof_P.Olofsson says:

PaV[143],
Yes, I wrote “In much statistical hypothesis testing…” which is where rejection regions appear. This is the Fisherian approach. You clearly do not understand what this discussion is about.

136. 136
Sal Gal says:

ribczynski (38) [and Prof. Olofsson],

Dembski abandons claims. He does not admit error. In “Specification: The Pattern That Signifies Intelligence,” he gives a nebulous explanation of how he is pulling together his past work. But, as is clear in your quote of the paper, he in fact quietly acknowledges that his “no false positives” objective in design detection was, is, and always will be unachievable.

The essence of a filter is that it blocks undesired stuff from getting through. If some events that pass through the explanatory filter are not designed, then there is little sense in continuing to call the inferential system a filter. The sad thing is that Dembski indeed stopped talking in terms of a filter, but did not alert his allies to the shift. I will not speculate on his motives.

The explanatory filter is a zombie. So is the argument from specified complexity. Dembski has said that his collaborative work with Marks is filling in the gaps in the mathematical arguments he sketched in No Free Lunch. This is strongly reminiscent of his claim in “Specification.” The interpretation of active information is, in the papers that have appeared at the EvoInfo website, implicitly Bayesian. That is, Dembski has shifted from a frequentist to a Bayesian perspective, and has not actively informed (pun intended) others of the fact.

I have been slapped down by anti-IDists for “failing to see” that active information is just complex specified information in disguise. (Et tu, Rib?) But, no, CSI is explicitly frequentist in formulation, and active information is formulated in terms of the “no free lunch” theorems for optimization. Dembski and Marks are trying to gain traction with the Bayesian interpretation of the theorems that was first advanced by Wolpert and Macready.

It is understandable that Dembski, having attempted to nail shut the coffin of Bayesianism, is not trumpeting the fact that he’s now taking a Bayesian approach. But somehow it doesn’t seem right for him to keep silent while his fellow IDists trot out his frequentist zombies over and again.

137. 137
Sal Gal says:

On the positive side, I want to say that I like the shift in Dembski’s work. What he and Marks write is sufficiently non-gelatinous that I can counter it. And I have no objection to people making clear arguments that I believe are wrong.

What most IDists do not appreciate is how terribly inappropriate it was for Dembski to breeze through pseudo-mathematical “arguments” in his earlier writing. There was in all of it an element of “trust me, I’m a doctor-doctor.” IDists believed that such a bright guy certainly was able to fill in the gaps, and was just trying to make things simple enough for them to read. In other words, they vested faith in Dembski.

I’m trying to tell IDists, without bashing ID itself, that they should learn from this experience. An “expert” must first supply the complete argument somewhere, and then simplify for the general readership. Dembski has effectively admitted now that he is unable to flesh out the arguments he previously sketched.

138. 138
Prof_P.Olofsson says:

Sal Gal[150],
An astute observation. Dembski’s use of the NFL theorems is indeed inherently Bayesian.

139. 139
PaV says:

PO:(#148)

Yes, I wrote “In much statistical hypothesis testing…” which is where rejection regions appear. This is the Fisherian approach. You clearly do not understand what this discussion is about.

In your paper in Chance, to refute Dembski’s argument, you use a “Baysian analysis” of the Caputo case, you come up with a improbability of the 1/2 chance possibility of 10^-11, and you state: “WHICH GIVES CLEAR EVIDENCE AGAINST FAIR DRAWING”. Did I get your attention. You again clearly make Dembski’s point; but you want to pass off your Bayesian analysis as Fisherian. You don’t seem to know what you’re saying.

You’re on the verge of being dishonest.

140. 140
StephenB says:

—–Professor P. Olofsson: “I just don’t see how humans designing objects have anything to do with intelligent designers designing humans.”

What do you mean with the phrase, “anything to do with?” I gather you are not alluding to the ridiculous prospect that the act of a cosmic designer could be, by definition, part of the same process as the act of a human designer. Perhaps you mean to say that you don’t see how the act of a cosmic designer could “have anything in common with” the act of a human designer.

Let’s go though it in abbreviated fashion:

[A] Affirmation:

Both acts of design leave clues in the form of functionally complex specified information

[B] Negation

Both constitute creative acts of intelligent innovation that cannot reasonably be explained by law or change.

Design and its meaning:

“An outline, sketch, or plan, as of the form and structure of a work of art, an edifice, or a machine to be executed or constructed.”

“An organization or structure of formal elements in a work of art; composition.”

“The combination of details or features of a picture, buiding, etc: the pattern or motif of artistic work.”

“The art of designing.”

“A plan or project; a design for a new process.”

ID suggests that the EFFECTS of human design and the EFFECTS of a cosmic design both point to these common elements. This concept is within everyone’s grasp—–including yours.

141. 141
Sal Gal says:

ribczynski (38),

You point out that Dembski indicates that non-design hypotheses should be rejected one-by-one. This is quite a surprising error on his part. The procedure should be to reject the disjunction (OR) of the hypotheses. That is, the null hypothesis should be

H0 = H1 V H2 V … V Hn,

where H1, …, Hn are hypotheses of how the event occurred by strictly materialistic processes. Dembski formulates, as indicated in the long quote you provide, design as “none of the above,” and thus “all of the above” materialistic hypotheses must be rejected jointly in favor of the (alternative) design hypothesis. His divide-and-conquer approach would reject only “each of the above, individually,” and not “all of the above, together.” The consequence would be an elevated probability of falsely rejecting materialistic explanation of an event.

Admittedly, matters are not as neat and clean as I suggest. Scientific hypotheses are not strictly logical. Common statistical practice is to penalize the test statistic (require a more extreme value for rejection of the null). Dembski does not describe how to do this with CSI (i.e., how much higher than 1 should the rejection threshold be when there are n null hypotheses?), and I believe it would have been awfully hard for him to have done so, had he stuck with CSI.

142. 142
Prof_P.Olofsson says:

StephenB[154],

I understand what you mean. Nevertheless, we know a lot about human design, nothing about “cosmic design.” This sounds mostly like a play on the word “design” to me.

143. 143
Prof_P.Olofsson says:

Sal Gal[155],
Very good point. I think this has been brought up before, maybe by Elliot Sober. It is less of a problem if you can order your hypotheses in some hierarchical manner. For example, if you want to test H0: p<=0.5, you test H0: p=0.5 instead.

144. 144
Patrick says:

Sal Gal,

Huh? I think you’re confusing issues. It’s not like Dembski “gave up on the EF” and that “the EF is a zombie”. I’ve been communicating with him privately for years and there’s been no indication of a position change like that. It’s all in your mind.

This new work is mostly relevant to Genetic Algorithms, which rely on what I called information-based “intelligent funneling” before Bill termed it active information. The selection filter and randomization function in a GA must be balanced properly in order to work (the search be funneled toward the target) and this is where active information comes in.

I would say that Dembski’s work on active information is related but distinct.

Active information is indispensable for increasing the probability of success of a search. Active information applied to a search should not be prescribed blindly, but must accurately reflect constraints on the target location. If in the search for “Ace of Clubs symbol” we are told that we are getting “warmer” when in fact we are getting “colder,” the active information contributed to the search can, depending on the algorithm crafted around this information, be negative and thus result in a probability of success less than that of random search. The NFLT presupposes the absence of active information.
….
The prior assumptions” and problem specific knowledge” required for “better-than-chance performance” in evolutionary search derives from active information that, when properly fitted to the search algorithm, favorably guides the search.
…..
The NFLT shows that claims about one algorithm outperforming another can only be made in regard to benchmarks set by particular targets and particular search structures. Performance attributes and empirical performance comparisons cannot be extrapolated beyond such particulars. There is no all-purpose “magic bullet” search algorithm for all problems.
….
Although commonly used evolutionary algorithms such as particle swarm optimization [9] and genetic algorithms [11] perform well on a wide spectrum of problems, there is no discrepancy between the successful experience of practitioners with such versatile algorithms and the NFLT imposed inability of the search algorithms themselves to create information [4, 10]. The additional information often lies in the experience of the programmer who prescribes how the external information is to be folded into the search algorithm. The NFLT takes issue with claims that one search procedure invariably performs better than another or that remarkable results are due to the search procedure alone

How would this be relevant to biology and ID you may ask? Well, besides Darwinists insisting that GA-based applications “more closely approximates the process of biological evolution” there is the fact that some ID-compatible hypotheses would require active information. In one hypothesis Darwinian mechanisms are taken into account by the Designer(s) and the architecture of life itself is configured to be modular, so that multi-functionality, gene duplication, cooption, and preadaptation, etc. are able to unmask secondary information.

Dembski’s recent work shows that in order to find the targets in search space active information is required. Besides “directed front-loading” there is the potential that design only holds true in regards to the OOL. The front-loaded active information is the design of the system (modular components, plasticity in the language conventions, foresighted mechanisms, etc), which allows the “evolving holistic synthesis” to function without there being a directly embedded plan.

Of course, this presumes that Darwinian mechanisms are capable of this task, for which we have no positive evidence at this time (heck, we do not even have POTENTIAL Indirect Stepwise Pathways to consider). I personally believe that given a system intelligently constructed in a modular fashion (the system is designed for self-modification via the influence of external triggers) that Darwinian processes may be capable of more than this. But that’s foresighted non-Darwinian evolution in any case, and even if there are foresighted mechanisms for macroevolution they might be limited in scope. As in, this hypothesis would require active information that allows for a “magic bullet” solution that is generalized in scope and capable of overcoming all barriers. Something Bill believes does not exist, nor have I seen or heard of such a thing in computer science.

But despite that major obstacle for this hypothesis personally I cannot reject it quite yet since we’re still struggling to comprehend the overall functionality and language of the entire biological system. BTW, this is the same reason I’m open to the possibility of the “evolving holistic synthesis” saving Darwinism.

Also,

The essence of a filter is that it blocks undesired stuff from getting through. If some events that pass through the explanatory filter are not designed, then there is little sense in continuing to call the inferential system a filter.

You’re arguing semantics. “Inferential system”. “Filter”. Whatever you want to call it. I think it’s been very clear for years that the reason he set a Probability Bound so (almost unreasonably) low was to ensure reliability. And if you think reliability is an issue name one instance of a false positive.

145. 145
StephenB says:

—–Professor O: “Nevertheless, we know a lot about human design, nothing about “cosmic design.” This sounds mostly like a play on the word “design” to me.”

We know that, according to our experience, functionally complex specified information always indicates design. Not often, not once in a while, but always. Science proceeds on the assumption that we live in a rational universe and that laws, chance, and design are all part of the overall operation of the cosmos. Given that fact, it is perfectly reasonable to infer that the presence of FSCI in cosmological formulations and biological design indicate the presence of intelligence just as surely as their presence in human artifacts indicates the presence of intelligence. The only way to argue against that proposition is to avoid the evidence or to suggest that we do not live in a rational universe at all.

146. 146
StephenB says:

Professor Olaffson: Also, you have yet to explain your comment to gpuccio @146: (Perhaps, I confused the issue by referring to you as PO)

gpuccio wrote, “humans can very easily generate CSI through design, and nothing else can…”

To that your responded (tongue in cheek, I gather?) “Conclusion: we have designed ourselves!”

I assumed that it was a typo, but since you have not responded, I gather that you actually did mean something by it. So, now I must ask you to explain how your conclusion follows from gpuccio’s comment.

147. 147
Prof_P.Olofsson says:

PaV[153],

…you want to pass off your Bayesian analysis as Fisherian.

No, I don’t. Bayesian inference is done by computing probabilities in the posterior distribution which is exactly what I do. Fisherian inference is done by computing probabilities under the null hypothesis. Both methods obviously aim at drawing some kind of conclusion in the end. You just don’t have enough background knowledge to follow my arguments.

You’re on the verge of being dishonest.

Nowhere near it! Besides, you should know by now that you can’t make me upset byt trying to insult me! 🙂

148. 148
Sal Gal says:

This sounds mostly like a play on the word “design” to me.

No matter that it comes from someone presenting as a logician, there is no logical basis for asserting that the intelligence of a putative designer of the physical universe is similar to that of a certain biological species that calls itself intelligent and considers itself capable of design as a consequence of its intelligence.

As Steve Fuller rightly points out, the assumptions that nature is the way it is because it was designed by an intelligent God, and that humans can understand nature by “getting into the mind of God,” have been, and may continue to be, useful heuristics in scientific investigation.

What I must insist is that people who assume (or who believe as a matter of faith in the truth of some scripture) not cloak their assumptions in pseudo-logic. There is no such thing as inductive learning without inductive bias (assumptions). Thus it is of crucial importance in discussion of what science is and what science should be to own up to assumptions, and not to “logicalize” them away.

149. 149
gpuccio says:

Sal Gal:

Your observations about Dembski’s work are interesting, but IMO very partial. My personal opinion is that he has expressed very important ideas both with CSI and with his new work about active information. I am sure that Dembski’s work, like anyone else’s, is in continuous evolution (pun intended), but that does not diminish its importance.

On the other side, what have his critics done? Darwinists refuse to even give a realistic model which can be analyzed quantitatively. Statistics and mathemathicians (including maybe you, if you are one) have given vague refutations of some formal aspects, completely ignoring the huge methodological, statistical and logical holes in the currently accepted theory. That’s, in my opinion, double standard and scientific prejudice.

Regarding your objections about the rejection in #155, I don’t see which are these multiple “non design hypotheses” which you mention. I have repeatedly affirmed that, given what we know about biological variation and the structure of genes and proteins, the only admissable hypothesis for biological random variation is the uniform, or a quasi uniform, distribution. Neither you nor Prof. Olofsson have really commented on that, or suggested what other distribution, if any, could be applied, least of all analyzed how that would change the results. The attempt to include non random factors in the statistical discourse is irrelevant. Random factors play such an essential role in darwinian theory that it is more than enough to falsify that part of the theory with a purely Fisherian approach, which would be anyway accepted in any other scientific field. The non random aspects (NS) can easily be falsified on a methodological basis, once the random part has been correctly quantified.

Let’s keep the thing simple, if we can. If a theory wants to explain something (the appearance of functional complex information) which does not usually appear by itself by means of a well known biological mechanism (random variation), it is the absolute duty of that theory to:

1) provide a realistic model of how that could happen (as detailed as possible; and believe me, all the information on molecular biology we have today can allow darwinists to try at least some specific models).

2) To show that the role of randomness in that model is quantitavely credible, and not a mere myth.

A theory which can’t do that is a complete failure. No ambiguous divagations about Bayesianism can change that.

150. 150
Prof_P.Olofsson says:

gpuccio[164],
Bayesianism can be discussed in its own right, whether you consider it “ambiguous divagations” or not. I didn’t initiate that discussion on this blog, but I will respond to comments and questions.

151. 151
Patrick says:

It’s as if he thinks that setting the UPB low magically compensates for the fact that he doesn’t know all of the possible material mechanisms.

Weird.

What’s weirder is that you give up the whole farm (you agreed we don’t know of ANY “potential” Indirect Stepwise Pathways that generates CSI) then act as if this is a huge problem for ID. Dembski did what he could by allowing for all the probabilistic resources in the entire universe for all time. Again, the causal history argument has been shot down before. And I ask you again, how should we somehow incorporate into calculations something we know nothing about? And finally yet again, we can calculate the informational bits in an IC system in the absence of unknown pathways.

For lack of time, it may be snarky but so far it seems I could leave by saying I could shorten the argument to “We don’t know so ID is wrong”.

152. 152
gpuccio says:

Sal Gal:

You say:

“a certain biological species that calls itself intelligent and considers itself capable of design as a consequence of its intelligence”

I think you could have been more precise. As I have observed many times, the concept of design originates for our experience of human design. So, we don’t “consider ourselves capable of design”. Your concept is just wrong. We define design on the base of what we observe ourselves doing.

I have also remarked that the process of design in humans is observed both subjectively and objectively: it is therefore doubly empirical. And now, thanks mainly to Dembski whom you so much criticize, we have even a formal assesmsent of the product of design: CSI.

Moreover, it is simply not true that “there is no logical basis for asserting that the intelligence of a putative designer of the physical universe is similar” to that of human designers. Or at least, if you mean “logical” as in logical deductive reasoning in mathemathics, I may agree with you. But the basis exists, and as said many times by me and by others (how many times must one affirm something to obtain that it is at least considered in the pertinent answers?) it is an “empirical” basis, and it is the formal similarity between CSI in human artifacts “and” in biological information, and nowhere else. That is a very good basis to apply our reasoning to empirical facts, and to infer design in biological information. You may not agree, you may not like it, but you can’t ignore that basis and just say that “there is no basis”.

You say:

“There is no such thing as inductive learning without inductive bias (assumptions). Thus it is of crucial importance in discussion of what science is and what science should be to own up to assumptions, and not to “logicalize” them away.”

That’s correct. I agree. But that does not give you, or anyone else, authority to decide what is “logicalization” and what is reasonable induction. Scriptures have nothing to do with that. Let’s just say that anyone has a general world view, and that anyone’s world view can be a cause of personal bias or, as you say, and I hope more often, “useful heuristics in scientific investigation”.

153. 153
Prof_P.Olofsson says:

Patrick[166],
In regards to my contribution, I have not made any claims about ID being right or wrong. My interest stems from the attempts by Dembski (and to some extent Behe) to use mathematics and statistics.

154. 154
gpuccio says:

Patrick (#166):

“I could leave by saying I could shorten the argument to “We don’t know so ID is wrong”.”

I think you have masterfully summed up the real essence of the discourse. What a pity that we are not allowed to say: “We don’t know so darwinian evolution is wrong”! And we have instead to show thousands of times very obvious arguments, without being even considered.

Maybe this is one case where general world views, in one of the two fields (guess which?) are more personal bias than useful heuristcs…

155. 155
gpuccio says:

Prof. Olofsson:

“Bayesianism can be discussed in its own right”

I have never denied that. I fully respect Bayesianism. It should be clear, however, that no objection to the ID theory on the basis of Bayesianism is empirically relevant.

156. 156
Prof_P.Olofsson says:

StephenB[161],
Yes, it was a joke, assuming that gpuccio did not mean exactly what he wrote.

“PO” is fine, “Olaffson” more questionable…

157. 157
IDskeptic says:

Has anyone got an easy to understand CSI calculation? I’m fairly math savvy…

158. 158
Patrick says:

Prof Olofsson,

Sorry if I seemed to be directing that at you. I think your criticisms are interesting and that if they lead to the tightening up of ID tools and methodology–if warranted–then it would be fruitful.

In any case I need to get going now.

159. 159
StephenB says:

—–PO: “Olaffson is more questionable.” Sorry about that. I will take note of the proper spelling. (Olofsson)

160. 160
Sal Gal says:

Patrick,

I am a scholar who despises propaganda, and I consequently annoy social activists on both sides of this debate. Let’s proceed, please, under the assumption that “I calls ’em like I sees ’em.”

I’ve always admired Dembski’s intellect. I genuinely hope that he makes the most of the years left to him as a scholar. He has plenty of adversaries who delight in seeing him fail, and I am not one of them. When I say that CSI is a zombie, keep it in mind that I say also that active information may have some engineering applications. Then again, I say emphatically that attempts to debunk evolutionary theory through consideration of active information are ill-founded.

Dembski’s latest treatment of CSI is dated June 23, 2005, at his website, and is identified as the “latest installment in the Mathematical Foundations of Intelligent Design.” It hinges on a radical extension of Fisherian hypothesis testing. If Bill has not taken just the statistical part of the work and tried to get it through peer review at a statistics journal, I would consider that bizarre. He would be forever famous as a statistician, all consideration of ID aside, if his statistical reasoning were correct. I don’t believe that it is, but he has the Ph.D. in the relevant area, and I do not. (Thus I leave it to people like Prof. Olofsson to scrutinize that aspect of his work.) But I can say that until it is established that Dembski has indeed made a monumental contribution to statistics, there is no reason to vest belief in his latest treatment of CSI.

