Does Bayesian Fuzziness Add to the Analysis?

_{Barry Arrington

July 15, 2013

Intelligent Design}

Share: Facebook; Twitter; LinkedIn; Flipboard; Print; Email

In comment 30 to this post Elizabeth Liddle writes:

I can think of lots of ways of testing specific design hypotheses, but they all involve a hypothesis involving a postulated designer. And IDists insist that this is irrelevant – that “Design detection” should only involve the observed pattern, not any hypothesis about the designer. This is ludicrous, frankly.

Let’s explore one of Lizzie’s prior forays into design detection, and we’ll leave it up to the onlookers to decide which side is “ludicrous.”

In a prior post I posed the following question to Dr. Liddle:

If you were to receive a radio signal from outer space that specified the prime numbers between 1 and 100 would you conclude (provisionally pending the discovery a better theory, of course) that the best theory to account for the data is ‘the signal was designed and sent by an intelligent agent?’

Dr. Liddle responded:

Yes. And I’ve explained why.

She expanded on her explanation:

Barry, I did NOT make the inference ‘based upon nothing but the existence of CSI’!

My inference had nothing to do with CSI.

It was a Bayesian inference based on two priors:

My priors concerning the probability that other parts of the universe host intelligent life forms capable of sending radio signals (high)

My priors concerning the probability that a non-intelligent process might generate such a signal (low).

Dr. Liddle’s problem can be summarized as follows:

1. Denying the design inference based on the prime number sequence is not an option. The inference is so glaringly obvious that to deny it would be absurd. Even arch-atheist Carl Sagan admitted this signal was obviously designed (when he used it as the basis of his book “Contact”). Therefore, were Dr. Liddle to deny the obvious design inference she would instantly lose all credibility.

2. So she asks herself: “How can I admit the design inference while continuing to deny the methods of ID proponents?”

3. Her solution: “I know. I’ll admit the design inference but cover up my admission with Bayesian fuzziness, and that will obscure the fact that I used the methods of the ID proponents while I continue to denounce those very methods.”

Notice how Dr. Liddle’s Bayesian “priors” add absolutely nothing to the design detection methods advocated by ID proponents. Here is a graph of the explanatory filter:

Let’s run the prime number sequence through the explanatory filter to see how.

1. We observe an event (i.e. a radio signal specifying the prime numbers between 1 and 100).

2. Is it highly contingent? Yes. We can exclude mechanical necessity.

3. Is it highly complex and specified? Yes. We can exclude chance.

4. The best explanation for the data: Design.

Now let’s see if Dr. Liddle’s Bayesian analysis adds anything to what we already have.

Prior 1: Estimate of the probability that other parts of the universe host intelligent life forms capable of sending radio signals: High

It is obvious that one’s prior estimate of the probability of the existence of intelligent life forms in other parts of the universe is utterly irrelevant to the design inference. How do I know? By supposing the exact opposite of course. Let’s assume that a person believes there is practically zero chance that other parts of the universe have intelligent life (as we have seen on this site, there is very good reason to believe this). If that person were to receive this signal he would have to revise his conclusion, because the signal is obviously designed.

We see, therefore, that whether one’s Bayesian prior regarding the probability of the existence of intelligence life forms is 0% or 100% makes absolutely no difference to the design inference. From this we conclude that Dr. Liddle’s first prior adds nothing to the analysis.

Prior 2: Estimate of the probability that a non-intelligent process might generate such a signal: Low

This, of course, is the explanatory filter by another name. How do we know that the probability that a non-intelligent process might generate such a signal is low? Because it is highly continent, complex and specified.

It is important to see two things:

1. When Dr. Liddle correctly inferred design from the prime number sequence she had one and only one data point: A radio signal specifying the primes between 1 and 100.

2. Dr. Liddle knew nothing about the provenance of the radio signal. In other words she made a design inference based on nothing but the pattern itself while knowing absolutely nothing about the designer. When she made her design inference she did not first make a hypothesis based on the “postulated designer,” for the simple reason that there was not a scintilla of data upon which to base that hypothesis other than the pattern itself.

Conclusion: Though she tried to cover it up with Bayesian fuzziness, Dr. Liddle did the very thing she now says is “ludicrous.”

Comments

Liddle and Arrington agree on this:
Your inference is only as good as your priors. If the data supporting your priors is very weak, then you probably shouldn’t stake too much on the table when you make your decision. However, if it’s very strong, then you might.
Liddle and Arrington agree that her priors here are:
Not quite thin air, Barry, as I explain above. But pretty thin air, I agree, in this case,
And that leaves the trap that Mark Frank fell into and I pointed out in 19. Let me explain it step-by-step: 1. Liddle can have no real confidence in her Bayesian priors. They are, as she concedes, pulled from pretty thin air. 2. Liddle also concedes the common sense conclusion that a Bayesian inference is only as good as the priors on which it is based. 3. Therefore, the Bayesian inference in this case is very weak. 4. This highlights the futility of applying a Bayesian approach in this case. The design inference based on a Bayesian analysis is very weak. Yet we know with all but certainty that the design inference is correct. 5. Therefore, an analysis that gives weak (non-existent really) support to a certain conclusion is useless.Barry Arrington_{July 15, 2013
July
07
Jul
15
15
2013
11:48 AM
11
11
48
AM
PDT}

While you catch up on my posts, Barry, because I think we have come slightly adrift, let me present a nice puzzle to you all reprobate gamblers (Sal, Barry, whoever else :)) You are a prisoner, about to be shot. But you are offered a possible out. You are presented with a large bag, apparently filled with coloured balls. You are told that the balls in the bag are from a factory that makes only red and blue balls. You are invited to withdraw one ball from the bag. It is red. You are invited to withdraw a second, but first you must guess its colour. If you are correct, you go free. If you are incorrect, you are shot. Which do you choose? And why?Elizabeth B Liddle_{July 15, 2013
July
07
Jul
15
15
2013
11:39 AM
11
11
39
AM
PDT}

Liddle: “So simply by using the fact that I have some reason to think an intelligent signaler is possible and no reason to think a non-intelligent signaler is possible, Design will come out ahead.” You write that you have no reason to think a non-intelligent signaler is possible, but surely you understand this is wrong. Non-intelligent sources send out radio signals all the time (quasars, for example). It is not the fact that a signal has been sent that is important. Both intelligent and non-intelligent sources send signals. The issue is whether there is anything about this particular signal that would lead to a design inference. And obviously there is. Tell me Elizabeth, what is there about this particular signal that separates it from a signal generated by, say, a quasar. Hint: The answer has the following three words that you can arrange in the correct order: information, specified, complex.Barry Arrington_{July 15, 2013
July
07
Jul
15
15
2013
11:36 AM
11
11
36
AM
PDT}

Barry:
Dr. Liddle, for the life of me it appears as though you are saying that the phrase “a hypothesis about the designer” is not equivalent to the phrase “a hypothesis involving a postulated designer.” If that is the case, please explain what you meant by the latter and how it differs from the former. Keep in mind that my overall point is that one can make a design inference while knowing absolutely nothing about the designer other than that he/she/it is intelligent. If you concede that statement, then we don’t disagree.
Let me try to say what I mean, in the hope that it may make sense of what I have previously said :) Dembski's EF, and CSI concepts are based on Fisherian null hypothesis testing. I think it is misapplied. I think if you want to use null hypothesis testing, or indeed any frequentist form of hypothesis testing, you need to have a specific hypothesis about the designer, or design method, that will enable you to properly calculate an appropriate null. However. I do think that it is perfectly reasonable to make an informed (even minimally informed) inference about which of two (or more even) general hypotheses regarding a phenomenon (Was it designed? Was it not designed?) are more likely, given the data, using a Bayesian approach, which, as I say, is extremely similar to "inference to best explanation". But, as you (and I, simultaneously as it happens!) state: GIGO. Your inference is only as good as your priors. If the data supporting your priors is very weak, then you probably shouldn't stake too much on the table when you make your decision. However, if it's very strong, then you might. But you are not testing your hypotheses, in the Popperian sense. You are simply trying to make the best decision between two alternatives given the best information you have. And the elegance of Bayes, as you will appreciate, is that it allows you to pull out an often counter-intuitive posterior given quite intuitive priors. So we can probably agree (let me slightly rephrase this from your formulation) that it is perfectly reasonable to make an inference about which of two hypotheses is the more likely to account for your data, without fleshing out the hypotheses much at all (although I would argue that you do need to make a stab at something about your priors). But it is not a test of your hypothesis, in the Popperian sense. Thus I see no reason to think the Design hypothesis for life is unreasonable, although it's not what comes out of the equation given my own priors. What I do think is faulty is the EF/CSI method, which gives you a positive for Design with a p value to die for, but which, I suggest, is invalid.Elizabeth B Liddle_{July 15, 2013
July
07
Jul
15
15
2013
11:35 AM
11
11
35
AM
PDT}

the fact that you know nothing about the probability of the existence of intelligent life
Other than the other fact that apparently intelligent life lives on Earth.LarTanner_{July 15, 2013
July
07
Jul
15
15
2013
11:31 AM
11
11
31
AM
PDT}