In short, I’m pointing out another instance of Bill not doing first things first. I’m not trying to nail him to a cross, but this seems to be a recurring problem with his claims.

Obviously, I have no way of proving that Dembski is not still mulling over CSI. I will not be surprised if he simply dumps the “Specification” zombie into a book, along with papers coauthored by Marks. In all sincerity, though, I think it will serve him better just to drop CSI, as it appears he has done.

161. 161
PhilipBaxter says:

I think CSI is a very interesting concept and if developed to the point where, as I understand it, a figure for the CSI in various constructs can be given then it’ll be a valuable tool for more people then just IDists.

gpuccio

We define design on the base of what we observe ourselves doing.

As a medical doctor, do you find that your work has influenced your belief (with regard to ID) or that your belief influences your work, or both? What have you observed yourself doing?

162. 162
tribune7 says:

Patrick — I think it’s been very clear for years that the reason he set a Probability Bound so (almost unreasonably) low was to ensure reliability.

Excellent point.

163. 163
Prof_P.Olofsson says:

gpuccio[170],
I’m not sure what you mean by “empirically relevant” but there are certainly relevant Bayesian objections to ID. Actually, even within the extended hypothesis testing paradigm, there are objections. For example, why is the no concept of power calcuilations in the filter? That is, why don’t we ask for probabilities to be computed under design hypotheses? Dembski’s reply seems to be that we don’t have to, because “logic does not require that rejected hypotheses are superseded” (approximate quote). Well, that’s an opinion, but I think the scientific method does not merely aim at ruling out what’s false, but also offering plausible alternatives.

And why should we infer “design” when we rule out “chance” (in the sense of “uniform distribution”)? All we can really infer is “not chance” which might just as well be hitherto unknown principles and laws of nature.

164. 164
Sal Gal says:

Patrick quoted Dembski and Marks:

The NFLT presupposes the absence of active information.

The statement is trivially true. It is a matter of how active information is defined. Marks and Dembski create the impression that the NFL theorems predicate no prior information, but this is simply false. Don’t take my word for it. Read carefully the first sentence of Wolpert and Macready’s “No Free Lunch Theorems for Optimization”:

The past few decades have seen an increased interest in general-purpose “black-box” optimization algorithms that exploit limited knowledge concerning the optimization problem on which they are run.

In fact, the algorithms do exploit knowledge of the finite domain and finite codomain of the cost (fitness) function, as well as knowledge that the function is total. (A function f is total when f(x) is defined for all x in the domain of f.) The so-called black box is in fact gray. And it is not even a very dark shade of gray.

NFL is a condition of statistical symmetry that arises when particular properties of the cost function are known. Take away some of that knowledge, and the symmetry is broken.

The definition of active information, not to mention its interpretation, depends critically on prior knowledge of the finite domain of the cost function. If all that is known about the domain is that it is, say, a subset of strings over some finite alphabet (think chromosomes of variable length, with no known upper bound on length, for concreteness), then the domain is effectively infinite. And in this case random search (uniform sampling without replacement) is undefined, because there is no uniform distribution on an infinite set of discrete objects. And the probability that random search locates the “target” is the baseline with respect to which active information is measured. Ironically, “active information” is meaningless in the absence of prior knowledge.

The upshot is that Dembski and Marks have developed analysis that applies only when quite a bit is known about the optimization problem. If you make the assumption, which present-day evolutionary biologists do not, that evolutionary processes search a space of genotypes, then how are you, even with access to scientific knowledge, going to bound the number of base pairs in the genotype? How big might nuclei become? Biological processes do not “know” the limit. Even if the limit on the size of the genotype were encountered, what would stop existing genetic mechanisms from extending DNA sequences (i.e., with lethal consequences)? Thus active information is undefined even for a “cartoon” model of evolution.

I’m all for Bill and Bob going where they can with engineering applications of active information. From my perspective, they stand to make folks think much more carefully about computational simulation of evolution. But I don’t like their tactic of shooting down the simulation, and then insinuating, with much waving of hands, that they’ve shot down theories of biological evolution.

165. 165
Sal Gal says:

gpuccio,

I have repeatedly affirmed that, given what we know about biological variation and the structure of genes and proteins, the only admissable hypothesis for biological random variation is the uniform, or a quasi uniform, distribution. Neither you nor Prof. Olofsson have really commented on that, or suggested what other distribution, if any, could be applied, least of all analyzed how that would change the results.

The relation of the size of the genotype (number of base pairs) to the apparent complexity of the organism is presently mysterious. But you can get from my last post that the distribution of trials on the space of all physically possible genomes diverges radically from the uniform.

What I have seen you do in the past is to fix the number of base pairs. Genotypes tend to get bigger over time. Someone correct me if I’m wrong, but my impression is that insertions of genetic material are much more common than deletions. If you insist on regarding biological evolution as operating on a space of genotypes — evolutionists no longer have such a simple view of things — you must take it into account that numbers and lengths of chromosomes vary. The distribution cannot be uniform.

166. 166
Prof_P.Olofsson says:

Sal Gal,
I don’t know if you are familiar with Olle Haggstrom’s critique of Dembski’s use of th NFLT. If not, I would like to recommend it.

167. 167
Prof_P.Olofsson says:

Speaking of NFLT, a nice counterexaple to Dembski’s claims in No Free Lunch is provided by the famous chloroquine resistance in the malaria bug. Obviously the bug’s restricted search did much better than would uniform search over the entire search space. Also obvious is that it doesn’t make sense to assume a uniform distribution over fitness function space (an assumption inherent in the conclusion of “on average” in the NFLT).

168. 168
tribune7 says:

But I don’t like their tactic of shooting down the simulation, and then insinuating, with much waving of hands, that they’ve shot down theories of biological evolution.

Sal Gal, I find your comments to be interesting, instructive and constructive but do you hold the theories of biological evolution to the same standards that you Dembski’s work?

If you make the assumption, which present-day evolutionary biologists do not, that evolutionary processes search a space of genotypes,

What assumptions do present-day evolutionary biologists make? Can we test them? Is it fair to consider probabilities when confronted with them?

169. 169

I wish I had time to respond adequately to this thread, but I’ve got a book to deliver to my publisher January 1 — so I don’t. Briefly:

(1) I’ve pretty much dispensed with the EF. It suggests that chance, necessity, and design are mutually exclusive. They are not. Straight CSI is clearer as a criterion for design detection.

(2) The challenge for determining whether a biological structure exhibits CSI is to find one that’s simple enough on which the probability calculation can be convincingly performed but complex enough so that it does indeed exhibit CSI. The example in NFL ch. 5 doesn’t fit the bill. The example from Doug Axe in ch. 7 of THE DESIGN OF LIFE (www.thedesignoflife.net) is much stronger.

(3) As for the applicability of CSI to biology, see the chapter on “assertibility” in my book THE DESIGN REVOLUTION.

(4) For my most up-to-date treatment of CSI, see “Specification: The Pattern That Signifies Intelligence” at http://www.designinference.com.

(5) There’s a paper Bob Marks and I just got accepted which shows that evolutionary search can never escape the CSI problem (even if, say, the flagellum was built by a selection-variation mechanism, CSI still had to be fed in).

170. 170
DaveScot says:

Bill,

Yeah, I recognized that problem for the EF in biology at first glance which is why I never liked talking about it with the flagellum. It works great if you have enough information about the probabilistic resources. It still makes me smile when I caught the guy who did the work on Harvard’s “Inner Life of the Cell” video red-handed using the EF to prove that the video clip used in Expelled was based on his work. I had to agree with him. The EF doesn’t lie. It’s a sound algorithm but like all algorithms it’s subject to rule of computing “Garbage In Garbage Out”. With something as complex as the flagellum it seems to be an intractible problem to qualify the input data well enough.

An even better example was when Panda’s Thumb presented us with 4 DNA sequences of several hundred bases each and challenged us to use the EF to find which of them contained evidence of design. Someone here, I think it was Patrick, found a Venter Institute watermark in one of them and we ran the watermark through the EF and reached a design inference. Venter used the principles behind the EF to design a hidden watermark that would stand up in court if he ever needed to show that someone stole DNA sequences from his artificial genome. And that WAS in biology.

171. 171
DaveScot says:

ribczynski

I asked you to stay out of this thread several days ago. If you continue to post to it I’m just going to delete them as I did just now.

Edit From Patrick:

Ahhh…and he was nice enough to concede that he does not have an answer in those deleted posts! When asked directly, “And what are these stepwise pathways?” the reply from ribczynski was “We don’t know them, and Dembski didn’t either.” He really gave up the farm with this argument. At this point we refer back to discussions on causal history, unknown laws, etc. How can we incorporate something we know nothing about into our calculations?

172. 172
Sal Gal says:

Well, Bill, I have a paper, “Sampling the Response of a Black Box in Bounded Time,” in review, and it’s conceivable that adding a citation of your work would be appropriate. From the acknowledgment:

The author thanks his esteemed adversary in matters of “no free lunch,” Bob Marks, for provoking thought on how to blacken the box.

Incidentally, I’m not withholding the paper now because it’s secret, but because it’s in horrible shape. I’m sure my brilliance dims in comparison to the genius of your math prof friend at the University of Chicago. But if he is like the several math geniuses I know, he is unlikely to mention that a serious problem with NFL analysis is that there is no model of anything physical. It seems to me that if you want to talk about biological evolution, you would want an analysis that takes time into account.

It is interesting to note that if the preconditions of the NFL theorems are satisfied, and running time is bounded, and extension of a sample never decreases quality (a reasonable assumption), then the best choice of optimizer is simply the one that is expected to yield the largest sample.

A very interesting thing about a physical box is that its response latencies are not only costs in sampling, but also a potential source of information. The distribution of latencies is certainly thick-tailed when the box may implement a partial function, and I think it is more reasonable to assume that it is thick-tailed than not when the box is known to implement a total function. The upshot is that there is an advantage in sometimes giving up on obtaining an output, and moving to a new input. Knowledge of the particular box is not required to gain this advantage.

Another interesting (I hope) point is that there must be some sort of physical channel through which inputs and outputs are transmitted. The bandwidth is finite, and with the NFL framework as my point of departure, I make the channel discrete. Consequently, I model inputs and outputs as strings over a finite alphabet. I see no way around an assumption that for inputs x and y, with x shorter than y, the probability that the latency for x is less than t is greater than the probability that the latency for y is less than t. (Clunky statement, but I can’t refer to expected latency.) This amounts to assuming that, all else being equal in the box’s response to the inputs, it takes longer to transmit x than to transmit y. And it justifies proceeding from shorter to longer inputs in sampling the response of the box when running time of the algorithm is bounded.

NFL is not only a consequence of knowing a lot about the function implemented by the box — knowing the finite input and output languages, and knowing that the function is total — but also of ignoring time.

[T]he analysis… does not concern itself with the computational engine used by the search algorithm, but rather concentrates exclusively on the underlying statistical nature of the search problem. The current probabilistic approach is [complementary] to computational complexity. Future work involves combining our analysis of the statistical nature of search with practical concerns for computational resources.

Well, the future is now. I’m probably making some mistakes in details, but the general outline of what I’m doing is good. And I am willing to say that I’ve benefited more from disagreeing with you than from agreeing with others addressing NFL. Before turning supercilious again, you might want to consider that while you’re doing all your ID stuff, this subgenius obsesses on one little area. And it’s actually an embarrassment for me, a computer scientist, to have gone 13 years before bringing time into the analysis.

173. 173
PaV says:

PO: #147:

Both methods obviously aim at drawing some kind of conclusion in the end.

Yes, they both have a cutoff of improbability, which, in Fisherian approach, per Dembski’s writings, is a rejection region, per se, but which in the Bayesian approach is simply “implicit” and only succeeds in being “parasitic” on the Fisherian approach.

You verge on being dishonest, not because I’m trying to make you angry, but because you’re on the verge of being intellectually dishonest. Even in your last post, you write this:

Bayesian inference is done by computing probabilities in the posterior distribution which is exactly what I do.

That’s wonderful. You calculate probabilities. But, how do you interpret them?

If you want to say that you simply compare one likelihood to another and choose between them, well, what does that prove exactly? As Dembski has pointed out, with more than one example, the likelihood approach (=Bayesian) can tell you A is more probable than B, yet, using the UPB, neither A nor B is a reasonable possibility, and any reasonable person would conclude the same. With these examples, Dembski has pointed out the limitations of the Bayesian approach in a logically convincing manner. It appears you simply don’t want to admit it. I can now see why Dembski doesn’t take the time to argue with his critics. It doesn’t seem to get anywhere.

174. 174
Sal Gal says:

“Thick-tailed” –> “heavy-tailed”

It’s late.

175. 175
gpuccio says:

Prof_P.Olofsson (#163):

“why don’t we ask for probabilities to be computed under design hypotheses?”

The design hypothesis postulates a specific causal factor (the designer and the design process) whose power to generate CSI in the case of human design is well proved, and which does not depend on random processes, as far as we know. So, I don’t understand how you could compute “probabilities”. Maybe because after all I am not a Bayesian. If you have a computer programmer, what are the probabilities that he will write his next software? I know, we can always bet, but for us Fisherians that is not a very good answer.

I don’t anderstand your reference to power calculations. Maybe I am wrong, but usually power is defined in relation to the beta error, which is the error of not rejecting a “false” null hypothesis. But, when we assess CSI, we do rject the null hypothesis, and so we are interested in the alfa error. The possible beta error should indeed worry darwinists, who do not reject the null hypothesis (although for unfathomable motives).

“the scientific method does not merely aim at ruling out what’s false, but also offering plausible alternatives.”

And we are indeed offering a plausible alternative. But, again, the alternative is offered through a correct causal model, based on empirical knowledge of the process of design. And it does not need any statistical evaluation because it does not utilize random procedures.

“And why should we infer “design” when we rule out “chance” (in the sense of “uniform distribution”)? All we can really infer is “not chance” which might just as well be hitherto unknown principles and laws of nature.”

Inferring “not chance” and then offering some correct alternative hypothesis based on causality and on a credible model is exactly what is routinely done in all hypothesis testing in medicine and biology. That’s what ID does. I can’t understand your objections.

Obviously, it is perfectly possible to offer many different causal models, once npn chance has been affirmed. That was exactly my point in my answer to you, when I affirmed that different HA may coexist, and that one is free to choose the “best explanation” among non-chance hypotheses, according to methodological and epistemological considerations.

ID does not pretend to be “absolute truth”. It’s “just” the best scientific explanation available. If and when “hitherto unknown principles and laws of nature” will be discovered, and a credible model for biological information made up on that basis, we will compare it with ID, and see which is best. Until then, we are the best, and that’s enough for us 🙂

176. 176
JT says:

From “Specification: The Pattern that Signifies Intelligence”, Dembski, p. 11:

[The vast majority of] sequences are … algorithimically incompressible. It follows that the collection of [compressible] sequences has small probability among the totality of sequences so that observing a [compressible] sequence is reason to look for explanations other than chance.

Is the truth of the above based on the assumption of the following proposition:

For all properties P:
If the vast majority of sequences do not have property P then it follows that the collection of sequences that do have property P has small probability among the totality of sequences so that observing a sequence that does have property P is reason to look for explanations other than chance.

Would this propostion be true for a property of being a specific sequence, or only for a property held by an infinite number of sequences.

Simple question presumably, but I can’t be tying up bandwidth too much at this hour.

177. 177

From “Specification: The Pattern that Signifies Intelligence”, Dembski, p. 11:

[The vast majority of] sequences are … algorithimically incompressible. It follows that the collection of [compressible] sequences has small probability among the totality of sequences so that observing a [compressible] sequence is reason to look for explanations other than chance.

Is the truth of the above based on the assumption of the following proposition:

For all properties P:
If the vast majority of sequences do not have property P then it follows that the collection of sequences that do have property P has small probability among the totality of sequences so that observing a sequence that does have property P is reason to look for explanations other than chance.

Would this propostion be true for a property of being a specific sequence, or only for a property held by an infinite number of sequences.

Simple question presumably, but I can’t be tying up bandwidth too much at this hour.

178. 178
Mark Frank says:

“The design hypothesis postulates a specific causal factor (the designer and the design process) whose power to generate CSI in the case of human design is well proved, and which does not depend on random processes, as far as we know. So, I don’t understand how you could compute “probabilities”. Maybe because after all I am not a Bayesian. If you have a computer programmer, what are the probabilities that he will write his next software? I know, we can always bet, but for us Fisherians that is not a very good answer.”

gpuccio

I am going to jump in the hope of saving PO having to respond to a zillion comments. No doubt he will correct me if I get it wrong.

There are several ways in which it makes sense to talk of probabilities relating to hypotheses which include design.

For example, to pick up on your pogrammer, let us suppose I want to assess whether a particular pattern of bits in an information channel, e.g. a radio signal, was the result of a programmer inserting them deliberately or noise generated by electrical interference.

To assess the programmer hypothesis you would take into account factors such as the availability and competence of such programmers, their access to the channel, possible motivations and their chances of success. It is fairly subjective, but we make estimates like this all the time. For example, was an explosion in a factory the result of a terrorist attack or an accident. Note that it is not just a question of estimating if there was such a programmer/terrorist would they succeed but also what are the chances that such a programmer/terrorist exists.

179. 179

Mark Frank wrote:
“For example, to pick up on your pogrammer, let us suppose I want to assess whether a particular pattern of bits in an information channel, e.g. a radio signal, was the result of a programmer inserting them deliberately or noise generated by electrical interference.

“To assess the programmer hypothesis you would take into account factors such as the availability and competence of such programmers, their access to the channel, possible motivations and their chances of success. It is fairly subjective, but we make estimates like this all the time…”

Perhaps true, but what is provocative about the Design Inference is the contention that we can determine something isn’t noise without any background information at all, just certain attributes of the signal itself.

180. 180
gpuccio says:

Sal Gal:

You raise a point which surprises me, and so I will comment on that. You say:

“If you make the assumption, which present-day evolutionary biologists do not, that evolutionary processes search a space of genotypes, then how are you, even with access to scientific knowledge, going to bound the number of base pairs in the genotype?”

And then go on discussing that concept, in your posts #164 and #165.

You seem to assume that ID examples in infering CSI are tied to the whole genome. But that’s not true. While there is no doubt that a whole genome exhibits CSI, it is not necessary to embark in the evaluation of such a complex example (after all, we IDists are generous people). It is absolutely enough to consider much simpler examples, where all considerations are easier. Or are you suggesting that, if a single protein need to be designed, a whole genome could arise by chance?

So, let’s go to simple examples. the simplest example in the world is a single protein gene. That’s the example I have always made. We can distinguish at least three different cases:

a) emergence of a “de novo” protein with a new function (many “de novo proteins, or orphans, ot TRGs, are well known, and I hope you will admit that “de novo” proteins had to appear many, many times in the course of natural history)

b) transformation of an existing protien into another one with the same, or simila, function

c) transformation of an existing protein into another one qith a new function

For each of this models it is perfectly possible to try an approximate computation of probabilities, even with the existing limits of our knowledge (many of which will be soon overcome with the increase of our understanding of molecular biology, as I have often argued). b) is obviously the “least embarrassing” for darwinian theory, and yet very problematic for it just the same, except for very trivial examples. a) and c), instead, imply insurmountable difficulties for the theory practically always.

For simplicity, again, I will briefly discuss case a). I can’t see how your objection about the number of bases can have any relevance. Usually, a known protein has an approximate length which varies little in the different species. For instance, myoglobin is approximately 150 aminoacids long in most species. There is obviously a relationship between length and function.

So, when I ptopose a model, say, of a 200 aminoacid protein, I am not saying that that is an absolute length. After all, if we compute the search space for 200 aminoacids, we are computing a lower limit (if we can call low 20^200). We can well include into the target space all possible sequences which are functional in that sense, even if their length is slightly different. That way, I think we are really helping darwinists. But you know, we IDists are generous people.

The concept is that it is completely impossible (empirically) that a de novo 200 aminoacid protein may arise by random variation, and it is not necessary to be sophisticated theorists to understand that. If that is true for a search space of 200 aminoacids, will it not be true for a whole genome? Perhaps we could remember here that modt proteins are mich longer, that the simplest known living being has a genome of hundreads of proteins, that the genome includes levels of organization and information which go well beyond the simple protein sequences, and so on. Oh, but yes, we have to discuss Bayesianism. All those things are not worth our attention…

A final, repetitive note about uniform distribution. Let’s talk anout my example: a single functional protein. What is meant here by uniform distribution is simply that all possible protein sequences have the same probability to be generated by random variation.

Now, we know that that is not “exactly” true. For instance, the genetic code is not really symetric, some aminoacids are represented by 4 codons, some by two, there are the stop codons, and so on. Therefore, even if the mutations at the nucleotide level were absolutely equi-probable, which probably is not the case, the aminoacids would anyway have different propabilities to occur in random sequences. We can well take into account those differences, but what would be the difference for our purpose? There would be no difference.

Why? Because whatever the probability distribution, it will have no special relationship with the functional sequences. For instance, one functional protein could be slightly more probable if it has an aminoacid content which compares well with the spontaneous occurrence of aminoacids, but another one could be in the opposite situation. And we have millions of different functional proteins.

Moreover, only a probability distribution which does not posit exceptional constraints ir really good to express all kinds of possible functional sequences. Functional sequences obey functional constraints which are at a higher, and independent, level in comparison to the laws which govern nucleotide variation.

If you cannot recognize the fundamental importance of these things, I really don’t know what to say more. Maybe you are interested in other kinds of problems, and I happily recognize your depth of thought and your competence as a scholar. But here, we are interested in a very practical problem: the origin of biological information. And on that we would like to discuss realistically.

181. 181
gpuccio says:

Mark Frank (#177):

“It is fairly subjective, but we make estimates like this all the time. ”

In social sciences, maybe. Not usually in medicine or biology. Biologicla sciences are still almost universally committed to Fisherian hypothesis testing (luckily, I would say). Why should ID behave differently?

182. 182
tribune7 says:

It seems to me that if you want to talk about biological evolution, you would want an analysis that takes time into account.

What amount of time should be taken into account? The age of the earth? The age of the universe? The UPB actually does and times it by a billion.

The beginning of biological life at 3.5 billion years ago? (Hopefully you understand that’s a guesstimate.)

The Cambrian explosion estimated at 530 million years ago although some estimates differ by 10s of millions of years?

PreCambrian bilaterians?

And how much credence can we give the fossil record?

From Wiki:

Another fossil, Vernanimalcula, has been interpreted as a coelomate bilaterian,[45] but may simply be an infilled bubble.[46]

183. 183
Prof_P.Olofsson says:

PaV[173],
It’s getting pointless discussing with you. You have no idea what we are talking about. I have refuted Dembski’s claim about the rejection region E* by showing that there is no such rejection region in a Bayesian analysis. Just read what I wrote about it and comment on that instead of going on about other things.

The “likelihood approach” you mention is not Bayesian. The key feature of Bayesian analysis is the interpretation of unknown parameters as random variables, hence the appearance of prior and posterior distributions. If you want to discuss these things, you need to grab a textbook and learn the basics first.

The interpretation of a small probability as something being unlikely has nothing to do with what method you use. In fact, the Bayesian approach is older than the Fisherian so there’s hardly any parasitism there.

Basta!

184. 184
Prof_P.Olofsson says:

Bill[169],

(5) Will be interesting to read. Let me guess that there is an assumption of a uniform distribution somewhere in the paper!

185. 185
Prof_P.Olofsson says:

gpuccio[180],
If it’s done right, yes. Here is an entertaining article that you might enjoy.

186. 186
tribune7 says:

Are you saying it is within the realm of reason to think 200 amino acids in a specific sequence solely by chance?

187. 187
tribune7 says:

Correction:

Are you saying it is within the realm of reason to think 20 or so different amino acids could arrange in a specific 200-long amino acid sequence solely by chance?

188. 188
StephenB says:

—–Professor O: I appreciate the fact that you are busy playing chess with several bloggers. Nevertheless, you deftly navigated away from an earlier point:

You wrote, “Nevertheless, we know a lot about human design, nothing about “cosmic design.” This sounds mostly like a play on the word “design” to me.”

We know that, according to our experience, functionally complex specified information always indicates design. Not often, not once in a while, but always. Science proceeds on the assumption that we live in a rational universe and that laws, chance, and design are all part of the overall operation of the cosmos.

Given that fact, it is perfectly reasonable to infer that the presence of FSCI in cosmological formulations and biological design indicate the presence of intelligence just as surely as their presence in human artifacts indicates the presence of intelligence. The only way to argue against that proposition is to avoid the evidence or to suggest that we do not live in a rational universe at all. To say that “we know nothing about cosmic design” is an inaccurate statement. In fact, we know that it contains FSCI, and that is a lot to know.

When we contrast that evidence for cosmic and biological design, which is real and concrete, against the lack of evidence supporting the alternative model, Darwinian UNGUIDED MACRO evolution, the conclusion is inescapable for any rational person. Design is a far more likely alternative than chance. Don’t let your foray into statistics, which is appreciated, mislead you about the reality of the big picture. The numbers associated with the CSI, which are debatable, are not synonymous with the fact of CSI, which is not.

189. 189
IDskeptic says:

Hey Stephen [188]. You say:

“We know that, according to our experience, functionally complex specified information always indicates design.”

I think many peoples problem with ID_implementation is that these things are not calculated mathematically very often and so become nebulous concepts.

Talking about FCSI is good, but I’m not convinced anyone can measure it.

190. 190
191. 191
Atom says:

IDskeptic wrote:

Talking about FCSI is good, but I’m not convinced anyone can measure it.

Three subsets of sequence complexity and their relevance to biopolymeric information: David L Abel and Jack T Trevors

Functional information and the emergence of biocomplexity: Hazen, R.M., Griffen, P.L., Carothers, J.M. & Szostak, J.W. (2007)

192. 192
193. 193
StephenB says:

—–ID skeptic writes, “I think many peoples problem with ID_implementation is that these things are not calculated mathematically very often and so become nebulous concepts.”

Let me offer two responses:

[A] q[v keijgnv[ky]14acdo je9 e;f=7,kfvor o torp itpe h[[i;y ,;k;tlr iwnri439eijd1-rri v0kv,fpe eo. ieoc dicok pf=rw-4ifp i oe ei82hf,v30 lc ioe peiw 39 o oewpoceui o kcuewi jciptpjropo

[B] Paragraph [A} was not designed, while this one was. That is the concept. There is nothing nebulous about it all all. And, yes, the probability that [A] was a chance formulation and [B] was not can be measured. Also, see Atoms link at 191.

194. 194
tribune7 says:

Professor, the paper seems to support Dembski (granted I was scanning at the end).

Dembski is using the strong form of null hypothesis. He makes a central prediction — that design has measurable qualities — and challenges the theory by attempting to reject it i.e. find something not designed that has those qualities.

If you find something not designed that has CSI, Dembski’s null hypothesis is false.

195. 195
Mapou says:

Ignore this comment if this has been rehashed in the past.

Nevertheless, we know a lot about human design, nothing about “cosmic design.” This sounds mostly like a play on the word “design” to me.

I agree with Stephen that design is design regardless or who is the designer or what is being designed. All designs share certain common charateristics. It seems that almost every discussion on UD ends up being about the distinction between design and non-design. Looking at the problem from an artificial intelligence point of view, I agree with other commenters that design always necessitates forethought, i.e. the ability to predict outcomes based on what-if scenarios. This presupposes the ability to learn from previous experience via interactions with a dynamic and partially predictable environment. Since events in nature are almost always probabilistic, it also presupposes the ability to ascertain the probability of future events.

In this light, is there a sure fire method to determine whether or not a particular system of arbitrary complexity required the ability to make predictions based on the probability of outcomes?

On a slightly different tangent (and I’m sure others have pointed this out), it appears to me that the sine qua non of survival is prediction. No lifeform can survive long unless it can reliably predict future events and conditions or unless some other agent does the prediction for it and prepares the living conditions accordingly. How can something that is so essential to survival evolve? Should it not be part of the equation from the beginning? Just a thought.

196. 196
Sal Gal says:

gpuccio,

And then go on discussing that concept, in your posts #164 and #165.

I don’t respond very methodically — sorry. A parent generates an offspring cell, not a genotype that grows a cell around itself. Allen MacNeill recently commented that the

“modern evolutionary synthesis” … is based on the idea that all phenotypic change is preceded (and caused by) genotypic change. This assumption, while warranted in the 1920s, is now known to be so inadequate a description of reality as to be essentially wrong.

Forgive me if this seems like appeal to authority, but I don’t have time today to go into my personal understanding of the matter.

BTW, I was in fact complimenting you in the multiverses thread.

197. 197
IDskeptic says:

There’s a lot on methodology, but few examples (and they tend to be cited from other sources) – I’m a pragmatist and I just don’t see people using these tools at all. Some of the examples seem to me to model a ‘spontaneous generation’, some looked at the odds of ‘this winning hand’ vs. ‘any winning hand’ (over specified?)and they all put ‘design’ where perhaps we should have ‘we don’t know’, ie are we sure that we have and understand all the natural processes in play?

198. 198
PaV says:

PO[183]:

It’s getting pointless discussing with you. You have no idea what we are talking about.

This is rather smug on your part. Do you believe that Bill Dembski understands Bayesian analysis? Do you believe that he understands the ‘likelihood approach’? Well, he comes to the same conclusion I do. So, why is it you want to be, or need to be, so smug?

I have refuted Dembski’s claim about the rejection region E* by showing that there is no such rejection region in a Bayesian analysis. Just read what I wrote about it and comment on that instead of going on about other things.

I’m afraid you’ve done no such thing. In your mind you may think that; but, in fact you have not done that. In fact, you’ve proved the converse. You make Dembski’s point. You know this, or should know this. That is why I am talking about verging on dishonesty.

When Dembski uses E*, rather than E, he is simply pointing out that using any permutation of 41 D’s and 1 R will gain you the same prior and posterior probabilities, which, if lumped together as a class of probabilities, have the same improbability using the Fisherian approach as the Bayesian approach comes up with analyzing just the one single instance that actually occured (Caputo’s list). And, applying a Bayesian/likelihood approach simply makes use of one element of this rejection region to ‘reject’ the chance hypothesis. (You’ll admit, I would think, that you reject the chance hypothesis in this case.) If this is not the case, then tell us why it isn’t so. Thus, in saying that there is “clear evidence” that the probability of 10^-11 for the 1/2 chance hypothesis rules that hypothesis out, you are “effectively”—not technically, not formally–employing a rejection region. If you count on such technical/formal distinctions for thinking that you’ve proven Dembski wrong, this is just nit-picking mathematical formalism that you’re using to “miss the forest for the trees”.

So, please answer this easy question: when you say that the 10^-11 probability rules out the ‘1/2 (or, fair) chance hypothesis’, are you, or are you not “effectively” employing a rejection region?

199. 199
Prof_P.Olofsson says:

StephenB[188],
I didn’t notice your previous comment, didn’t mean to evade you. I don’t have much to say though. I still think we’re playing with the word “design” here which we, of course, understand very well as a general concept but only because we know of human design. We have never seen an example of “cosmic design,” unless we already believe in it a priori. Sure, we can say “this looks designed so it must have been designed by somebody who is sort of like us” but I don’t think that’s a valid argument. As our insular friend kairosfocus might put it, we’re in “Plato’s cave.”

I’ll leave it at that, the issue is probably too philosophical for me. I’m more interested in the mathematical and scientific debate.

200. 200
Atom says:

IDSkeptic,

You’re welcome.

K.D. Kalinsky’s paper gives a good real life example from biology dealing with human Intelligent Design.

I don’t think the concepts are as “nebulous” as you think. Sure, sometimes we have to estimate the functional target space, but it isn’t unusual to approximate and give upper/lower bounds in science.

You should give KD’s paper a good read and see if it doesn’t become a lot clearer. (Perhaps you just haven’t seen FSCI/CSI presented in a clear way up to this point?)

Atom

PS Your point about “all natural processes” could come across as the “magical ordering principles” objection that was slain earlier in this thread. Either you have evidence for something or you don’t. One can always hold up scientific progress in the hope that a new discovery will overturn our current theories, but it is better to explain the current facts based on what we currently know, not based on what could possibly be true.

201. 201
Sal Gal says:

gpuccio,

It is absolutely enough to consider much simpler examples, where all considerations are easier. Or are you suggesting that, if a single protein need to be designed, a whole genome could arise by chance?

There’s a subtle and seductive fallacy that is generally present in the argument from improbability, of which the argument from specified complexity is a species. There is no accounting for the framing of the system of interest.

The “best” statistical model of empirical data does not maximize the probability of all events. Loosely, explanation is a matter of shifting probability mass from some events to others. It’s generally easy for an adversary who seeks to highlight the improbability of some event or another to find an event of low probability. That is, the scientist does not construct the model in an adversarial fashion, and the person arguing from improbability is an adversary. At first blush, it seems that a scientist defending against all adversaries, rather than attempting to explain, would go with uniform probability — just what Dembski likes. In other words, perhaps the best way to defend against an argument from improbability is to explain nothing.

In his latest version of CSI, Dembski includes an ad hoc “pattern dredging” penalty.* This does not account fully for “drawing the target around the arrow.” It accounts for definition of events, but not for selection of a particular event.

In the theory of Kolmogorov complexity, a theorem states that a string of 0’s and 1’s is algorithmically random only if it contains a logarithmically long substring that is very orderly (algorithmically compressible). If you home in on that substring, or certain substrings of that substring, you are going to infer design of an entity that is in fact necessary for global randomness. It’s not just the naming of the pattern you see, but your framing of it in the first place. I don’t think Dembski has accounted for the framing.

*If memory serves, there is an ordering of the descriptions the semiotic agent potentially applies to events. The penalty is the negative logarithm of the position of the description in the ordering. Why adding this quantity to “complexity” (the self-information of the event) is appropriate is an utter mystery to me.

202. 202
Prof_P.Olofsson says:

PAV[198],
Didn’t mean to be smug, just pointing out the obvious fact that most people here, including you, don’t have the required background to discuss advanced statistical matters.

As for Dembski, he certainly misrepresents Bayesian analysis. Whether it is on purpose or out of ignorance, I don’t know.

The 10^-11 probability is computed in the posterior distribution. It is not the probability of a rejection region. A rejection region in this example is a subset of the possible sequences of Ds and Rs, such as E*. There is no such set that corresponds to the 10^-11 probability.

Your talk about “same prior and posterior probabilities” illustrate further that you don’t understand the topic. There are no priors or posteriors in hypothesis testing.

The two approaches are entirely different, the crucial difference being whether p is viewed as an unknown constant or as a random variable. The fact that small probabilities indicate ulikely events have nothing to do with it.

Easy question for you: Do you understand where the 10^-11 probability comes from?

203. 203
IDskeptic says:

Cool chatting with you Atom – a really appreciate the collegial tone.

I’ll go over the articles in depth tonight.

I must just comment on “Either you have evidence for something or you don’t.”, though. ID is essentially about ‘not having evidence for X, Therefore Y as far as I can tell.

204. 204
JT says:

I’m still trying to get the following answered, if anybody feels like they have a good handle on the paper Dembski mentioned in the thread last night.
Is the implication that if a sequence exhibits any property that is highly unusual (e.g. compressibility) then that is an indication of design (given a certain level of probabilistic resources and presumably not including properties only present in a finite set of sequences).

——————————-

From “Specification: The Pattern that Signifies Intelligence”, Dembski, p. 11:

[The vast majority of] sequences are … algorithimically incompressible. It follows that the collection of [compressible] sequences has small probability among the totality of sequences so that observing a [compressible] sequence is reason to look for explanations other than chance.

Is the truth of the above based on the assumption of the following proposition:

For all properties P:
If the vast majority of sequences do not have property P then it follows that the collection of sequences that do have property P has small probability among the totality of sequences so that observing a sequence that does have property P is reason to look for explanations other than chance.

Would this propostion be true for a property of being a specific sequence, or only for a property held by an infinite number of sequences.

205. 205
Atom says:

IDSkeptic,

No worries; I’m only here to help people think through issues of design, not to start fights with people I don’t know. (I’m not rude in person, so why be rude on the net?)

ID has both negative and positive parts, as gpuccio mentioned before. Sure, we rule out unguided mechanisms based on our full (current) knowledge of those mechanisms, their capabilities and how they behave. But then having eliminated unguided mechanisms, we test the capabilities of guided (intelligent) mechanisms. We see that intelligence as a causal class can achieve the effect(s) we observe, so therefore, intelligence becomes the best (tentative) answer. Furthermore, by studying examples of intelligent causation and artifacts, we can learn whether or not their are objective signs of intelligent activity. If the effects we see also share these “signs”, our confidence in the Design inference becomes that much stronger.

206. 206
Prof_P.Olofsson says:

I don’t know if anybody is enjoying my exchange with PaV, but here is the issue:

In the Caputo example, there is an observed sequence of 40 Ds and 1 R. This single sequence, call it E, has probability less than 1 in a trillion. However, in order to rule out a fair drawing, we need to put it together with all other sequences that carry the same, or more, evidence of cheating. This set of 42 sequences, call it E*, has a probability of less than 1 in 50 billion. In statistical jargon, E* is a rejection region. This is how it’s done in hypothesis testing; whatever the observed sequence is, it must be put together with other sequences to form the rejection region.

Dembski’s claim is that a Bayesian analysis must also use rejection regions such as E*. This is his claim of “parasitism.” However, he is wrong. A Bayesian analysis uses a prior distribution of the parameter p, then computes the probability of the single sequence E given p, then put the two together into a posterior distribution of p. Inference is now done by computing probabilities in the posterior distribution (which, I remind, is a distribution for the probability p). In my article, I give one example of how a Bayesian analysis is done. Whatever the observed sequence E, it will never be put into a set E*.

Maybe somebody else can chip in and help explain.

Or maybe we can just leave it at that.

207. 207
tribune7 says:

Clearly, Professor, a Bayesian analysis shows that there is no cheating by Democrats in New Jersey.

Anyway, it sounds like you and Dembski are having a dispute over definitions (Bayesian means this, no it means that) and is irrelevant to the interpretation of the results.

208. 208
Prof_P.Olofsson says:

tribune[207],
Clearly, Caputo=Capone+Bluto.

It’s not about defintions and it’s not a dispute.

209. 209
Rude says:

ID Skeptic 203: “ID is essentially about ‘not having evidence for X, Therefore Y’ as far as I can tell.”

Not quite. ID says that design is design until someone negates it. Darwin foolishly thought he could, and right or wrong materialism embraced him with all its heart.