Liddle “But I haven’t based it [i.e., the design inference”] on ‘nothing but the existence of a pattern’, Barry!” Of course you have Elizabeth, your denials to the contrary notwithstanding. As I have shown, your Bayesian priors are meaningless. Therefore, your Bayesian analysis is meaningless. Yet you and I both know that the design inference is indisputable. [Setting aside ludicrous appeals to crystal lattice structures that leave voids in a prime number pattern]. And what is that indisputable design inference based on if it is not based on a wholly unsupportable Bayesian analysis? Why, it is based on nothing more than our knowledge that nature is incapable of generating the Signal while intelligent agents produce Signals of this sort routinely. In other words, the inference is based on nothing but the pattern itself and that fact that it contains complex specified information that nature cannot produce. Barry Arrington_{July 15, 2013
July
07
Jul
15
15
2013
11:25 AM
11
11
25
AM
PDT}

And as I explained in comment 11, your priors are literally meaningless. You’ve pulled them out of thin air. Barry
Keep in mind that I am not criticizing Bayesian analysis as such. Indeed, I use it all the time at the poker table. I am saying that Bayesian analysis in this instance does nothing but cover up the fact that you know nothing about the probability of the existence of intelligent life prior to getting the Signal, and that is a huge coverup. In other words, when the Bayesian priors are garbage the Bayesian posteriors will also be garbage, a classic example of the GIGO phenomenon at work.
On the contrary, as I explain above, it does the reverse of "cover up" the paucity of the data that supports the priors. It makes it absolutely explicit. But as my post there has only just loaded (for some reason it got stuck in cyberspace for a short while) I'll say no more until you've had a chance to read it.Elizabeth B Liddle_{July 15, 2013
July
07
Jul
15
15
2013
11:18 AM
11
11
18
AM
PDT}