But we are designers. It’s what we do. Even as we talk, sequencing one novel clause after the other, we are designing. The whole of humanity is a laborotory for the study of design, its production, its character.

Materialism maintains that biological organisms arose via chance and necessity, or for some now maybe just necessity (front loading all the way down), and that the human mind that produces design simply “emerges” from neurons as bile emerges from the liver.

Evidence? There ain’t none, none whatsoever, but you gotta have faith because materialism is the law of the land. For isn’t there a judge who said so?

210. 210
tribune7 says:

Clearly, Caputo=Capone+Bluto.

Good. And just as clearly we can conclude amino acids don’t form into a protein by chance, right?

It’s not about defintions and it’s not a dispute.

When you say: “As for Dembski, he certainly misrepresents Bayesian analysis. Whether it is on purpose or out of ignorance, I don’t know” it seems kind of personal.

211. 211
Prof_P.Olofsson says:

tribune[210],
I shouldn’t have written “ignorance,” I realize now that didn’t sound nice. I have no personal quibbles with Bill D.

212. 212
tribune7 says:

I shouldn’t have written “ignorance,”

Or “on purpose” 🙂

Now, can we clearly conclude amino acids don’t form into a protein by chance?

213. 213
Prof_P.Olofsson says:

tribune[212],
No.

214. 214
JT says:

(206 – PO)

I’m going throught the addendum (Addendum 2: Bayesian methods) that specifically addresses all this at the end of Dembski’s paper mentioned above.

So one would compare P(E|D) and P(E|C) and go with the one that was larger, where D = design and C = chance. Dembski also says its actually necessary to compute P(E*|D) and P(E*|C). But whether its E or E* If the only thing you knew about an event is that it was designed what could that possibly tell you about its probability that you didn’t know previously? A designer can design uncompressible strings as well. It seems that P(E|D) would always equal P(E|C). Dembski says himself in this addendum that “Bayesian methods are inadequate to draw design inferences”, so I guess his only choice is to denigrate them.

215. 215
Prof_P.Olofsson says:

As for Bill D and Bayesianism, my most benevolent interpretation is that he thinks about using the event E* in a computation with Bayes’ formula. However, that’s not what Bayesian inference is about.

216. 216
JT says:

“A designer can design uncompressible strings as well”

Maybe not – anything generated by an algorithm is pseudo random, so maybe if we knew it was designed we would know it couldn’t be truly random. But then nothing is truly random.

217. 217
JT says:

Sorry to come to this discussion late- I get what Dembski is saying now: If you knew something was designed that would be based on background knowledge, such as that the sequence was compressible.

218. 218
gpuccio says:

Sal Gal:

#196:

About Allen MacNeill and his “list of engines”: I have discussed a few times with him here and I must say that, even if he is certainly a “fair” interlocutor, I remain very much disappointed of his scientific approach and of his way of dealing with problems (just my opinion). McNeill seems to be an enthusiast of any vague concept which smells of novelty and appears to be trendy, but if you ask him to substantiate his claims, you obtain nothing. He seems not to be interested in causal models, to the point of comparing the science of origins more to a natural history than to a true scientific discipline.

That said, the idea that non genetic factors (epigenetics) and adaptation (neo-Lamarckinsm) may play a role in evolution is certainly interesting, and it can have a role both in darwinian theories and in design theories (indeed, more in the second). But if you think that what I call “neo-neo-darwinism (McNeill, evo-devo and similar) has significantly contributed to ranovate the causal paradigm of neo-darwinism, then again you are not realistic.

Whatever the approach, whatever the reasoning, it is sadly true that, out of ID and classical neo-darwinism, no other explicit theory exists of how biological information could have emerged. Whatever you may say, biological information (or at least the part we know) is written in DNA. The sequence of proteins is written in DNA. We know no other place where it is written, or where it could have originated. Therefore, all my examples about protein genes and protein sequences remain valid, and your discourses about the non centrality of genetics and the citation of McNeill’s hyperboles seem to be still another diversion from the important issues.

219. 219
tribune7 says:

Now, can we clearly conclude amino acids don’t form into a protein by chance? . . . no

OK, professor, I’m not the statistician but are you saying that because we have never seen amino acids form a protein we don’t know the prior distribution of p hence we can’t compute the probability of amino acids forming one?

220. 220
PaV says:

PO[206]:

No one needs it pointed out to them that Bayesian analysis and Fisherian analysis are different. The mathematical approach is clearly different. That should be clear to all. No one disputes that. But the fact that Bayesian and Fisherian approaches are different is neither here nor there when it comes to the fundamental question Dembski raises. What’s at stake is which approach should be used—elimination or comparison. Dembski, in making his choice of elimination, points out the pitfalls of the Bayesian/likelihood approach (Sober) when it comes to making inferences, and then goes on to show that Bayesians are, in the end, dependent on Fisherian analysis and its concomitant rejection regions. But he nowhere claims that, within their statistical methodology, Bayesians use a formal rejection region. What’s, then, at stake is this, where you write:

Dembski’s claim is that a Bayesian analysis must also use rejection regions such as E*.

I don’t see where Dembski anywhere makes such a claim. He claims that, ultimately, effectively, Bayesians are “parasitic” on the rejection region E*, the reason being that it is this rejection region that Fisherian analysis would rule out as completely improbable (since anything with probability less than 1% or .1%, as you’ve indicated, is rejected), and that, without admitting it, Bayesian’s, when analyzing event E, and then coming to the conclusion that it should be “rejected”, are simply being “parasitic” on the Fisherian analysis, which, prior to analysis, sets up a clear rejection region which, in the Caputo case, is, ,a priori, comprised of events E*.

That’s his claim. And this is not what you claim Dembski is saying.

Now, again, can you answer this easy question (or are you avoiding it?): when you say that the 10^-11 probability rules out the ‘1/2 (or, fair) chance hypothesis’, are you, or are you not “effectively” employing a rejection region?

221. 221
gpuccio says:

Sal Gal:

#201:

I am not sure I understand fully what you mean here, and again I will leave some specifics, which really don’t seem to be relevant to my discourse, to Dembski or others who have the necessary competence.

I must say. however, that I am again disappointed by your affirmation that the argument of CSI is an “argument from improbability”. I disagree. It is rather the confutation of an “argument from probability” which has been affirmed for dacades and which is blatantly false.

All of you who make that kind of argument seem to forget that it is darwinian evolution which has used an “argument from probability” where the probability does not exist in appropriate measure. You seem to forget that it is the model of darwinian evolution which pretends that random variation may create from noise functional information complex enough to be selected. That’s simply not true. Confuting that lie is not an “argument form improbability”. It is sound scientific reasoning.

You seem to forget, again, that the specification we find in proteins is a “functional” specification, and not a specification due to compressibility. Therefore, whatever objections you may bring to Dembski based on reasonings about compressibility can be true or false, but they bear no relationship to the biological issue.

According to darwinian theory, nucleotides vary for random mechanisms, from point mutations to deletions and insertions, duplications, and so on. According to darwinian theory (it’s not me who am saying that!) no such mechanism of variation is in any way related to function. According to darwinian thery, there must exist a model where we can see how such random variation has the power to create the kind of information we observe, even with the help of NS, where that help can realistically apply.

That is darwinian theory, not mine. That’s the theory which has “radically changed” modern science! And if that theory is not able to produce one single credible model, and hides behind myths with the complicity of all the academic envirnment, that’s certainly not the fault of any “argument from improbability”. The only improbable thing, here, seems to be that I may ever receive any explicit response from you about these arguments (if possible, without your appealing to MacNeill’s authority).

It’s a pity, because I hold you in great esteem.

222. 222
Prof_P.Olofsson says:

PaV[220], You write:

What’s, then, at stake is this, where you write:

“Dembski’s claim is that a Bayesian analysis must also use rejection regions such as E*”

I don’t see where Dembski anywhere makes such a claim.

The best way to see it is probably to read what Demsbki writes, that’s what I did. “Elimination vs. Comparison.” Page 11:

This event—E*—consists of 42 possible ballot line
selections and has improbability 1 in 50 billion. It’s this event and this improbability on which the New Jersey Supreme Court rightly focused when it deliberated whether Caputo had in fact cheated. Moreover, it’s this event that the Bayesian needs to identify and whose probability the Bayesian needs to compute to perform a Bayesian analysis.

and

E* is the rejection region (and therefore specification) that counts the number of Democrats selected in 41 tries. That’s what the court used and that’s what Bayesians use.

and on page 12:

Bayesians routinely consider such composite events

(If they do it “routinely,” just one little example would be easy to provide, wouldn’t it? I’ve asked for it before but never gotten any.)

So when you say

Bayesians are “parasitic” on the rejection region E*

it’s not true because they never use E*, never ever.

223. 223
Prof_P.Olofsson says:

PaV[220],
I’m breaking up my comments to keep them from running too long. You go on:

Now, again, can you answer this easy question (or are you avoiding it?): when you say that the 10^-11 probability rules out the ‘1/2 (or, fair) chance hypothesis’, are you, or are you not “effectively” employing a rejection region?

224. 224
Prof_P.Olofsson says:

JT[214],
There are obvoius problems with likelihood inference or Bayesian inference, as I mention in the article that started this thread. What is P(E|D)? How could we possibly know? Some might claim that P(E|D)=1 because the designer would never do it differently. Some might claim that P(E|D)=0 because a designer would never do it this way. And what is P(D)? Richard Dawkins would say 0, Ken Ham would say 1. Dembski brings up these points and we are in perfect agreement. However, he then goes further trying to completely discredit Bayesian inference based on a misrepresentation on what Bayesian inference really is.

225. 225
Sal Gal says:

Prof. Olofsson,

I wanted to reread Haggstrom, and just did. I actually don’t think his response to Dembski is fair. He indicates, without citing sources, that that the uniform is sufficient, but not necessary, for NFL. He seems to pull “exchangeability” out of thin air as another sufficient condition. If he had cited sources, he’d have revealed that the condition appeared in the literature in 2004, two years after Dembski’s No Free Lunch.. It is actually necessary as well as sufficient, and has been stated as P(f) = P(f o s) for all functions f and all permutations s of Dom(f).

Haggstrom should have used the 2004 result, which he obviously knew, to strengthen Dembski’s argument. I consider this standard practice. All Haggstrom does in rebutting Dembski is to show unreasonable the assumption of a uniform distribution on fitness functions. In 2005, Dembski perhaps would not have insisted on a uniform distribution. (That said, I will mention that the sufficiency of independent and identically distributed f(x), not necessarily uniform, for NFL was proven in 1996. BTW, how does i.i.d. uniform f(x) constitute less prior knowledge that i.i.d. f(x)?)

A key issue Haggstrom does not address, no matter that it was emphasized by Wolpert and Macready, is that the existence of regularity (structure) in a fitness function (landscape) does not ensure that the search algorithm exploits it. Dembski and Marks are correct in criticizing him for this omission.

Actually, if you are going to choose the algorithm arbitrarily, uniform P(f) is a wonderful circumstance. Almost all functions are algorithmically random, or nearly so. It follows that for almost all f, almost all algorithms obtain a very good solution with a modest number of function evaluations — and the number does not depend on the size of the search space! (A similar result appeared in the NFL literature in 1996.)

There’s a much higher probability of severe mismatch of algorithm and fitness function when the function is algorithmically compressible (realizable in the known universe) than when it is algorithmically random (“typical” in mathematical abstraction, but physically unrealizable). Haggstrom actually strengthens the hand of Dembski and Marks by making a case for the compressibility of the fitness function (“clustering” of fit individuals in the search space). Again, the choice of search algorithm was a non-issue under Dembski’s (and Wolpert and Macready’s) assumption of uniform P(f).

226. 226
StephenB says:

—–Professor O: “Sure, we can say “this looks designed so it must have been designed by somebody who is sort of like us” but I don’t think that’s a valid argument. As our insular friend kairosfocus might put it, we’re in “Plato’s cave.”

With respect, I must remind you that I presented you with scientific evidence which you rejected on philosophical grounds. That FSCI is an empirically anchored fact is undeniable. That it appears in nature and in the work of human designers is also a fact that can be observed and studied on that basis. If you think that there is another scientific explanation for FSCI in cosmology and biology, reason dictates that you at least hazard a guess as to what it might be, even if you disagree with the probability estimates associated with the arguments.

—–PO: “I’ll leave it at that, the issue is probably too philosophical for me. I’m more interested in the mathematical and scientific debate.”

You dismissed a scientific fact with a philosophical shrug. The fact remains that FSCI is all around us and you wave it off it without a qualm. It is just as easy to ignore the facts of science by immersing oneself in mathematics as it is to ignore mathematical arguments by immersing oneself in scientific facts. Unless I misunderstand you position, you are not only arguing against Dembski’s mathematical approach; you are also arguing against the significance of FSCI in general. Under the circumstances, both subjects should be fair game, should they not?

Further, the problem of comparative anlaysis cries out for an answer: Which scientific theory do you find more problematic? Is it ID’s precise mathematical formulations, which are concrete enough to be tested by a professional statistician, or is it Darwinism’s non-existent mathematical formulations, which are too abstract to be tested by anyone?

227. 227
Prof_P.Olofsson says:

Sal Gal[225],

He seems to pull “exchangeability” out of thin air as another sufficient condition. If he had cited sources, he’d have revealed that the condition appeared in the literature in 2004, two years after Dembski’s No Free Lunch.. It is actually necessary as well as sufficient, and has been stated as P(f) = P(f o s) for all functions f and all permutations s of Dom(f).

Which is equivalent to exchangeability. Haggstrom doesn’t pull it out of thin air; it’s more or less trivial as you want the joint distribution of (f(x1),…,f(xn)) to be unchanged under permutation which is the very definition of exchangeability. Although more general than i.i.d., exchangeability is still completely unrealistic in biology.

BTW, how does i.i.d. uniform f(x) constitute less prior knowledge that i.i.d. f(x)?)

Good point! Of course it constitues more prior knowledge because it is more restrictive. It is a common misconception that assuming a uniform distribution means “assuming nothing” when, in fact, it is a model assumption just like any other.

228. 228
Prof_P.Olofsson says:

StephenB[226],
I didn’t mean to make you upset. I just don’t have much to say on this particular point. What scientific fact did I dismiss? I have never argued against FCSI; in fact, I don’t even know what the acronym stands for.

229. 229
JT says:

Prof Olofsson –

re Dembski on E*:

“it’s this event that the Bayesian needs to identify and whose probability the Bayesian needs to compute to perform a Bayesian analysis.”

“That’s what the court used and that’s what Bayesians use.”

“Bayesians routinely consider such composite events”

What Dr. D means is that they are doing this in effect or implicitly. He uses some of the same quotes in Addendum 2 on Bayesian Methods (from the aforementioned “Specification…” paper) and then in the closing sentence of that section, makes his implicit use of “implicit” explicit:

Specification’s role in detecting design, far from being refuted by the Bayesian, is therefore implicit throughout Bayesian design inferences.[emphasis added]

I did not follow his reasoning but will take another look at it.

230. 230
Sal Gal says:

gpuccio,

I do not accept that there is information in any material entity. There is information in observation and interpretation. Give an alien intelligence a DNA strand, and the intelligence probably responds, “Huh?” Then again, an embodied alien possibly tastes the DNA and says, “Sweet!”

231. 231
StephenB says:

—-Professor O: Why would you think that I am upset? My style is to waste few words, so I don’t normally begin with a preamble exhorting everyone to be at peace, though I hope that they are.

Sorry about the FSCI (functionally specified complex information.) Just call it CSI (complex specified information) I refer to the pattern which is always left when humans design things. It also appears in nature, which serves as a strong indication for most people that randomness may not explain all that there is.
So, It seemed odd to me that you would question the mathematics behind the principle and yet have no interest in the principle itself, especially in light of the fact that you were obviously once interested in it long enough to find a rationale for dismissing it. You imply, for example, that ID suggests that “someone like us” did the designing. Obviously, that is not the case, so naturally I wondered how you came up with that.

Also, I thought that this, too, was a fair question: Which scientific theory do you find more problematic? Is it ID’s precise mathematical formulations, which are concrete enough to be tested by a professional statistician, or is it Darwinism’s non-existent mathematical formulations, which too abstract to be tested by anyone?
Still, you are free to be selective about which things you care to discuss. Freedom of speech is a beautiful thing, and most of us enjoy that privilege, unless of course we want to argue for ID in an academic environment. So, you are definitely entitled to not be interested in the subject matter that drives the entire debate.

Meanwhile, I bid you a good evening as I withdraw from the dialogue until tomorrow.

232. 232
JT says:

Here are some quotes from Bill D. from “Addendum 2 on Bayesian Methods” in “Specfication: The Pattern that Signifies intelligence”.

Its hard to imagine who would be arguing against him on this point. Richad Dawkins? Bayesian Theorists? Or was he just really coming down hard on other ID proponents.

Here in this addendum, I want to focus on what I regard as the most damning problem facing Bayesian approach to design detection, namely that it tacitly presupposes the very account of specification that it was meant to preclude.

In the above Dembski seems to be following the strategy that the best defense is a good offense implying that Bayesian inference is pernicious to an extreme degree. However, in the same sentence he admits that the Bayesian approach was specifically intended to preclude the types of specifications he employs.

Bayesian theorists see specification as an incongruous and dispensable feature of design inferences”

“… a Bayesian use of specification might look as follows: given some event E and a design hypothesis D, a specification would assist in inferring design for E if the probability of E conditional on D is increased by noting that E conforms to the specification. [emphasis added]”

So he didn’t get this Bayesian use of specification from any of the actual Bayesian theorists he mentions by name. Rather, he speculates what that usage might be. This is the Bayesian usage of E* he is actually alluding to the rest of the addendum, this hypothesized version he created. And he is asserting the Bayesian theorists would just use his own method for inferring the prior probability that E was designed. So the biggest fault of Bayesian theorists is using Dembski’s own method to compute prior probabilities of design without giving him credit (and denying they’re doing it to boot.) At least in this speculative scenario that Dembski proposes.

Looking at the above now, it seems like I must have read it somewhere before, though I couldn’t say where.

I thought it was my own, but I’m not that mean. I’m just mad because nobody’s answered my question in 204.

233. 233
Sal Gal says:

Prof. Olofsson,

To be a bit more blunt, Theorem 1 of Wolpert and Macready’s “No Free Lunch Theorems for Optimization” says, in essence, that the distribution of the observed sequence of fitness values does not depend on the search algorithm. The notion that it says something about averages is mathematical folklore, and Haggstrom repeats it while claiming to explain what NFL is “really about.” This is a minor transgression, because the folkloric NFLT does follow from the actual theorem. The zinger is Haggstrom’s repetition of the utterly false notion that the average time to obtain a satisfactory solution does not depend on the algorithm. I repeat my quote of Wolpert and Macready:

[T]he analysis… does not concern itself with the computational engine used by the search algorithm, but rather concentrates exclusively on the underlying statistical nature of the search problem. The current probabilistic approach is [complementary] to computational complexity. Future work involves combining our analysis of the statistical nature of search with practical concerns for computational resources.

How closely did our tutor read the one paper, dated 1997, on NFL he cited in his 2006 paper on NFL? Did he not Google “no free lunch theorem” and go to the top-ranked site, no-free-lunch.org, which points to the leading references on NFL?

Haggstrom doesn’t pull it out of thin air; it’s more or less trivial as you want the joint distribution of (f(x1),…,f(xn)) to be unchanged under permutation which is the very definition of exchangeability.

Excuse me for doubting that someone who misses what Haggstrom misses easily saw what entered the literature only after 9 years of research into NFL by a number of bright people. No one has given a simple proof that there’s NFL iff the distribution on function is invariant under permutation of the domain. Perhaps Haggstrom can give a simple proof of the 2004 result by drawing on results in the 2005 book on probability theory he cites for “exchangeability.” In any case, my point that he should have strengthened Dembski’s argument stands.

Although more general than i.i.d., exchangeability is still completely unrealistic in biology.

It is physically unrealistic. Streeter gave a good discussion in 2003 (“Two Broad Classes of Functions for Which a No Free Lunch Result Does Not Hold”). English made similar points in two 2004 papers regarding invariance under permutation. However, it is not so clear that “exchangeability” cannot hold approximately for relatively small subsets of the domain. (Think in terms of n-wise independence, but not (n+1)-wise independence, of distributions.) Was it not obvious to Haggstrom that there might be NFL for non-exhaustive search, even when there is not NFL for exhaustive search? Gee, obvious stuff is much more obvious after someone points it out than before, isn’t it? 😉

If I were merely underwhelmed by the paper, I would not be grousing. If Haggstrom had merely gotten the matter of NFLT’s applicability to algorithm running time wrong, I would not be grousing. It is the combination of his tone of high authority, his neglect of literature review, his failure to strengthen Dembski’s argument when it was appropriate, and his crucial error that really bugs me.

234. 234
Sal Gal says:

Prof Olofsson, I wrote,

BTW, how does i.i.d. uniform f(x) constitute less prior knowledge that i.i.d. f(x)?)

You replied,

Good point! Of course it constitues more prior knowledge because it is more restrictive. It is a common misconception that assuming a uniform distribution means “assuming nothing” when, in fact, it is a model assumption just like any other.

I saw a claim in an online forum that it’s been established that there’s not NFL in inductive learning for iid f(x). Wolpert proved that there’s NFL for bias-free learning. Supposedly there are generally-superior biases for iid, but not iid uniform. I find it interesting — a bit disquieting, actually, because I do not understand — that things work differently in learning than in search.

235. 235
Sal Gal says:

Sal Gal: “that really bugs me.”

Let me be the first to point out that I really do need to get a life.

236. 236
PaV says:

PO:

Easy question for you: Do you understand where the 10^-11 probability comes from?

From a normalized binomial distribution:

41[43/41][(1/2)^40]x[1/2]= 1.995 x 10^-11.

Your answer to my easy question was “not”. I think to say that you are not “effectively” using a rejection region, when, in fact, you’re rejecting a chance hypothesis based on a calculated imporbability level, means we have little we can discuss honestly.

Look at JT’s comment in [229].

237. 237
Mark Frank says:

Your answer to my easy question was “not”. I think to say that you are not “effectively” using a rejection region, when, in fact, you’re rejecting a chance hypothesis based on a calculated imporbability level, means we have little we can discuss honestly.

PaV 236

I am going to have a go as you seem to be out of sync with PO. Forgive me if I “explain” things which are already obvious to you – it is hard to know where to start.

I think maybe you are confusing two uses of the word “reject”. An informal ordinary English sense – and the formal sense as used in hypothesis testing.

If you do a Bayesian analysis and calculate that the probability of a hypothesis being true is 10 to minus a zillion then it is reasonable to conclude that the hypothesis is false (actually even this is open to debate under some circumstances but let’s go with it). You can describe this as rejecting the hypothesis in the informal English sense. I am sure that PO would be happy to reject a hypothesis in this informal sense based on a very low calculated probability. But it is not an example of a formal rejection as in hypothesis testing and does not require a specifying a rejection region.

As I am sure you are aware, in the case of formal hypothesis testing you do not calculate the probability of the hypothesis. You set a rejection region based on some criterion and see if the observed results fall into the rejection region. It is a completely different logic. Although it is used widely, this process has a long history of conceptual and practical problems because basically you are not calculating what you want to know. You are calculating the probability of the outcome, not the hypothesis. Hence the guarded and formal use the word “reject”. It might be quite sensible to reject a hypothesis in this formal sense and then for other reasons decide it was probably true.

The key point is that Dembski was claiming that somehow the Bayesian calculation depended on having a rejection region in mind (implicitly or explicitly). One way to demonstrate this is to consider the situation where there are alternative rejection regions. A common example is one-tailed or two-tailed hypothesis testing. In most stats text books there will be some discussion as to when the rejection region for a t-test or similar should be one-tailed or two-tailed. In hypothesis testing the decision to reject or not to reject may well be different in the two cases. However, a Bayesian calculation will be entirely unaffected by the choice and does not have take this into account.

238. 238
gpuccio says:

Sal Gal (#230):

“I do not accept that there is information in any material entity. There is information in observation and interpretation. Give an alien intelligence a DNA strand, and the intelligence probably responds, “Huh?” Then again, an embodied alien possibly tastes the DNA and says, “Sweet!””

OK, this at least is a clear statement. And you are obviously entitled to your personal opinions.

Unfortunately, it is IMO a very untenable statement, and one that almost nobody would accept, and with reason.

So, just for the sake of discussion, I ask you: do you really mean it?

Do you really believe that there is no objective reality which we can call “information” in any material reality? Not in matter itself, obviously, but in the form and order which material reality presents?

Do you really believe there is no information, objectively, in Shakespeare’s works (just to stick to an abused example)? Or in software? (Maybe Bill Gates has a rather different opinion, and lawyers seem to support him)

And above all, do you really believe there is no information in function? That an alien intelligence, seeing “a living cell”, would just swallow it, and never wonder what it is and how it works? (Well, obviously, if the alien is hungry enough, that’s a possibility: there are priorities, after all)

Are you suggesting that the concepts of cause and effect, of function, and so on, are merely human? Just a convention for us inhabitants of earth?

Are you suggesting that our sending messages (or trying to receive them) from aliens is stupid?

You say that there is information in observation and interpretation. That’s true, but does not imply what you seem to imply.

I agree that the information in a poem means nothing if there is not some conscious intelligence which understands its meaning. That does not mean that the information is not objectively in the written poem. It is there as a specific form, but like all information it can be detected only by a conscious agent. But the conscious agent is not making it up in the moment of reading. Otherwise, we could find meaning in any random sequence of letters, and we can’t. Otherwise, different readers could not get the same meaning (approximately) when they read the same poem.

It is true that meaning and information are not there in matter itself. They must be perceived. That’s exactly the point I have made many times. But they are “written” in matter. We know no “disembodied” information in the material world.

But the same is true of things like color, weight, mass, energy, force, and so on. None of these “properties” is in matter itself (that’s certainly bad news for materialists). They are certainly human concepts, but they do refer to some objective reality which we conventionally call “matter”. They are, in a sense, information. Even Schroedinger’s equation is not in matter itself, and yet it is one of our best approximations to what we call “the objective world”.

Would an alien be interested in Schroedinger’s equation, or simply in Newton’s law of gravitation? Or would he just eat it?

So, are you saying that there is no color, mass, weight, energy, force, quantum events, “and” information in any material entity?

In a sense, I agree. those things, including information, are not in matter itself, but in our understanding of reality. Matter itself is only a concept, a category of our understanding of reality.

On that we agree. But how would that be any argument specifically about information? Are you “not accepting” the whole category of science? Or of rational knowlegde? Or simply of knowledge itself?

Or are you just evading my arguments, as usual?

239. 239
tribune7 says:

Mark,

As this thread continues, it seems what this is all about is that somewhere Dembski said something was Bayesian and now others are saying HA HA , no it wasn’t.

And none of this has a whit to do with concluding that it is quite unreasonable to think the amino acids formed a protein by chance yet perfectly rational to infer design for the protein.

240. 240
Mark Frank says:

Re 239

This thread is about PO’s paper. One of the significant items in the paper was his refutation of Dembksi’s claim that Bayesian inference is parasitic upon specification.

It is relevant because much of the argument for ID falls apart if you adopt a Bayesian approach. That’s why it is important for Dembski to try and find holes in it. Otherwise why would he bother?

241. 241
Patrick says:

Note to other readers who want to comment on this conversation:

Raw assertions and links to Panda’s Thumb do not sit well with me. Please a) make an argument yourself and b) do not distract from the the main conversation.

242. 242
Patrick says:

It is relevant because much of the argument for ID falls apart if you adopt a Bayesian approach.

Ignore other schools of thought for the moment. Explain how the “argument for ID fall[s] apart” in itself.

243. 243
tribune7 says:

It is relevant because much of the argument for ID falls apart if you adopt a Bayesian approach. That’s why it is important for Dembski to try and find holes in it. Otherwise why would he bother?

Professor Olofsson’s paper seems to hinge on the meaning of Bayesian and the dispute becomes one of technicalities and semantics. Dembski, in his Caputo example, looks as though he is being generous by attempting to apply Bayesian standards, and Professor Olosfsson even then concedes that a design inference would be correct with Bayesian standards.

I don’t think there is any reason for Dembski to use Bayesian standards in the Caputo example (or ID in general) other than to show the strength of his point.

In fact the rebuttal seems to be HE DIDN’T APPLY BAYESIAN STANDARDS AND EVEN IF HE DID HE WOULD BE RIGHT!!!! DADGUMMIT!!!

With regard to the Behe, Professor Olofsson is critical using probabilities acquired from literature without seeming to ponder why the probabilities from standard evolutionary literature might be so crude yet so accepted.

244. 244
gpuccio says:

Again, for the nth time: 99% of biological and medical science have never used a bayesian approach. Most biologists and mediacl doctors don’t even know what a bayesian approach is. Believe me, that is true.

So, why should ID, which is a biological theory, bother? Dembski probably bothers because he is a mathematician and a statistician. I certainly don’t.

245. 245
Mark Frank says:

242

Explain how the “argument for ID fall[s] apart” in itself.

Patrick – in principle I would love to do this – but it would spark off yet another branch on this very long thread and I think that is bad idea.

246. 246
Mark Frank says:

Again, for the nth time: 99% of biological and medical science have never used a bayesian approach. Most biologists and mediacl doctors don’t even know what a bayesian approach is. Believe me, that is true.

gpuccio

In your medical career have you never come across the sensitivity, specifity of a test and how the positive predictive value varies with base rate? That’s all Bayesian.

247. 247
Patrick says:

Again, for the nth time: 99% of biological and medical science have never used a bayesian approach.

On a side note, that’s changing due to advances in computing speed. See this article from the FDA for an example:

Guidance for the Use of Bayesian Statistics in Medical Device Clinical Trials

248. 248
Mark Frank says:

243 and 244

This thread was about PO’s paper. His point was simply that Dembki was wrong to discredit Bayesian inference in the way he did. You may consider that technical and semantic. However, it is true.

No one is denying that in the Caputo case that there is very strong evidence that he was cheating. Any method that came to a different conclusion would be suspect.

How important is Bayes to ID? Well that goes beyond POs paper. I think it is very important. Bayes asks the right question. How probable is the hypothesis given the data?Hypothesis testing in all its forms doesn’t.

There is a great illustration in the Cohen paper which PO links to above. Look at the case of testing for schizophrenia on page 999. Using pure hypothesis testing then if the null hypothesis is “patient is normal” and they test as schizophrenic there is only a 3% chance of getting this result if they are normal so the null hypothesis should be rejected and they should be carted off to hospital. Using Bayesian techniques it becomes apparent that 60% of people who test as schizophrenic are in fact normal.

Often there are severe practical problems in applying Bayesian techniques. It can be difficult or almost meaningless to apply prior probabilities. But you should be aware of what hypothesis testing is leaving out and how important it is.

249. 249
PaV says:

Mark Frank [237]

I think maybe you are confusing two uses of the word “reject”. An informal ordinary English sense – and the formal sense as used in hypothesis testing.

I’m not confusing them. I’ve stated above (and if you’re coming late to the debate, I can’t blame you for not having read everything on this terribly long thread) that I’m not talking about a “rejection region” in the ‘technical, formal, mathematical’ sense, but rather in an ‘implicit, effective’ way. That is Dembski’s point. He says that Bayesian analysis is “parasitic” on the rejection region; he did not say they’ve set up, analyzed, and made conclusions using the rejection region explicitly.

What Dembski points out is that we, as humans, have an inner sense of when something is utterly improbable, and that this sense is at work whether or not we use Bayesian analysis or Fisherian analysis, which, of course, uses explicit rejection regions. Looked at another way, if Bayesian analysis simply calculates probabilities and went no farther, well, let’s face it, computers can, and do, do this quite well. So, using computers, we have the results, we have numbers flowing from Bayesian analysis. Now what? Well, unless we, as humans apply some kind of ‘rejection region’ when it comes to probabilities/improbabilities, then no conclusion can be reached. I don’t see how anyone can disagree with this. And so once a Bayesian ‘rejects’ a hypothesis, then, implicitly, he/she has a rejection region in mind. IOW, Fisherians put the ‘rejection region’ up front, and Bayesians use it as a last step (although, given the nature of the question being asked, only a relative difference might be necessary, and, then, no implicit rejection region would be required).

250. 250
tribune7 says:

Using Bayesian techniques it becomes apparent that 60% of people who test as schizophrenic are in fact normal.

Mark, is Dembski using null hypothesis in the manner it was used in testing for schizophrenia as per the article or is he using it as advocated by Popper, as per the article, i.e. taking a central prediction and challenging the theory by attempting to reject it?

251. 251
Prof_P.Olofsson says:

PaV[236],
No, that is not where it comes from (and I have no idea what probability it is that you compute).

The 10^-11 figure comes from computing the probability P(p<1/2) where p has a beta distribution with parameters 2 and 41. This distribution is what you get as a posterior when your prior is uniform on [0,1].

I have repeatedly pointed out that a Bayesian analysis starts by viewing the unknown parameter as a random variable, thus having a probability distribution, in this case on [0,1]. I don’t care that you keep insisting that I’m dishonest. I just feel sorry for you that you don’t want to learn.

Again: the idea that a small probability indicates something unlikely has nothing to do what type of analysis you use.

And again: Bayesian inference does not use rejection regions. A rejection region is a clearly defined concept in statistics and it is in this context Dembski mischaracterizes Bayesian inference. Your qualifier of “effectively” has no meaning other than trying to obfuscate the issues.

E pericoloso sporgersi!

252. 252
PaV says:

PO:

I don’t care to learn all about Bayesian analysis–of what use is it to me?

And…..I don’t need to learn it to come to the same conclusion that Bill has come to. I see, and agree to, his logic in all of this.

When you write “Your qualifier of ‘effectively’ has no meaning other than trying to obfuscate the issues”, I can do more than to say that your ‘rejection’ of this qualifier effectively seals you off from the conclusion Dembski makes, and with which I agree. However, I don’t think reasonable people would think Dembski’s conclusion unreasonable. So why the denial?

As to your honesty: can you, in all honesty, write that Dembski has said that Bayesian analysis ‘uses’ the rejection region E* without adding the qualifier “implicitly”? Tell me, is that being honest?

When someone knows better, and pretends not to know better, they’re being dishonest. At least that’s how I see it.

253. 253
gpuccio says:

Mark Frank:

I think it should be clear by now that my intent is not in any way to discredit bayesianism. I am really convinced that probably bayesianism has much to offer. But my point is that you cannot criticize ID for using fisherian hypothesis testing: it remains the cornerstone of actual biological and medial research, and I don’t think that anyone would discredit those sciences for using hypothesis testing as their main tool (OK, maybe sensibility and specificity are bayesian concepts, but no MD is aware of that: thank you for the tip).

So I am simply saying that, unless you use double standard blatantly, you cannot discredit ID because it is based on fisherian hypothesis testing.

Foe the rest I may agree with you, but not when you say:

“This thread was about PO’s paper. His point was simply that Dembki was wrong to discredit Bayesian inference in the way he did. ”

That’s not true. PO’s paper discussed, among other things, Dembski’s views about bayesianism, and that’s exactly the part which I have not dealt with in my review (I usually try to avoid talking about what I don’t know). But the whole purpose of PO’s paper went well beyond that: it was clearly intended as a general refutaion of ID with specific examples of how ID arguments always fail when observed more carefully. The Bayesian issue was really only part of it, and certainly the least important part for the general reader.

Regarding the other points, instead, which were about issue that I think I can understand, I must say that I indeed have dealt with them in detail, one by one, in my post-review which can now be found on the site graciously provided by kairosfocus. And PO has practically not answered my points, even though some minor aspects have been debated here.

During the discussion on this thread, which is reaching some record size, I have repeatedly tried to re-express, clarify and detail my points in all possible manners. I must see that I have not seen in return, not criticisms, not discussions, practically nothing. Most of the debate has been on collateral, minor points. Even Sal Gal, who has seriously and proficiently engaged in the debate, giving many deep insights, seems reluctant to deal with the substantive biological application of all our theoretical chatting.

So, I hope for the last time, I repeat: we are dealing with biology here, with empirical science. Any refined theoretical perspective is welcome, provided that it is in some way referred to the “real thing”.

A final note about PO’s paper: I understand that it was originally intended for a statistical journal, and that explains some of the technicalities. But the reason that it was then published on Talk Reason is that it appears and is intended as a general refutation of ID by a qualified scholar. It is as such that I have taken the liberty of reviewing and criticizing it. It is as such that it has been the object of this thread for (how many are they): more than 250 posts.

254. 254
Prof_P.Olofsson says:

Sal Gal[233],
It is not folklore. W&M’s Theorem 1 states an equality between two sums where summation is over f. Divide both sides by the cardinality of the function space and you have equality between two averages which are expected values under the assumption of uniformly chosen f. Haggstrom’s formulation is equivalent to W&M’s formulation, as Haggstrom clearly shows in his paper. I think Haggstrom’s treatment is simpler because it focuses on the probabilistic properties of the observed values f(x1),…,f(xn). Because these are i.i.d, the average time does not depend on the algorithm so if he makes such a claim, it is true.

The general conclusion in the NFLT is that the joint distribution of the vector [f(x1),…,f(xn)] remains unchanged under permutation, which is precisely the definition of exchangeability. How do we guarantee it? By the type of permutation invariance you mention. It’s intuitively clear that this condition is necessary and sufficient; a formal proof should not be hard to state.

I think the probabilistic approach is very helpful (naturally, because I’m a probabilist) and it certainly shows that W&M’s Theorem 1 is more or less trivial. I’m not an expert in optimization so I’m not familiar with all the extensions and variants of the NFLT. I don’t know if US is the right forum for this discussion, but if you’re interested, drop me an email.

255. 255
Sal Gal says:

gpuccio,

Or are you just evading my arguments, as usual?

No, I get overwhelmed by the length. I tried to give you an accurate, pithy, and thought-provoking response. I was in bed earlier this week, recovering from oral surgery. Now I have work backed up.

I am the person who wrote 13 years ago that conservation of information underlies “no free lunch.” My interest in ID dates to when I learned that Bill Dembski was referring to COI and NFL. It is relatively easy for me to dash off remarks on NFL.

A problem I have with your comments, and those of most UDers, is that you shift in a New York minute from the formal notion of CSI to your informal intuitions of information. The “complexity” of an event in CSI is its self-information, which is defined with respect to some probability distribution on events. That is, for event E, I(E) = -log P(E). I never, ever forget that, and I am always mindful of the fact that the distribution comes from a human being (interpreter of observations). Observation of an event does not give us a probability distribution, no matter the impression you might get from the term “self-information.” It is helpful, I think, to keep in mind that I(E) is sometimes referred to as a surprisal. The observer is surprised, not the event.