Barry:
Here’s the problem with Liddle’s analysis. Prior to receiving the signal we have zero data upon which to reasonably assign any probability whatsoever to the existence of intelligent life other than that on earth. The most famous attempt, of course, is based on Drake’s equation. But as Crichton has pointed out: The problem, of course, is that none of the terms [in the equation] can be known, and most cannot even be estimated. The only way to work the equation is to fill in with guesses . . . As a result, the Drake equation can have any value from ‘billions and billions’ to zero. An expression that can mean anything, means nothing. Speaking precisely, the Drake equation is literally meaningless . . . Liddle’s estimate of the probability of intelligent life in the universe is meaningless, and her estimate of the probability that intelligent life in the universe can send a radio signal is also meaningless.
I think this is a pretty fair point. I think trying to weigh up probabilities in the absence of any data on which to construct a frequency distribution is pretty pointless. What emerges from a Bayesian posterior is only as good as the priors you plug in. If they are well-founded, you can have confidence in your posterior. If they aren't, well, it's GIGO. Nonetheless, for the example you gave me, my inference would be "design" because on the little data I have, the priors for intelligent signal-sending life are higher (after all, I know of at least one intelligent signal-sending life form) than for a prime-number-generating non-intelligent process (frequency distribution of data = []). So simply by using the fact that I have some reason to think an intelligent signaller is possible and no reason to think a non-intelligent signaller is possible, Design will come out ahead. So using the very little data I have, I would conclude Design, using that reasoning. But then you asked me to consider it again, for an non-living designer. Well, I simply have no basis in which to make such a prior estimate at all, but as an exercise, I showed you how it could be done, possibly for someone such as yourself who does consider that a non-living designer is a substantial non-zero possibility, presumably based on your personal experience.
Liddle and Frank say that the advantage of Bayesian thinking is that it makes you estimate the probability in a structured way. Uh, explain to me again how pulling probabilities out of nothing but thin air* and putting them in an equation is an advantage.
Not quite thin air, Barry, as I explain above. But pretty thin air, I agree, in this case, not surprisingly as it is a completely fictional scenario.
Of course it is not an advantage. Again, it is obvious what Liddle is doing (and Frank is applauding). Liddle has pulled meaningless probabilities out of thin air and plugged them into a Bayesian equation. The result looks rigorous until you realize that when you boil it all down it is all but meaningless and if by chance it does have some meaning, that meaning is “my gut tells me there is intelligence out there and they sent the signal.” Yeah, that’s rigorous alright.
I didn't claim it was rigorous. It's as flakey as heck. But if a gun was held to my head, and I had to choose, at least it would give me a basis on which to pick, using the tiny amount of data at my disposal. A Bayesian inference is only as good as the support for your priors.
*I had to resist with all my might the temptation to write “your a– –” instead of “nothing but thin air,” but that would have been vulgar.
No problem, Barry. I have heard (and used) worse :) But let me try to clarify the point that both Mark and I are making: To compute CSI, you need to pull an number out of your ahem. Specifically you need to plug in P(T|H) where H is "the relevant chance hypothesis taking into account Darwinian and other mechanisms". There is no way to compute this. You just have to pull it out of your donkey based on your priors about the probability of "Darwinian and other mechanisms". It is, indeed, the exact equivalent of P(S|¬I) in my equation above. But unlike my equation, it does NOT require you to make any estimate of P(I). It is therefore precisely as flakey in one parameter, and simply omits the other. And then it adds another parameter, which as far as I can see is based on a mistaken assumption regarding the upper bound on the size of the universe. But the worst thing about it is that that P(S|¬I), though present, is never, ever, in my experience of seeing CSI arguments, calculated for anything other than "random independent draw". In other words, the first step of the EF, never actually makes it into the second step. So while my figure might have been a wild stab, at least it was based on some kind of data, and balanced by an estimate of P(I) and P(S|I) given more, CSI simply buries all this donkey extraction stuff, and then triumphantly claims Design, p<10^-150! My more modest p=.91 for the probability of Design, given the data, seems rather more seemly. My Little Pony extraction, perhaps :)Elizabeth B Liddle_{July 15, 2013
July
07
Jul
15
15
2013
11:16 AM
11
11
16
AM
PDT}

RDFish: The “design hypothesis” is not a meaningful explanation of anything at all. It is like trying to explain planetary motion with the “movement hypothesis” without saying what is doing the moving – it tells us precisely nothing.
Wrong. Because designers are intelligent, that is, intelligence manifests various properties. What would you call something that has the following properties? ... 1. "Understanding" of natural forces. 2. "Foresight" of future events based on the arrangements and relationships of various natural forces. 3. "Intent" to alter the course of future events. I call it "intelligence." Apparently SETI agrees, because their entire objective is to find something "out there" that exhibits the marks of something that only an intelligent source could generate. The term "intelligence" is not vacuous, and neither is a design hypothesis. Let's try to explain the first 100 primes from a distant radio source. It would either be due to chance and necessity of natural forces, or Something With the Properties Cited Above, i.e., "intelligence." Both are plausible explanations given what we know about entities that have those properties, namely, humans. A design hypothesis in this case would be one that says, "an intelligent entity or entities caused the transmission of these primes. It is not plausible that cause and necessity of natural forces did it." That's essentially the hypothesis SETI is operating under. Sidebar: does this mean the intelligence has to be from a "human?" I don't think so. What cogent case can you make that any such intelligence would necessarily be human?CentralScrutinizer_{July 15, 2013
July
07
Jul
15
15
2013
11:12 AM
11
11
12
AM
PDT}

Liddle:
I’m saying that in order to make a design inference using Baysian reasoning, we need to consider our priors for the probability of a number of things, including the probability of a Designer. Ideally that probability should be based on some kind of data, if you want to have confidence in your posterior.
And as I explained in comment 11, your priors are literally meaningless. You’ve pulled them out of thin air. Keep in mind that I am not criticizing Bayesian analysis as such. Indeed, I use it all the time at the poker table. I am saying that Bayesian analysis in this instance does nothing but cover up the fact that you know nothing about the probability of the existence of intelligent life prior to getting the Signal, and that is a huge coverup. In other words, when the Bayesian priors are garbage the Bayesian posteriors will also be garbage, a classic example of the GIGO phenomenon at work.Barry Arrington_{July 15, 2013
July
07
Jul
15
15
2013
11:11 AM
11
11
11
AM
PDT}