The frequentist approach to probability is sometimes called “objective,” but this does not mean that probability or information is in an object. Put simply (perhaps too simply for Prof. O), it means that with repetition of a controlled experiment we can arrive objectively at the relative frequencies of outcomes. But there is never any way to rule out the possibility that with stricter control of the experiment the entropy (average self-information) of the empirical distribution of outcomes will be reduced. When an intelligent (whatever that means) entity improves the control of an experiment, it is not creating information, but exploiting what it has learned through observation of the contribution of extraneous factors to the outcome.

Now if Dembski or Olofsson were to engage me in further discussion of frequentism, I soon would be out of my depth. What I’m much better equipped to talk about is algorithmic information. The problem is that I would go on interminably. I will say, perhaps cryptically, that Bill Gates can launch a disk containing the binary image of Vista into space without much fear of piracy, even under the assumption that intelligent aliens retrieve it. There are several reasons, but I’ll stick with the simplest one. The aliens don’t know the computer. A physical computer observes and interprets a physical system that we call a program. And there exists no algorithm for inferring the computer from the program, or even for inferring that the program is a program.

As I wrote above, “I do not accept that there is information in any material entity. There is information in observation and interpretation.”

BTW, I read someone’s account of how he defined an algorithm for converting DNA sequences to music. Perhaps the Designer has written the “music of the sphere” in genomes, and our preoccupation with life causes us to miss their True Meaning. 😉

256. 256
tribune7 says:

Sal Gal, I like your posts.

257. 257
Prof_P.Olofsson says:

PaV[252],

I don’t care to learn all about Bayesian analysis–of what use is it to me?

Just a crazy thought I had: since you have so much to say about Bayesian inference, you might want to learn what it is. I’m probably way out of line here.

I don’t think reasonable people would think Dembski’s conclusion unreasonable.

Reasonable people who understand the topic would.

As to your honesty: can you, in all honesty, write that Dembski has said that Bayesian analysis ‘uses’ the rejection region E* without adding the qualifier “implicitly”?

Yes. Didn’t you read the quotes I gave before?

But how does the Bayesian identify this event?

referring to the event E*. The answer is that the Bayesian doesn’t, because he never uses it. But feel free to add “implicitly” in as many places as you want, his claims are still not true. Bayesian inference does not use rejection regions, whether explicitly or implicitly or any other itly.

258. 258
Prof_P.Olofsson says:

PaV[252],
You have the following very interesting comment to make, regarding Bayesian inference:

And…..I don’t need to learn it to come to the same conclusion that Bill has come to.

In fact, it’s so revealing I won’t even comment, just let it sit there for everybody to ponder! 🙂

259. 259
Prof_P.Olofsson says:

gpuccio,[253],
You wrote

[…]about PO’s paper: I understand that it was originally intended for a statistical journal, and that explains some of the technicalities. But the reason that it was then published on Talk Reason is that it appears and is intended as a general refutation of ID by a qualified scholar.

Not sure what you mean by “originally intended.” The paper was published in Chance and obviously addresses Chance readers. I’m glad that it ended up in the blogosphere but that hasn’t changed its contents. It is not intended as a general refutation of ID.

260. 260
Atom says:

PO wrote:

In fact, it’s so revealing I won’t even comment, just let it sit there for everybody to ponder!

PO, let’s be cordial. The generous way to interpret PaV’s remark would be that he arrived at the same conclusion using a different method than Dembski, rather than your implied “I don’t know it, so I’ll just agree with Dembski.”

Maybe it was a poor choice of wording, but I think we should give each other the benefit of the doubt when it comes to interpretation of comments and not just look to score quick rhetorical points.

Atom

261. 261
Prof_P.Olofsson says:

gpuccio[244] and Mark Frank [246],

gpuccio said:

Most biologists and mediacl doctors don’t even know what a bayesian approach is. Believe me, that is true.

Oh, I believe you! In fact, there is a famous example how only a small fraction of Harvard medical doctors could correctly apply Bayes’ rule to compute a false-positive rate of 16%. Almost half of them instead gave the answer 95% which was the test’s sensitivity!

262. 262
tribune7 says:

Professor, using Bayes what would be the false-positive rate of proteins forming a flagellum?

263. 263
Prof_P.Olofsson says:

Atom[260],
The little 🙂 was meant to indicate that it was a light-hearted comment. I know that PaV is a nice guy and very passionate, as only an Italian can be!

It’s more than a rhetorical point though. I have very clearly pointed out where Demsbki goes wrong. PaV seems to be under the impression that “rejection region” refers to any situation where we rule out hypotheses based on statistical analysis. However, if you read Dembski’s “Elimination vs. Comparison” chapter, he clearly refers to rejection regions in their formal, statistical meaning. He’s wrong, and I show how a Bayesian analysis is done. I understand that most people here don’t have the necessary background to understand it, but I think it’s a little over the top to claim that you don’t have to understand it, don’t want to understand it, but can still claim that I am wrong.

Now tell PaV to be cordial and stop insinutating that I’m dishonest!

264. 264
Prof_P.Olofsson says:

Prof Olofsson –

What Dr. D means is that they are doing this in effect or implicitly.

He claims more than that, but even if he didn’t, he would still be wrong. Bayesian inference does not use the rejection region even implictly. In Bayesian inference, you only use what you have observed (in the Caputo case, the sequence E) and never have to put it into a rejection region (in the Caputo case, E*). One problem is that many here only know about Bayesian inference from Demsbki’s writing, and if he is incorrect, how would you know it?

265. 265
IDskeptic says:

The talented PO says:

“Oh, I believe you! In fact, there is a famous example how only a small fraction of Harvard medical doctors could correctly apply Bayes’ rule to compute a false-positive rate of 16%. Almost half of them instead gave the answer 95% which was the test’s sensitivity!”

I think you stole that example from yourself – on your own website 😉

266. 266
Prof_P.Olofsson says:

tribune7[243],
Didn’t mean to ignore you…you say:

Professor Olofsson’s paper seems to hinge on the meaning of Bayesian and the dispute becomes one of technicalities and semantics.

It’s not about semantics and certainly not about the “meaning of Bayesian.” All concepts that are discussed are well defined and have precise meanings. It is not about “technicalities” although it is technical by its very nature.

267. 267
gpuccio says:

Sal Gal:

I have really appreciated your post #255. I am sorry if I have pressed you too much, and pleace accept my best wishes for your health.

Maybe sometimes passion for the things I believe in makes my posts too vehement. I apologize for that. But I enjoy most confrontation with interlocitors of value, like yourself. I get rapidly tired of those who go on repeating trivialities.

Well, I will not press you further. Just know that I am perfectly at ease with your definition of information, and with any serious approach to quantifying information in biological beings. Be it as “self information” (which, if I understand correctly, is Shannon’s H) or by a purely frequentist approach, the important thing is to understand that no theory of biological complexity can avoid the quantitative approach. And I am very confident that any serious quantitative approach will decree the end of the classical darwinian theory, and open the way to the search for alternatives. Serious alternatives, like ID (but any other credible theory is welcome). My point is that a theory which declares to be able to explain one of the biggest mysteries in the observable universe cannot hide behind dogma, authority, vagueness, and refuse confrontation with serious refutaions.

By the way, an interesting attempt to apply “self-information”, or something similar, to proteins can be found in a paper by Dursto, Chiu, Abel and Trevors:

“Measuring the functional sequence complexity of proteins”

freely available on Internet.

A sort of preliminary introduction to that work is a previous paper by Abel and Trevors:

“Three subsets of sequence complexity and their relevance to
biopolymeric information”

That too is free on Internet.

I think that’s a promising way to approach the problems of biology. It’s likely that you will find formal problems in those papers, because your approach is naturally rigorous (I hope you never read medical research). And there are certainly errors. Science advances by errors, most times. But the ideas are stimulating, and raise true problems that nobody else usually addresses.

I think we are losing great opportunities because of this stupid war netween darwinism and ID. But you should know who really wants the war. If the academic powers had not tried to suffocate the instances of ID by raw force, and to relegate ID out of the scientific arena, the debate could be deep and stimulating, and science, true science, would profit and prosper. There is a lot of work to do on the information we are gathering from genomes, transcriptomes, proteomes, and various other “omes”. And the flat and innatural perspective of a wrong theory will not certainly help to find the right direction.

Finally, just a note about your example of the computer. In general I agree with you. It is obvious that information (I am using here the term in the sense of meaningful information, let’s say CSI, and not of self-information) cannot be grasped or interpreted even by a conscious intelligent being if the context is completely unknown. You say: “there exists no algorithm for inferring the computer from the program, or even for inferring that the program is a program”. I agree on the first part, a little less on the second (there are ways by which sometimes we can recognize a language even without knowing how to decipher it). But those are details. In general, detection of meaning requires context.

But in our case (biological information) we do have the context. We have the cell, the living beings. We have both the software and the computer. We don’t have dead stings of DNA. We have DNA actively working. That’s why I asked you why an alien should not be intersted to understand how a living cell works. Because he could understand it, just as we do.

Well, that’s enough. I am running long again. Please feel free not to answer this post! Or even not to read it 🙂

268. 268
Prof_P.Olofsson says:

IDSkeptic[265],
Hey, I forgot about that! First I stole it from the New England Journal of Medicine though.

Sorry, Wiley dropped the discount…

269. 269
tribune7 says:

Didn’t mean to ignore you

And as long as you are not ignoring me, Professor how about (262)? 🙂

270. 270
Prof_P.Olofsson says:

StephenB[231],

—-Professor O: Why would you think that I am upset? My style is to waste few words

Very good, my style too!

So, you are definitely entitled to not be interested in the subject matter that drives the entire debate.

I am not uninterested as much as I am unqualified. Other can probably enter that debate with more thoughtful points. I entered this particular thread because it is named in my honor! There is a lot of stuff going on here and I don’t have time to deal with most of it, so I stick to what I know best. I’m not saying you don’t have valid points but I will not attempt to respond to them, sorry.

Meanwhile, I bid you a good evening as I withdraw from the dialogue until tomorrow.

And I wish you a happy return to UD and a good weekend!

271. 271
Atom says:

PO wrote:

Now tell PaV to be cordial and stop insinutating that I’m dishonest!

PaV, stop insinuating that Prof. O is dishonest.

There, now play nice you two. 🙂

272. 272
Prof_P.Olofsson says:

tribune[262],

Professor, using Bayes what would be the false-positive rate of proteins forming a flagellum?

Probably about 7% or 52% or something like it. Or 0.3%.

273. 273
Prof_P.Olofsson says:

Atom[271],
You are one smooth rapper!

274. 274
tribune7 says:

Professor, using Bayes what would be the false-positive rate of proteins forming a flagellum? Probably about 7% or 52% or something like it. Or 0.3%.

Now, seriously 🙂

275. 275
PaV says:

PO [258]:

Tell me, PO, do I need to know all about Beta distributions, solve the integrals involved with posterior probabilities to figure out that at some point Bayesians, just ’cause it seems right to them, say, “Oh, this is too improbable.” The exact moment that you choose to reject a hypothesis as too improbable, then you are making use, implicitly, of Fisher’s rejection regions. That is simply clear. Should you choose to hide behind the fact that nowhere is a rejection region “mentioned” in Bayesian analysis to maintain that you are not doing any such thing, that’s your business. You’re being pedantic. I don’t have a PhD in Statistics. I don’t care to have one. Dembski does. I follow his logic as he explains the bsic moves that Bayesians make, and it makes sense to me. I read your paper, and you rendering of what Dembski has said, and your version of why he’s wrong, and it strikes me as entirely missing Dembski’s point. Why should I listen to you then?
Try convincing me, instead of trying to bully me.

In [207] you write:

Inference is now done by computing probabilities in the posterior distribution (which, I remind, is a distribution for the probability p). In my article, I give one example of how a Bayesian analysis is done. Whatever the observed sequence E, it will never be put into a set E*.

You state that the posterior probability is, in fact, a distribution for the probability p. So, the probability of p=1/2 is 10^-11. Would, then, the probability of 10^-11, that is p=10^-11 have a probability of almost one in this distribution? If so, it would seem then that as we move from p=10^-11 (almost zero)to p=1/2, the probability of this likelihood, according to your posterior distribution around p, should go from almost 1 to 10^-11. If this distribution were graphed, we would see the probability fall off from a probability of almost one at p=almost 0, to a proability of almost zero (that is, 10^-11) at p=1/2. Since you consider this latter probability too low for serious consideration, then what else can be said other than in doing so you have, effectively, marked off a rejection region somewhere along this distribution as the distributuion moves into a “tail” region before, and around, p=1/2. Dembski’s question to you would be: “Where will your cutoff be?” That is, where is the point along this continuous distribution that you decide that the improbability is too low. Once you’ve decided, then, in effect, a “rejection region”—equivalent to a Fisherian rejection region–has effectively been established. For you, obviously 10^-11 is too low, so that the cutoff is at least that low.

276. 276
Prof_P.Olofsson says:

PaV,

The exact moment that you choose to reject a hypothesis as too improbable, then you are making use, implicitly, of Fisher’s rejection regions. That is simply clear.

No, that is simply wrong (sorry for being so direct). First, arguments from improbability were around long before Fisher was even born. Second, more importantly, as pointed out many times, Dembski discusses rejection regions in their statistical meaning, not your “colloquial” sense.

277. 277
Prof_P.Olofsson says:

PaV,

Would, then, the probability of 10^-11, that is p=10^-11 have a probability of almost one in this distribution?

No.

278. 278
PaV says:

Mark Brown:

Re: Shandin’s paper:

If he had used the UPB as the rejction region, he would not have made the mistake of a false positive.

Maybe you can help me: I’m still scratching my head over his P(D|Ho), where Ho=normal people. If it is “given” that the people tested are normal, it it is a “given”, why would you conclude otherwise? It seems to me the way to understand the P(D|Ho) is that there is a chance of 3% (you’ve corrected Shandin properly) of getting a positive for normal people.

279. 279
tribune7 says:

I suspect this thread will be wrapping up soon, and I would like to thank the professor for his participation before it does.

It was enjoyable and interesting.

And while I can’t arbitrate as to whether Dembski or the Professor is right about the use of Bayesian standards, I would like to note that no way, no how can proteins form a flagellum by chance, and whatever standard says you can’t is clearly the right one 🙂

280. 280
Prof_P.Olofsson says:

PAV,

Try convincing me, instead of trying to bully me.

Sorry you don’t want to listen to Atom’s advice.
Anyway, I am trying to convince you and I have had help from Mark Frank but you don’t want to listen. Let’s say that you can say that a Bayesian analysis “rejects a hypothesis” in your colloquial meaning of the concept. It is still irrelevant to Dembski’s claims. He never makes any such claims. He talks about rejection regions in their statistical meaning, as subsets of the type E*. They never appear in a Bayesian analysis. His claim is that you have to form the E* event in a Bayesian analysis. He is wrong. You don’t. You never do. I don’t know how many times I have to repeat it.

281. 281
PaV says:

PO:

What’s the distribution look like then?

282. 282
Prof_P.Olofsson says:

PaV[278],
You cannot use the UPB in any practical application because you can then never reject anything. Nobody would be diagnozed with schizophrenia so obviously no false positives, but no true ones either.

283. 283
tribune7 says:

Nobody would be diagnozed with schizophrenia so obviously no false positives, but no true ones either.

But if all you were looking for was a single false positive and wanted to give to getting one every advantage, the UPB would be be appropriate and very generous too boot, right?

284. 284
Prof_P.Olofsson says:

PaV[281],
Like this.

285. 285
Prof_P.Olofsson says:

tribune7,[283],
I don’t understand what you mean.

286. 286
tribune7 says:

I don’t understand what you mean.

The goal of ID is create a method in which design can be ascertained. To do this the chance of getting a false positive is maximized. To do this the UPB — which is arrived at by multiplying the number of elementary particles in the observable universe, the maximum rate per second at which changes in physical states can occur and the estimated age of the universe times 1 billion — is used.

If a specified event has a probability less than the UPB, it is designed.

This means many items for instance, such as the clause “for instance” that are designed will not register as such but a single thing that is not designed that does register falsifies the theory.

So if you don’t care about false negatives but have a strong concern about a false positive, then the UPB is very practical.

287. 287
StephenB says:

Professor Olofsson @270.

“I am not uninterested as much as I am unqualified. Other can probably enter that debate with more thoughtful points. I entered this particular thread because it is named in my honor!”

Fair enough. Who can argue an overture of humility and an appeal to relevance.

288. 288
StephenB says:

Oops! I mean, Who can argue [with] an overture of humility and an appeal to relevance.

289. 289
Prof_P.Olofsson says:

tribune[286],
Sure, but in practical applications, you can’t use a significance level of 10^-150. It would mean that nobody would ever be diagnosed with any disease, no new medications would ever reach the market, and Caputo would not be considered a cheater!

290. 290
Prof_P.Olofsson says:

StephenB[287],

overture of humility and an appeal to relevance.

You forgot “and a touch of vanity”!

Have a good weekend yall.

(Linguistic parenthesis: I, Swede living in Texas, think the word “yall,” without apostrophe, should be considered a proper English word!)

291. 291
IDskeptic says:

PO, not meaning to derail, but in your medical false positive example, we are told ” that the test is 95% accurate” – isn’t this though an unwarranted assumption that false positives and false negatives have the same odds (ie we’re using an average of two, perhaps asymmetric, mechanisms)?

292. 292
crater says:

I, Swede living in Texas, think the word “yall,” without apostrophe, should be considered a proper English word!

That is fine, but y’all is singular. When referring to a group of people, such as you are here, the correct term is all y’all.

293. 293
tribune7 says:

Sure, but in practical applications, you can’t use a significance level of 10^-150.

It depends on the practical application. If the point is to make it as easy as possible to falsify a controversial new theory it works great 🙂

Caputo would not be considered a cheater!

Which is the idea. Caputo was found to be a cheater with a exponentially less burden of proof than the one Dembski is providing for the design of the flagellum, protein, DNA, the universe etc. 🙂

And if you have gone offline, have a great weekend in TX.

294. 294
PaV says:

PO [284]

“Like this”:

The distribution I describe is but the symmmetric partner of the one you’ve linked to. I wasn’t sure which way it would go, from 1/2 up, or 1/2 down. Nonetheless, the argument is the same, and, looking at the distribution, I feel vindicated.

295. 295
Prof_P.Olofsson says:

PaV[294],
What distribution have you suggested?

296. 296
Sal Gal says:

gpuccio,

I read all of it. You do make me think — that’s another compliment. We’ve agreed on quite a bit in the past. Remember Cloud of Unknowing, by any chance?

Something you made me think of, with your mention of “meaningful information,” is that “semiotic agent” is an event of high specificity — Dembski does indicate that semiotic agents are physical. That is, some semiotic agents can give very short descriptions of a low-probability event that includes themselves. (Surely the probability of materialistic processes giving rise to a semiotic agent is lower than the probability that they give rise to a “bidirectional rotary motor-driven propeller.”) This would lead to inference that semiotic agents are intelligently designed. But the inference invokes a semiotic agent, and is performed by a semiotic agent. Something is amiss here, don’t you think? It appears that design inference requires a designed entity.

297. 297
JT says:

[201] Sal Gal

In the theory of Kolmogorov complexity, a theorem states that a string of 0’s and 1’s is algorithmically random only if it contains a logarithmically long substring that is very orderly (algorithmically compressible). If you home in on that substring, or certain substrings of that substring, you are going to infer design of an entity that is in fact necessary for global randomness. It’s not just the naming of the pattern you see, but your framing of it in the first place. I don’t think Dembski has accounted for the framing.

This seems to be a very severe challenge to Dembski’s work, and very interesting. (Wonder if anyone else in this thread actually read it).

So, any compressible sequence of necessary length to indicate design could in fact be a part of a larger incompressible string. Hopefully everyone understands the significance of that.