The “design hypothesis” is not a meaningful explanation of anything at all. It is like trying to explain planetary motion with the “movement hypothesis” without saying what is doing the moving – it tells us precisely nothing. Cheers, RDFish
A slight but respectful disagreement. Planets move according to an elipse. It would be wrong to say the mechanism is an elipse, the mechanism is gravity that creates the elliptical motion. Designs look to be designed. It is formally wrong to say design is the mechanism of design. It is correct however to say it looks analogous to something a human could design if the human hand the resources and intelligence. Whether intelligence is the root cause, there is nothing wrong to say it looks like the product of human-like intelligence or something greater. I don't argue vigorously whether it is a scientific claim, but the impression of design in certain cases is clearly there. In the case of pulsars, the impression of design was false. No great loss, no disaster. If ID is wrong, no great disaster to the human enterprise either, imho.scordova_{July 15, 2013
July
07
Jul
15
15
2013
11:05 AM
11
11
05
AM
PDT}

Mark Frank:
Barry – if the prior probability is really unknown then it forces you to confront that fact and therefore the incredible uncertainty in your conclusion. That uncertainty is real and it is better to recognise it. An alternative that brushes over the uncertainty and thus allows you to announce that something is certain is not superior.
Mark, it seems like you are saying that if we cannot reliably estimate the probability of the existence of other intelligent life prior to receiving the Signal then there is “incredible uncertainty” in the conclusion that the Signal is the result of intelligent design. If that is what you are saying, then your statement is preposterous and patently absurd. It is the exact opposite of the truth. Far from being incredibly uncertain, the design inference based on the signal is all but certain. As I said, even a man who was perhaps the most famous materialist atheist of the latter half of the twentieth century (Carl Sagan) would agree that the design inference is all but certain. Can it be that you really believe the gibberish that you wrote? I have a hard time accepting that.Barry Arrington_{July 15, 2013
July
07
Jul
15
15
2013
11:05 AM
11
11
05
AM
PDT}

Liddle quotes Arrington: “A corollary to the point of the OP is that your statement that one needs to make a hypothesis about the designer before one can make a design inference is false.” Liddle then writes:
I am not aware of having said such a thing, although I’d like to see the quote and check the context. Certainly if you wanted to use null hypothesis testing, you would need a hypothesis about the designer, and I’ve certainly said that. I’ve also said, as you quote in the OP, that “I can think of lots of ways of testing specific design hypotheses, but they all involve a hypothesis involving a postulated designer.”
Dr. Liddle, for the life of me it appears as though you are saying that the phrase “a hypothesis about the designer” is not equivalent to the phrase “a hypothesis involving a postulated designer." If that is the case, please explain what you meant by the latter and how it differs from the former. Keep in mind that my overall point is that one can make a design inference while knowing absolutely nothing about the designer other than that he/she/it is intelligent. If you concede that statement, then we don’t disagree.Barry Arrington_{July 15, 2013
July
07
Jul
15
15
2013
10:56 AM
10
10
56
AM
PDT}

To say intelligence can’t be described in mechanical terms in principle is an unsupported assumption, however.
Agreed, thank you for the correction. Salscordova_{July 15, 2013
July
07
Jul
15
15
2013
10:52 AM
10
10
52
AM
PDT}

RDFish: Moreover, if we cannot explain what “intelligence” is in mechanical (or non-mechanical!) terms, then it is not helpful to offer it as an explanation for anything. It is not “intelligence” that designs cars and computers and Mt. Rushmore, it is “human beings”. It is not “intelligence” that designs eyeballs and blood clotting cascades and flagella, it is… nobody has any idea.
So if we were to continously detect the first 100 primes from 2 to 541 from a distant source, you would not at least provisionally put the tag of "intelligently generated" on the phenomenon? How would you tag it?CentralScrutinizer_{July 15, 2013
July
07
Jul
15
15
2013
10:50 AM
10
10
50
AM
PDT}