But as I write this, now I’m confused. If a substring were compressible, then so would be any larger string it was part of. Now I don’t get it.

I should add that I and everyone else should certainly hope that in the end Dembski’s work proves to be at its core of lasting scientific value. I absolutely do personally hope that.

I thought it was significant that he came on the thread a couple of days ago and pointed out the essay of his which he indicated was the most up-to-date and succinct treatment regarding CSI: “Specification: The Pattern That Signifies Intelligence”. I’ve spent a fair amount of time in it the last couple of days, trying to nail down answers to some questions I’ve had for quite a while regarding the theory, namely because the theory has been presented in so many different forms in the past, and for different types of audiences, with subsequent updates and revisions, etc.

There is one specific aspect of it that seems very troublesome to me if I understand it correctly. First a few quotes:

Are there patterns that, if exhibited in events, would rule out their original occurrence by chance?

To see that the answer is yes, consider the following sequence..

(?R) 0100011011000001010011100101110111000000010010001101000101011001111000100110101011110011011110111100

this sequence was constructed simply by writing binary numbers in ascending lexicographic order

(?R) cannot plausibly be attributed to chance.

The crucial difference between R and (?R) is that (?R) exhibits a simple, easily described pattern whereas R does not. To describe (?R), it is enough to note that this sequence lists binary numbers in increasing order. By contrast, R cannot, so far as we can tell, be described any more simply than by repeating the sequence. Thus, what makes the pattern exhibited by (?R) a specification is that the pattern is easily described but the event it denotes is highly improbable and therefore very difficult to reproduce by chance. It’s this combination of pattern simplicity (i.e., easy description of pattern) and event-complexity (i.e., difficulty of reproducing 16 the corresponding event by chance) that makes the pattern exhibited by (?R) – but not R – a specification.
….
the simpler the patterns and the smaller the probability of the targets they constrain, the larger specificity

It seems clear to me that what the above quotes indicate, is that in Dr. Dembski’s conception of CSI, a sequence of a hundred million 1’s would indicate design more strongly, have higher complex specified information, and be less likely to occur by chance than a hundred million bit long string encoding a chess program.

To me personally, this seems utterly absurd, so in the past I was never even sure if I understood it correctly (maybe I still don’t).

If you have a hundred million bit long string that is all 1’s and the question is did it occur by pure chance or by some other method, then that “some other method” could be a trivial little program that output 1 a hundred million times. Is someone saying that little program is intelligent, or possibly that it could only have been created by someone intelligent? That would truly be ridiculous. The probability of getting a sequence of a hundred million 1’s by chance is equal to the he probability of that tiny program “output 1” occuring by chance. Obviously that would be easier than getting a program by chance that output a chess program.

In all honesty, I just haven’t been able to make it through all the formal parts of the paper yet (maybe I never will), but the previously quoted statements from the informal sections of the paper seem unequivocal. Please someone help me out and tell me if I’m missing something or if what he says in the formal sections of the paper revises everything he says above. (Note: I certainly do understand about things like replicational resources, etc. AFAIK, I actually do understand the theory – its just not competely adding up.

Also I posted what seemed like a pretty simple question in 204 (which could actually be answered merely ‘yes’ or ‘no’) which no one’s answered yet.

298. 298
gpuccio says:

Sal Gal:

Thank you again for the compliments: I am embarrassed.

Was Cloud of Unknowing your previous nickname? I certainly remember that with pleasure…

I agree with you that a semiotic agent is more specified than a flagellum. Indeed, I am convinced that the specification grows exponentially as you progress to higher levels of organization. For instance, the “regulation network” of protein synthesis, whose final effectors are transcription factors, and whose stricture and identity remains largely mysterious, is certainly more complex than the protein sequences themselves.

Just think that we have no idea of how each specific cell at each specific moment can choose a specific “transcriptome” from the same generic genome. That implies:

a) choosing a correct subset of transcribed proteins among the 20000 protein genes of the human genome, including all the possible variations implied by the new “one gene – many proteins” paradigm, which could bring the potential proteom inthe range of the hundreds of thousands.

b) choosing the correct transcription rate which can give the correct relative concentrations of all the proteins selected for the transcriptome

c) choosing, and maintaining, the correct cronological sequence in transcription

d) checking for esrrors and unwanted deviations

Can you imagine how big the search space of possible transcriptomes must be? That’s why we stick to simple examples like protein sequences, which are better understood and whose search space is however in the “interesting” range of 20^100 – 20^1000.

And think of another example which is often overlooked: the spacial and functional organization of the central nervous system. In humans, it is formed of about 10^11 neurons. Each of them is supposed to form an average of 1000 connections, for a total of 10^14 connections. Those connections have obviously to follow some order, and that order seems to be one of the crucial components which allows the expression of a semiotic agent. Can you imagine the search space of 10^14 connections, even if it were only a problem of controling theis spacial order (which it is not)?

Finally, you say:

I agree. That is true, I think of any inference, and not only of design inference. Indeed, I think that God, who is not designed, does not know by inference. So, inference is probably a process inherent in the designed structures used by consciousness to express itself effectively at the consciousness-mind-matter interface.

299. 299
PaV says:

PO:

The distribution I describe in [275]. Yes, since you’re correcting papers, the entire distribution of probabilities will sum to 1. With that correction, though, I think that the distribution you show can’t be right. If you’re graphing p along the bottom, and posterior probability vertically, the peak value should be a p=10^-11, which is as close to zero as you can get, not near p=1 which you show in the distribution. AS I expected, per the distribution you’ve shown, there is “tail” region, much like a normal distribtuion, and in which, of course, Fisherian “rejection regions” are found. This doesn’t bode well for your argument.

300. 300
Prof_P.Olofsson says:

PaV[299],
You do not describe this distribution in [275]. Your talk about p=10^-11 and p=1/2 is meaningless as they are completely different. The second “p” is the probability with which we choose “D.” The 10^-11 figure is a probability about this second “p” which is a random variable. I know it’s confusing to talk about “probabilities of probabilities” but that’s what we’re dealing with here. The 10^-11 figure is the probability that p<=1/2 computed as an integral in the beta distribution I showed you. This distribution is correct (your doubts notwithstanding) and its peak is at 40/41. The peak is near 1 beacuse we observed so many D’s. Finally, there is nothing to “sum” as this is a continuous distribution.

PaV, I have patiently tried to explain Bayesian inference to you, even took the time to plot a graph for you. You said that you don’t want to learn it. Fine, let’s then leave it there and you ought to stop arguing about it. The professor in me says that you might still be interested in understanding the basics so I will send you an email explaining in more detail. As for this particular discussion, I’m done, unless somebody else is interested in continuing.

301. 301
PaV says:

PO:

In earlier posts I suggested that the distribution was of one probability about another. I also indicated that it was a continuous distribution. And, finally, in case you’ve forgotten, an integral is a “sum”.

Again, the pedantry.

As to your posterior distributon: what a bizarre way to analyze the Caputo case. The most “probable” assumption is that the “probability” at play in the Caputo case is that of 40/41 or greater.. Wow. What a strange way of looking at what really happened.

As to your [282], why would you want a “false positive” when you’re dealing with all normals?

As to the email, don’t expect a response.

302. 302
Prof_P.Olofsson says:

PaV[301].

As to your [282], why would you want a “false positive” when you’re dealing with all normals?

You don’t want false positives and you’re not dealing with all normals. You have a test that is such that if somebody is normal, there is a <5% (I assume not much less) chance that he tests positive. So if we use a significance level of 5%, we classify those who test positive with schizophrenia. We catch most cases of schizophrenia but also get a lot of false positives. If we use a very low significance level, such as the UPB, we would not classify anybody with schizophrenia so we’re missing the 2% that are schizophrenic and get no false positives.

303. 303
Prof_P.Olofsson says:

Here’s a question for everybody: In the Caputo case, he was accused of cheating because he drew 40 D’s in 41 drawings that were supposed to be fair. IT was not known how he had done the drawings. If he used a roulette wheel and let 00 denote R and the numbers 0–38 denote D, the observed sequence is quite likely to occur by chance alone.

(1) How do we rule out that he used a roulette wheel (real or computer simulated)?

(2) Why do we need to rule it out?

304. 304
tribune7 says:

If he used a roulette wheel and let 00 denote R and the numbers 0–38 denote D, the observed sequence is quite likely to occur by chance alone.

Well, yes.

(1) How do we rule out that he used a roulette wheel (real or computer simulated)?

Because Caputo was convicted of cheating. If he used a roulette wheel designed to give Dems 39 out of 40 shots to get top ballot position he wouldn’t have been cheating since such an outcome would be expected.

(2) Why do we need to rule it out?

Because if a roulette wheel was used to determine the sequence of amino acids in the protein it would prove design since roulette wheels are designed objects 🙂

Seriously, if there was some necessity that caused amino acids to sequence themselves correctly to form proteins, proteins to form a flagellum etc. that would change things.

But it is safe to say that blind chance cannot account for them.

And what is this necessity, anyway? 🙂

PaV, cheer up.

305. 305
Prof_P.Olofsson says:

StephenB[307],
A few hours later now…maybe this hour is finer!

I think, perhaps, you misunderstand me.

I’m just asking how we can rule out that he did not use a flawed randomization device that would favor Democrats, whether it’s a roulette wheel or any of the things you suggest, or anything else that is chance but not 50-50 chance.

And next, I wonder why we need to rule out such biased chance.

My own answers are (1) we can’t and (2) we don’t have to. Using a roulette wheel would also constitute cheating. All we are really interested in is whether 50-50 chance was at work.

Dembski’s answer is that we rule out (1) by taking Caputo’s word and (2) because we have to rule out all chance with the filter, not just 50-50.

(That is, the old Dembski who wrote The Design Inference. The new Dembski does not believe the filter works.)

Finer? 🙂

306. 306
JT says:

I think the following is an explanation for why ruling out the uniform chance hypothesis for formation of the bacterial flagellum effectively rules out all other chance hypotheses as well.

Suppose you have some set of preexisting conditions, processes laws, etc. that resulted in the formation of a bacterial flagellum. Suppose these preexisting conditions do not have a mind as conceived by ID, but rather are just physical processes. So assume it was these preexisting conditions that resulted in the existence of a bacterial flagellum, and not uniform chance.

If y is any entity with the property of being a bacterial flagellum, then any preexisting conditions capable of producing y can be labelled f(x).

It is the case that Prob(y) <= Prob(f(x)).

So in other words the probability of getting a bacterial flagellum by uniform chance is less than or equal to the probability of getting (by uniform chance) any prexisting conditions that would result in the formation of a bacterial flagellum. This follows from Kolmogorov/Chaitin Algorithmic information theory.

So that’s why you only have to consider the uniform chance hypothesis.

I know that my usage of notation is laughable, but I believe my point is valid.

307. 307
Prof_P.Olofsson says:

PS. StephenB,

Why “better yet”? Actually, I chose the roulette example because it’s a typical chance example and it coincides remarkably well with the observed outcome; the most likely outcome is 1 R and 40 D’s. Other than this amusing coincidence, it has no significance per se.

308. 308
JT says:

I completely misspoke.
I meant to write the following:

It is the case that Prob(f(x)) <= Prob(y).

So in other words the probability of getting a bacterial flagellum by uniform chance is greater than or equal to the probability of getting (by uniform chance) any prexisting conditions that would result in the formation of a bacterial flagellum. This follows from Kolmogorov/Chaitin Algorithmic information theory.

Actually I believe the two values are equal so just takes out the <.

309. 309
Patrick says:

The new Dembski does not believe the filter works.

I wish Bill had taken the time to explain his comment. The only qualifier he added was “pretty much”…which does not explain his position adequately. But to say that “it does not work” or “it is a zombie” is a gross over-simplification.

Before he wrote it I had expressed via email my belief that the old formulation of the EF was too simplistic (which was also pointed out here). This is not so say that it does not work in practical applications but that it’s limited in its usefulness since it implicitly rejects the possibility of some scenarios since “[i]t suggests that chance, necessity[law], and design are mutually exclusive.” For example, the EF in its original binary flowchart would conflict with the nature of GAs, which could be a called a combination of chance, necessity, and design.

In regards to biology when the EF detects design why should it arbitrarily reject the potential for the limited involvement of chance and necessity? For example, in a front-loading scenario a trigger for object instantiation might be partially controlled by chance. Dog breeding might be called a combination of chance, necessity, and design as well. This does not mean the EF is “wrong” but that it’s not accurate in its description for ALL scenarios. The current EF works quite well in regards to watermarks in biology since I don’t see how chance and necessity would be involved and thus they are in fact “mutually exclusive”.

Personally I believe that the EF as a flowchart could be reworked to take into account more complicated scenarios and this is a project I’ve been pondering for quite a while. Whether Bill will bother to do this himself I don’t know.

310. 310
Prof_P.Olofsson says:

Patrick,
What I find interesting is that the premise in The Design Inference is that they three are mutually exclusive. If he now thinks they are not, he obviously does not agree with that book (which I believe was his doctoral dissertation in philosposhy). Many of his critics such as Perakh and Elsberry had pointed it out early and now he seems to be in agreement.

This is all fine, there is no shame in admitting error (the shame is in not doing it) and let your theories evolve, and whether the EF works or not is not tied to Dembski the person.

Pardon my ignorance but what are GA’s?

Maybe I shouldn’t have added the parenthtic comment in [305], I just meant to point out that when I wrote about “Dembski’s answer,” I was referring to his writings, not necessarily what he would write now. I’d still like people here to ponder my points about the Caputo case.

311. 311
Mark Frank says:

JT 308

It sounds interesting but I can’t make sense of what write. Can you give an example?

Thanks

312. 312
StephenB says:

—-Professor Olofsson wrote, “Actually, I chose the roulette example because it’s a typical chance example and it coincides remarkably well with the observed outcome; the most likely outcome is 1 R and 40 D’s. Other than this amusing coincidence, it has no significance per se.”

The fact that you would even consider the subject in that context shows that either your ability to apply mathematics to real live situations is seriously in doubt or else you are willing to resort to irrelevant criticisms to discredit the process of design inference. You don’t have to remind me that it has no “significance per se.” You need to remind yourself of that fact before you introduce it.

Surely, you must understand that anyone who would use a roulette wheel in a manner that you suggest would know beyond doubt that he was cheating. The whole point of your example was to imply that Caputo may have innocently used a random process to get those unlikely results, and therefore may not have been guilty. Do you think that this point was lost? Your question was not a joke; it was a challenge—and the challenge was answered. That is why I dramatized the point with my own similar examples. Thus, it appears that your desire to discredit the explanatory filter compromises your objectivity and—-perhaps your analysis of Bayesian statistics?

313. 313
JT says:

[310] what are GA’s
genetic algorithms

[312]Surely, you must understand that anyone who would use a roulette wheel in a manner that you suggest would know beyond doubt that he was cheating

Could a program be mislabled?

Could a random number generator be malfunctioning?

Why not say the roulette wheel was cheating?

It occured to me, since all random number generators are intelligently designed does that mean even a ballot randomly selected by Caputo was intelligently designed?

314. 314
Prof_P.Olofsson says:

StephenB[312],
Obviously it was a mistake to use the roulette wheel as an example as you are now obsessing over it. Anyway, you misunderstand my point. You write

The whole point of your example was to imply that Caputo may have innocently used a random process to get those unlikely results, and therefore may not have been guilty. Do you think that this point was lost?

That’s not my point at all. The question is not whether he was innocent, of course he wasn’t. The question is now the EF is used in this simple example. We can easily rule out p=1/2 vs p>1/2 with a hypothesis test. After that, I ask (1) how do we rule out the values p>1/2 and (2) why do we have to do it?

I just put these two questions out there for discussion and I don’t understand why you feel the need to resort to personal attacks.

315. 315
tribune7 says:

Professor, to make a quasi-semantic point, chance is not design which is not necessity.

Now, all three could be involved in the coming about of an object or event but the part that was design was not chance.

Why should design be rejected as being even part of the cause of an event if a null hypothesis shows that the probability of an event — if it’s specified i.e. conforms to an independently given pattern as per Dembski — is less than the UPB?

Actually, why should it not be presumed?

316. 316
JT says:

[311]

The objection is that Dembski’s calculations establish the probability of a bacterial flagellum being thrown together at a point in time, such that that molecules randomly floaiting around just by happenstance one day converged into the configuration of a bacterial flagellum, where no type of organism existed before.

But, what if preexisting conditions in physical reality favored the formation of at least certain key attributes of a bacterial flagellum at a rate that was much higher than blind chance. The argument goes that Dembksi’s arguments do not address this, and the probability he calculated could be too low as a result.

Well what I was saying was, suppose those preexisting conditions are such that they directly account for the formation of every key attribute of a bacterial flagellum. IOW, let’s just take it for granted that some identifibable physical process alone, sans ID, can completely account for the production of a bacterial flagellum from nothing. So the probability of getting a bacterial flagellum is equal to 1, and we have this physical process that preceded it to account for it, but now we can’t account for the origin of that physical process. Well what I’m saying is the probability of getting that physical process by uniform chance cannot be greater than the probability of getting a bacterial flagellum by uniform chance. Even if this physical process itself was directly caused by something that preceded it, you will eventually have to hit a point of origin, where nothing preceded it by blind chance or something else, and the probability of that point of origin for bacterial flagellums occurring by uniform chance cannot be greater than the probability of a bacterial flagellum itself occurring by uniform chance.

It should be obvious why the cause for a bacterial flagellum cannot be more likely to occur than a physical flagellum itself. Just thinking about this statement for a few seconds should explain why. But to expand on this, in algorithmic information theory, the probability of a particular binary string C (e.g. 100011001…) is equal to the probability of the smallest program-input that will generate C as output. So that’s why the probability of a cause for a bacterial flagellum is equal to the probability of a flagellum itself.

But actually, I don’t think that Dembski actually computes probabilities this way, because as I pointed out in 297, he thinks that a really long but really simple pattern e.g. (101010101010101010101010101010101010101010101….) could not have been created by chance, but a program that could output such a string is tiny. So there appears to be some fundamental problems with his analysis.

I don’t rule out the possibility that I am in error regarding what Dembski says or in my own analysis above.

One final point – a point of origin for the explanation of bacterial flagellums need not have been created by either blind chance or “intelligent design”. It could have always existed.

317. 317
JT says:

[311]

The objection is that Dembski’s calculations establish the probability of a bacterial flagellum being thrown together at a point in time, such that that molecules randomly floaiting around just by happenstance one day converged into the configuration of a bacterial flagellum, where no type of organism existed before.

But, what if preexisting conditions in physical reality favored the formation of at least certain key attributes of a bacterial flagellum at a rate that was much higher than blind chance. The argument goes that Dembksi’s arguments do not address this, and the probability he calculated could be too low as a result.

Well what I was saying was, suppose those preexisting conditions are such that they directly account for the formation of every key attribute of a bacterial flagellum. IOW, let’s just take it for granted that some identifibable physical process alone, sans ID, can completely account for the production of a bacterial flagellum from nothing. So the probability of getting a bacterial flagellum is equal to 1, and we have this physical process that preceded it to account for it, but now we can’t account for the origin of that physical process. Well what I’m saying is the probability of getting that physical process by uniform chance cannot be greater than the probability of getting a bacterial flagellum by uniform chance. Even if this physical process itself was directly caused by something that preceded it, you will eventually have to hit a point of origin, where nothing preceded it by blind chance or something else, and the probability of that point of origin for bacterial flagellums occurring by uniform chance cannot be greater than the probability of a bacterial flagellum itself occurring by uniform chance.