But I haven't based it on "nothing but the existence of a pattern", Barry! I'm really mystified as to why you should be telling me I'm saying things that I plainly have not said! I'm saying that in order to make a design inference using Baysian reasoning, we need to consider our priors for the probability of a number of things, including the probability of a Designer. Ideally that probability should be based on some kind of data, if you want to have confidence in your posterior. You also need a prior about the probability of the signal without a designer. And you should probably also have a probability for the probability that if there is a Designer, he/she/it would send a radio signal, but we can set that as 1 if you like. That is a long way from "nothing but the existence of a pattern"! But it DOES include the probability of an immaterial designer if that is what you are interested in. So reports of my concessions are greatly exaggerated! As for your comment:
A corollary to the point of the OP is that your statement that one needs to make a hypothesis about the designer before one can make a design inference is false.
I am not aware of having said such a thing, although I'd like to see the quote and check the context. Certainly if you wanted to use null hypothesis testing, you would need a hypothesis about the designer, and I've certainly said that. I've also said, as you quote in the OP, that "I can think of lots of ways of testing specific design hypotheses, but they all involve a hypothesis involving a postulated designer." However, null hypothesis testing is not the only way of making an inference, and I have suggested the Bayesian reasoning given above. As the priors for an immaterial designer are poorly supported (though not absent), I don't think one could have a great deal of confidence in the inference, but an inference is nonetheless perfectly possible, and indeed reasonable. But a Bayesian inference is not a test of a hypothesis - it is a way, as Sal says, of ranking hypotheses - specifically, of ranking the probability of a hypothesis being true, given the data, but the ranking is extremely dependent on the input priors. On the other hand, if we did have a specific design hypothesis, for example the "front-loading" hypothesis, that would be very testable, or could be. It could make specific predictions not predicted by an evolutionary model and so could go head-to-head with evolution. As I keep saying, Barry, I have no problem in principle with attempting to detect design, and I have even no problem in principle, in trying to figure out how to detect a Designer who transcends physical laws. The point I was trying to make in the exchange you quote is that I do not think that the EF, or indeed CSI, work. I think they represent a completely invalid application of Fisherian null hypothesis testing, and I think ID could do a lot better.Elizabeth B Liddle_{July 15, 2013
July
07
Jul
15
15
2013
10:41 AM
10
10
41
AM
PDT}

#11 Barry - if the prior probability is really unknown then it forces you to confront that fact and therefore the incredible uncertainty in your conclusion. That uncertainty is real and it is better to recognise it. An alternative that brushes over the uncertainty and thus allows you to announce that something is certain is not superior.Mark Frank_{July 15, 2013
July
07
Jul
15
15
2013
10:32 AM
10
10
32
AM
PDT}

E Liddle: Barry, I did NOT make the inference ‘based upon nothing but the existence of CSI’! My inference had nothing to do with CSI.It was a Bayesian inference based on two priors: My priors concerning the probability that other parts of the universe host intelligent life forms capable of sending radio signals (high) My priors concerning the probability that a non-intelligent process might generate such a signal (low).
So, Dr Liddle, I'm wondering, when it comes to the particular DNA replication system that we find in nature, and the processes going on within cells, what do you consider the probably of those being the result of natural forces, and what probably of them being the result of a human-like intelligence?CentralScrutinizer_{July 15, 2013
July
07
Jul
15
15
2013
10:28 AM
10
10
28
AM
PDT}

Hi Sal,
What if the mechanism can’t be described in mechanical terms (like intelligence)
The behavior of humans and other animals that we call "intelligent" cannot presently be explained in mechanical terms, because we don't understand how that behavior is generated. To say intelligence can't be described in mechanical terms in principle is an unsupported assumption, however. Moreover, if we cannot explain what "intelligence" is in mechanical (or non-mechanical!) terms, then it is not helpful to offer it as an explanation for anything. It is not "intelligence" that designs cars and computers and Mt. Rushmore, it is "human beings". It is not "intelligence" that designs eyeballs and blood clotting cascades and flagella, it is... nobody has any idea. The "design hypothesis" is not a meaningful explanation of anything at all. It is like trying to explain planetary motion with the "movement hypothesis" without saying what is doing the moving - it tells us precisely nothing. Cheers, RDFishRDFish_{July 15, 2013
July
07
Jul
15
15
2013
10:26 AM
10
10
26
AM
PDT}