It should be obvious why the cause for a bacterial flagellum cannot be more likely to occur than a physical flagellum itself. Just thinking about this statement for a few seconds should explain why. But to expand on this, in algorithmic information theory, the probability of a particular binary string C (e.g. 100011001…) is equal to the probability of the smallest program-input that will generate C as output. So that’s why the probability of a cause for a bacterial flagellum is equal to the probability of a flagellum itself.

But actually, I don’t think that Dembski actually computes probabilities this way, because as I pointed out in 297, he thinks that a really long but really simple pattern e.g. (101010101010101010101010101010101010101010101….) could not have been created by chance, but a program that could output such a string is tiny. So there appears to be some fundamental problems with his analysis.

I don’t rule out the possibility that I am in error regarding what Dembski says or in my own analysis above.

One final point – a point of origin for the explanation of bacterial flagellums need not have been created by either blind chance or “intelligent design”. It could have always existed.

318. 318
JT says:

correction:
you will eventually have to hit a point of origin, where nothing preceded it by blind chance or something else

you will eventually have to hit a point of origin, where nothing preceded it but blind chance or something else

319. 319
Prof_P.Olofsson says:

tribune7,[315],
Sorry, I don’t quite understand your quasi-semantic point.

320. 320
StephenB says:

Professor O: I am sorry if I upset you. It was not my intention to hurt your feelings. The objective was simply to call attention to the fact that it is possible to obsess over the math and lose track of the reasons for using it. The original point of the EF exercise was, after all, was to show that that, regardless of the method he used, Caputo was almost certainly cheating. Your example made it appear that he may well have been innocent and that he could have used a perfectly valid method for arriving at his skewed result. Obviously, that is not the case, and since we all now agree that your example was inappropriate, nothing more need be said. So I will be happy to withdraw my comment about your objectivity, since it is not needed to make the point. We will just shrug it off as on oversight.

321. 321
Sal Gal says:

Suppose that New Jersey law called only for random ordering of candidates’ names on the ballot, and that Caputo used an American roulette wheel as Prof. Olofsson indicated (Republican first if and only if the outcome is 00). Then Caputo satisfied the letter of the law.

Our feeling that Caputo violated the spirit of the law indicates our biased interpretation of “random” as “uniformly at random.” We can argue that he “added information” only by invoking our bias in favor of the uniform and by imputing intention to Caputo as an agent. Consider that any formal calculation of CSI cannot take properties of an intelligence into account, and that swapping the Democrats and Republicans on the ballots cannot change the CSI.

Why would we not convict Caputo for violation of election law if he listed Republicans first 40 times, and a Democrat first just 1 time? He would in fact be guilty if he believed mistakenly that the second position is better for the candidate. If he did not believe this, then was he inept, or was he capricious, in the performance of his duty? The only way to know if Caputo caused the probability of listing a Republican first deviated from .5 by intelligent design is to read his mind. Thus there is no empirical basis for inferring intelligent design here.

By the way, if you did not know that ignorant voters tend to select the first candidate on a ballot, would you not be inclined to say that the one anomalous ballot was an error?

322. 322
Sal Gal says:

A related note for readers familiar with Dembski’s technical papers:

Dr. Dembski likes to invoke the Principle of Indifference to justify his analyses in terms of uniform distributions on sample spaces. But the idea in that principle is that a modeler in some sense “adds information” if s/he arbitrarily selects another distribution. To my knowledge, Dembski has never attempted to justify turning the Principle of Indifference on its head and asserting that something has “added information” to the data-generating machinery if the distribution of the data diverges from the uniform.

Now reexamine what I said about Caputo, keeping in mind that his choice of data-generating machinery may in fact reflect indifference to making the random distribution of the data equal to a particular distribution. The uniform distribution is just the average of all distributions. If the actual distribution is not average, why should we attribute this to intelligence rather than indifference?

323. 323
tribune7 says:

tribune7,[315],
Sorry, I don’t quite understand your quasi-semantic point.

You said chance, design, necessity are not mutually exclusive. They are.

While something could have come about by a combination of chance/design/necessity the part that came about by chance did not come about by design, and vice versa.

So chance could have come about in the development of the flagellum as we know it without ruling out design.

So why do we axiomatically rule out design when dealing with proteins, flagellum etc.?

324. 324
JT says:

tribune7, it was Dr. Dembski himself who said above that chance, design and necessity are not mutually exclusive. My guess is that was his euphemistic way of saying he finally realized that design and necessity are indistinguishable (with necessity being his term for mechanism).

325. 325
tribune7 says:

Thus there is no empirical basis for inferring intelligent design here.

That would only be because it involves New Jersey. We should be polite and simply refer to it as design. 🙂

326. 326
gpuccio says:

JT:

Dembski considers all forms of specification. Compressibility is one of them. A sequence like the one you cited is highly compressible. Highly compressible sequences are a tiny subset of the whole search space, and are easily recognizable by intelligent beings. They are specified and, if complex, they cannot come out of purely random causes. In that case we have to consider the whole complexity of the sequence, and not the “compressed” complexity. In other words, you will never obtain that sequence of 01 by flipping a fair coin.

But the problem with compressible sequences is that they can often be the product of necessity. that would be the case if the sequence of 01 were the output of a computer program. So, to define such a sequence as specified one must be sure that no necessary cause is present.

I think that Dembski correctly analyzes all forms of specification because that’s important for his general theory. Moreover, compressible sequences are specially interesting for the fact that they are easily recognized by consciousness.

But, that said, we must remember that all that has no real importance for the question of biological information. In fact, the specification in biological information is of the functional type, and is not linked to compressibility. Functional specification is much easier to treat, and can easily be considered out of the range of necessary laws. It bears no relationship to compressibility. So, functional sequences are really pseudo-random: only their functional meaning in a specific context defines them as “different”.

We can call functional CSI as FSCI (functionally specified complx information). That is obviously a subset of CSI proper, but practically the only subset which is really relevant for biological issues.

327. 327
Sal Gal says:

tribune7,

So why do we axiomatically rule out design when dealing with proteins, flagellum etc.?

In 321-322, I challenge the notion that there can be any empirical basis for saying that something that is not empirically observable causes the empirical distribution to deviate from the average. In the scenario of an indifferent creator of distributions, the uniform distribution is no more likely than any other. One must actually violate the Principle of Indifference to say that the data should be distributed in one way, and not another.

You have to bring something non-empirical to the table to argue that empirical distributions should “look” one way and not another. Juergen Schmidhuber is well aware of this. He applies a speed prior in inductive learning of programs to describe data and solve problems. This amounts, in his mind, to saying that the Great Programmer does things in a particular way, and he owns up to it as an essentially religious belief.

This brings me back to something I have said in various incarnations at UD — empirical science and logic will never “prove” the existence of the non-material. It should be evident at this point that I know a fair amount of math and science. I am here to say that the only important things I know are unspeakable absurdities. They are a matter of private experience, not empirical observation or proof. I object strenuously to the devaluation of knowledge that is not of a sort that permits public argument for its correctness.

328. 328
Sal Gal says:

gpuccio,

Quick response. I’ll check back in the wee hours.

Functional specification is much easier to treat, and can easily be considered out of the range of necessary laws. It bears no relationship to compressibility.

You’ve strayed onto my turf. Dembski is not the only one who has his uses for Seth Lloyd. Even taking gravitational degrees of freedom into account, Lloyd bounds at about 2^400 bits the information storage capacity of the known universe. Let’s consider a physical box that takes seven 64-bit numbers as inputs and supplies a single 64-bit number as output. There are

(2^64)^(2^448)

possible functions implemented by the box. It follows from basic coding theory that almost all of those functions require about 2^454 bits to encode, no matter how clever you are with the design of the code. That is, for a fixed representation, almost all functions are incompressible in the sense that the description of the function requires about as many bits as an explicit listing of all the input-output pairs. Only a minuscule fraction of the functions “fit” in the known universe. With some technical argument, I can get to the statement that the box can implement only functions with highly compressible descriptions. (The technical mess relates to non-uniqueness of the representation, and that different functions have descriptions of different length in different representations.)

Now there is nothing exceptional about the class of functions I just described. And when you speak of biological function, you are not that far removed from my physical box. Perhaps you meant something I did not understand. But I have just shown that implemented functions have highly compressible specifications (descriptions).

329. 329
JT says:

gpuccio [326]:

Dembski considers all forms of specification. Compressibility is one of them. A sequence like the one you cited is highly compressible. Highly compressible sequences are a tiny subset of the whole search space, and are easily recognizable by intelligent beings.

As far as I can see (and do correct me if I’m wrong by giving specific quotes) the only type of specification that Dembski mentions in that entire paper (the one he mentioned above and which I’ve made repeated reference to) is compressibility.

They are specified and, if complex, they cannot come out of purely random causes.

Its not a matter of them being complex. If they’re of a certain uncompressed length and described by a simple pattern, then to Dembski they’re created by something other than chance and Dembski’s term for “Not chance” is “Intelligence”. I understand that you have your own interpretation of what he says – I don’t think its what he says.

In that case we have to consider the whole complexity of the sequence, and not the “compressed” complexity. In other words, you will never obtain that sequence of 01 by flipping a fair coin.

If the sequence can be produced by a simplisitic process then its only the probability of that simplistic process that is relevant, IMO – not the uncompressed length of the output.

But the problem with compressible sequences is that they can often be the product of necessity. that would be the case if the sequence of 01 were the output of a computer program. So, to define such a sequence as specified one must be sure that no necessary cause is present.

That I believe is where the argument from ignorance comes in in ID. Dembski would grant that “Sure a mechanism could cause this – but if you think there is one, you have to point it out, otherwise we’re going to assume its Intelligence”, where for him, Intelligence is something other than mechanism (or at least historically it has been for him).

I think that Dembski correctly analyzes all forms of specification because that’s important for his general theory. Moreover, compressible sequences are specially interesting for the fact that they are easily recognized by consciousness.

They’re also recognizable by mechanisms, if by “recognize” you mean a change in behavior closely correlated with encountering some specific compressible sequence. If you mean “have a subjective experience of the sequence”, well who knows what that means. But I hope you didn’t mean that humans have some general ability to distinguish compressible sequences from non-compressible ones, because no such ability exists anywhere.

But, that said, we must remember that all that has no real importance for the question of biological information. In fact, the specification in biological information is of the functional type, and is not linked to compressibility. Functional specification is much easier to treat, and can easily be considered out of the range of necessary laws. It bears no relationship to compressibility. So, functional sequences are really pseudo-random: only their functional meaning in a specific context defines them as “different”.

To the best of my knowledge, you are absolutely wrong on this. The only type of specification that Dembski considers is compressibility. IOW, that is what he is exploiting for the purpose of his specific arguments. The only thing relevant about “biological information” from the standpoint of Dembski’s arguments is that such information is compressible.
I stand ready be corrected if you can provide specific quotes from that paper (which after all he identified in this thread as the most up to date statement regarding what CSI is all about.)

Note: I’ll carry this discussion on for as long as you want I guess. I had statements I wanted to make previously – I sincerely hope they were comprehended. I am certainly willing to clarify them. I should be willing to defend them as well, but its not like I’m itching for a debate.

(Note: I have idea how the formatting of this will turn out because I’m not getting the preview function currently.)

Regards

330. 330
Prof_P.Olofsson says:

StephenB[320],
No hurt feelings here. Now, you write

The original point of the EF exercise was, after all, was to show that that, regardless of the method he used, Caputo was almost certainly cheating

which is still not true. The original point was to ask how the EF is applied. As it requires all chance hypotheses to be rejected, (1) how do we reject those with p>1/2 and (2) why do we need to?

I have now repeated the two questions many times. Any answers?

331. 331
Prof_P.Olofsson says:

tribune7[323],
You say that they are, Dembski says that they are not. I can only conclude that you disagree.

332. 332
Prof_P.Olofsson says:

Sal Gal[321],
Let me repeat Demsbki’s answer: we rule out all hypotheses with p>1/2 by taking Caputo’s word (my emphasis because I think it’s pretty weak). My answer is that we can’t rule out all these hypotheses but it doesn’t matter, because he could also have cheated by using nonuniform chance. There is no need to insist on ruling out all chance hypotheses and infer “design” but that’s what the EF insists upon.

333. 333
JT says:

clarification:

I said:

“If the sequence can be produced by a simplisitic process then its only the probability of that simplistic process that is relevant, IMO – not the uncompressed length of the output.”

I meant that that is what is relevant in reality, not the design inference. But I guess you’ve given me some information I solicited previously if you’re confirming that the size (in bits) of a process capable of generating a string is not relevant in I.D, only the uncompressed length of the generated string itself.

334. 334
JT says:

gpuccio:

another clarification [329]

I wrote:

That I believe is where the argument from ignorance comes in in ID. Dembski would grant that “Sure a mechanism could cause this – but if you think there is one, you have to point it out, otherwise we’re going to assume its Intelligence”, where for him, Intelligence is something other than mechanism (or at least historically it has been for him).

Sometimes I think I actually agree with Dembski, and he is just incapable of expressing his ideas in a succinct way.

I understand, as I alluded to in 316, that if you trace back the mechanistic causes for something, that you will eventually hit something that wasn’t caused by a mechanism. At that point I do agree that all you have left as an explanation is blind chance or something else. If ID want’s to call this “something else” “Intelligent Design” that’s fine. But that doesn’t mean humans themselves operate on the basis of some nebulous unspecifiable thing. And let’s be clear – for ID, Intelligence is unspecifiable as it cannot be characterized as a mechanism for them. And just to repeat what I said in 316, whatever that first cause is it could have just existed forever, so you don’t even have to invoke “Intelligent Design” as a cause for it.

OK that’s it hopefully.

335. 335
JT says:

I have to clarify the following as well or someone will misunderstand [329]:

“If the sequence can be produced by a simplistic process then its only the probability of that simplistic process that is relevant, IMO – not the uncompressed length of the output.”

By simplistic process I don’t mean flipping a coin. I mean any simplistic process that could generate the string 100% of the time. “Necessity” in ID parlance.

336. 336
PaV says:

PO[302]:

We catch most cases of schizophrenia but also get a lot of false positives. If we use a very low significance level, such as the UPB, we would not classify anybody with schizophrenia so we’re missing the 2% that are schizophrenic and get no false positives.

If you are willing to accept a test that is 97% reliable, then you will get false positives 3% of the time. If you use the UPB, you’re saying you won’t accept false positives. When the conditional hypothesis assumes a ‘normal’ population, then why would you want to accept false positives.

Two questions for you:
(1) According to Shandin, the true probability of getting a false positive is around 60%. Does that mean if the test were conducted twice on the same individual that the likelihood of that person getting a false positive 36%.

(2) For the posterior distribution you graphed, you’ve said that the cumulative porbability for p<=1/2 of 10^-11 is a level you consider too low to be likely. IOW, you reject it. Now the cumulative probability for p=.1, or p=.2 or .3 or .4 would all be lower. Would you reject those probabilities as well?

337. 337
JT says:

gpuccio [329,326]

I wrote:
“I hope you didn’t mean that humans have some general ability to distinguish compressible sequences from non-compressible ones, because no such ability exists anywhere.”

Let it not be said I can’t comprehend someone else’s point of view.

Upon reflection, I do understand that humans see patterns (compressible sequences) in things. We look for patterns. If we don’t discern a pattern in something, then for us it is noise (i.e. uncompressible) whether it is in reality or not. Furthermore, it seems to me that the patterns we perceive are often internal to us, that is we see patterns dictated by the patterns inherent in our internal coginitive and sensory makeup, which serve to filter out all but what they are capable of detecting.
This is all mechanistic, not some unexplained attribute of a nonmechanism, i.e. “Intelligent Agency”, IMO.

338. 338
Prof_P.Olofsson says:

PaV[336],you won’t accept false positives.

When the conditional hypothesis assumes a ‘normal’ population, then why would you want to accept false positives.

339. 339
tribune7 says:

I challenge the notion that there can be any empirical basis for saying that something that is not empirically observable causes the empirical distribution to deviate from the average. In the scenario of an indifferent creator of distributions, the uniform distribution is no more likely than any other. One must actually violate the Principle of Indifference to say that the data should be distributed in one way, and not another.

But there is a reality and this includes the existence of design. Attempts to quantify it are certainly not beyond the pale.

And I suspect Dembski’s methods have grounds for improvement — it is a fairly new thing, after all — but it is a pretty solid at bat.

You have to bring something non-empirical to the table to argue that empirical distributions should “look” one way and not another.

Sal Gal, that almost sounds like something Godel would say 🙂

But then again he went ahead and proved God anyway.

I am here to say that the only important things I know are unspeakable absurdities. They are a matter of private experience, not empirical observation or proof.

I agree with this. And unlike Godel, ID will not prove God nor is it designed to do so.

It will, however, illustrate the dangerous and cruel absurdity that society should be predicated on the belief that everything happened by accident and that there is no absolute purpose for our existence.

340. 340
Prof_P.Olofsson says:

PaV[336],
Pleae ignore [338], I pressed return by mistake. You write

When the conditional hypothesis assumes a ‘normal’ population, then why would you want to accept false positives.

Well, you wouldn’t, but everybody isn’t normal. You have a population that is composed of normals and schizophrenics and you have two conditional probabilities, one for normals and one for schizophrenics. Among the normals, 3% test positive and among the schizophrenics, 97% test positive. As there are so few schizophrenics, most positive cases are actually miclassified normals. It’s a common problem in any type of screening.

(1) According to Shandin, the true probability of getting a false positive is around 60%. Does that mean if the test were conducted twice on the same individual that the likelihood of that person getting a false positive 36%.

Good question. No, the 36% (60% of 60%) would assume independence. Here, an individual tested twice would likely get the same test result twice (although that of course depends on the test and I don’t know what it is) so the false-positive rate would still be 60%.

(2) Yes.

341. 341
RoyK says:

PO, you see where PaV is going here. By getting you to admit that you would “reject” if p=.1, or .2 or .3 or .4, he will say that these constitute an implicit “rejection region.” Can you please respond to this “implicit” logical trap?

342. 342
Prof_P.Olofsson says:

By the way, what’s often done in screenings for a disease is to first use a test with high sensitivity (will catch most disease cases but also many normal cases). Then the positive cases are restested with another test that has higher specificity (will weed out most false positives). The reason for not using the second test from the beginning is that it’s more expensive or time-consuming. I have personal experience in testing for TB for my green card application. First I tested positive with the skin test because I’d been vaccinated which gives a lot of false positives. Then I was cleared with a chest X ray. More specific because it actually looks at the lungs and is not affected by previous vaccinations.

343. 343
Prof_P.Olofsson says:

RoyK[341],
I know, but I have already explained to him in an email the difference between a rejection region and a “rejection region.” He said I shouldn’t expect a reply but I still hope he read it. We can also look back at Mark Frank’s comment [237] or my comments [276] and [280]. I said I wouldn’t continue this debate but I reneged because PaV asked questions that I can answer. We’ll see how much stamina I have!

344. 344
IDskeptic says:

I’d love a free online statistics correspondence course…

345. 345
Prof_P.Olofsson says:

gpuccio, if you’re still around:

I intended to reply to all your criticism but we haven’t even made it to Behe yet. It usually happens and the last time I debated at UD, we set a length record (the “Archie Bunker” thread). I have comments about Behe as well but I think we’ll take them “off the air” if you’re still interested.

346. 346
Prof_P.Olofsson says:

IDskeptic[344],
It’s free, it’s online, and it might end abruptly at any time! 🙂

347. 347
Sal Gal says:

This discussion continues at a new thread Barry A created.

Thanks, Barry.