Mark Frank: “One of the big advantages of Bayesian thinking is that makes you estimate the probability in a structured way rather than just saying ‘it’s obvious.’” Liddle: “Yes, indeed, Mark.” Here’s the problem with Liddle’s analysis. Prior to receiving the signal we have zero data upon which to reasonably assign any probability whatsoever to the existence of intelligent life other than that on earth. The most famous attempt, of course, is based on Drake’s equation. But as Crichton has pointed out:
The problem, of course, is that none of the terms [in the equation] can be known, and most cannot even be estimated. The only way to work the equation is to fill in with guesses . . . As a result, the Drake equation can have any value from ‘billions and billions’ to zero. An expression that can mean anything, means nothing. Speaking precisely, the Drake equation is literally meaningless . . .
Liddle’s estimate of the probability of intelligent life in the universe is meaningless, and her estimate of the probability that intelligent life in the universe can send a radio signal is also meaningless. Liddle and Frank say that the advantage of Bayesian thinking is that it makes you estimate the probability in a structured way. Uh, explain to me again how pulling probabilities out of nothing but thin air* and putting them in an equation is an advantage. Of course it is not an advantage. Again, it is obvious what Liddle is doing (and Frank is applauding). Liddle has pulled meaningless probabilities out of thin air and plugged them into a Bayesian equation. The result looks rigorous until you realize that when you boil it all down it is all but meaningless and if by chance it does have some meaning, that meaning is “my gut tells me there is intelligence out there and they sent the signal.” Yeah, that’s rigorous alright. *I had to resist with all my might the temptation to write "your a-- --" instead of "nothing but thin air," but that would have been vulgar. :-) Barry Arrington_{July 15, 2013
July
07
Jul
15
15
2013
10:25 AM
10
10
25
AM
PDT}

Liddle: “But I haven’t [confirmed the point of the OP].” Of course you have. The point of the OP is that one can make a design inference based on nothing but the existence of the pattern. A corollary to the point of the OP is that your statement that one needs to make a hypothesis about the designer before one can make a design inference is false. In fact, as you now concede, one need know absolutely nothing about the designer (other than that he/she/it is intelligent, i.e., capable of design) to make that inference. You confirmed that point in your comment 1 and you reconfirmed it in your comment 8. Thank you.Barry Arrington_{July 15, 2013
July
07
Jul
15
15
2013
10:14 AM
10
10
14
AM
PDT}

Sorry, my wording was awful. We can rank possibilities in Bayesian reasoning according to various mechanisms, but what if the real mechanism is missing? What if the mechanism can’t be described in mechanical terms (like intelligence).
Well, we can certainly rank the posteriors, given our priors and likelihoods. I don't see that I've done anything wrong. I don't think I have to describe intelligence in mechanical terms to give it a substantial prior. (This is slightly weird: I am trying to persuade an IDer that it is just fine to put a prior on an unknown mechanism, and that unknown mechanism is a Designer.... Ah well :))Elizabeth B Liddle_{July 15, 2013
July
07
Jul
15
15
2013
09:57 AM
9
09
57
AM
PDT}

Barry:
Where in comment 1 do you make any hypothesis about the designer other than that he/she/it is “intelligent life”? The answer, of course, is that you do not.
Indeed.
Your own comment 1 confirms the point of my post, which is that a design inference can be based on nothing but the observation of the pattern itself without reference to any information about the designer other than that he/she/it is intelligent.
My method would work just as well if my prior for a living designer was infinitessimal, but my prior for a non-living designer was, say, .01. In fact, rather than re-do the calculation, just substitute :
So let me set my prior on the probability of an intelligent ~~life~~ designer somewhere, quite high, say .01. And let me set my prior on the probability an intelligent ~~life being able~~ designer choosing to send radio signals to earth as lower, let’s say .00001
It works just as well. We still do not need to know anything about the designer, and we can still conclude design, given those priors. They are not mine, but for someone, perhaps yourself, with good reasons for having them (direct experience of the Divine, for instance) they could be perfectly reasonable.
I will address other issues in your response in later comments, but I wanted to first thank you for confirming the point of the OP.
But I haven't. Your OP seemed to be saying that my Bayesian approach "added nothing" to the EF. I took the example of a SETI scenario, and plugged in priors about an Extra Terrestrial, because I thought that was what we were talking about. But as I have just shown, the Bayesian approach works just as well for a supernatural designer - as I point out, it's a rather more algebraic equivalent of the Inference to Best Explanation, which quite a few IDers are quite keen on. My point is not that the Bayesian approach adds anything to the EF, but that it is a much better alternative. And certainly does not preclude making a Design inference. It just that with the Bayesian approach, you have to make your priors explicit. With the CSI approach they are hidden (in that P(T|H) I keep banging on about). I think you have assumed (well, you have explicitly said this) that because I am critical of ID, I am nitpicking at the methodology because I don't like the inference. This is simply not true.Elizabeth B Liddle_{July 15, 2013
July
07
Jul
15
15
2013
09:55 AM
9
09
55
AM
PDT}

Can you explain this, Sal? I don’t understand what you are saying.
Sorry, my wording was awful. We can rank possibilities in Bayesian reasoning according to various mechanisms, but what if the real mechanism is missing? What if the mechanism can't be described in mechanical terms (like intelligence).scordova_{July 15, 2013
July
07
Jul
15
15
2013
09:46 AM
9
09
46
AM
PDT}

Bayesian ranking works if you have alternative mechanisms and your list of mechanisms is complete. It will fail if the mechanism in play is not on your list.
Can you explain this, Sal? I don't understand what you are saying.
FWIW, pulsars were once considered possibly of alien origin because of their clock like precision…. Suppose someone did think pulsars were evidence of ID? So what? The damage done would hardly be worth notice in the scheme of things.
Yes indeed they did. It was the Cambridge team with Jocelyn Bell, who, as it happened, had been at my boarding school and I was friends with her younger sister. It was all very exciting - hence my reference to the "LGM" hypothesis - the Little Green Men. We had daily news updates on the school hall blackboard!Elizabeth B Liddle_{July 15, 2013
July
07
Jul
15
15
2013
09:34 AM
9
09
34
AM
PDT}

Bayesian ranking works if you have alternative mechanisms and your list of mechanisms is complete. It will fail if the mechanism in play is not on your list. We may not be able to prove an intelligence exists in the formal sense to create the aliens that gave the SETI signal, but we can use CSI to demonstrate the SETI signal resembles a design. CSI is provisionally a measure of resemblance of design based on available distribution functions. FWIW, pulsars were once considered possibly of alien origin because of their clock like precision.... Suppose someone did think pulsars were evidence of ID? So what? The damage done would hardly be worth notice in the scheme of things. But in the realm of uncertainty, the real question is which idea (ID or Darwinism) is the better wager? Darwinism is a terrible wager. Not even in terms of eternal life, but if we devalue human life, when in fact it is special, what is the cost of Darwinism if Darwinism is wrong? We might never be able to settle with ID is true or not, but like a businessman wagering on the best investment in light of unending uncertainties, what is the best wager on truth. I'm not betting on Darwin, it's a negative expectation bet on many levels. That wager reasonable inference that I've not seen any rebuttal for.scordova_{July 15, 2013
July
07
Jul
15
15
2013
09:28 AM
9
09
28
AM
PDT}

Elizabeth, let’s go back to your original statement:
I can think of lots of ways of testing specific design hypotheses, but they all involve a hypothesis involving a postulated designer.
Where in comment 1 do you make any hypothesis about the designer other than that he/she/it is “intelligent life”? The answer, of course, is that you do not. Your own comment 1 confirms the point of my post, which is that a design inference can be based on nothing but the observation of the pattern itself without reference to any information about the designer other than that he/she/it is intelligent. I will address other issues in your response in later comments, but I wanted to first thank you for confirming the point of the OP.Barry Arrington_{July 15, 2013
July
07
Jul
15
15
2013
09:24 AM
9
09
24
AM
PDT}

Yes, indeed, Mark. And that is of course where we both disagree with Barry. We simply do not accept that we can make a design inference in a vacuum. In fact, I could (and probably should!) have responded to Barry with a much shorter post: We cannot evaluate the probability of an event without considering the generative processes that might have produced it. It's the error at the heart of Dembski's method. And it's hiding there, in plain site, in the old EF: You can only reject "mechanical necessity" if we have exhausted not only known mechanisms but also unknown mechanisms. Which is clearly impossible. So the filter is doomed from the start.Elizabeth B Liddle_{July 15, 2013
July
07
Jul
15
15
2013
09:19 AM
9
09
19
AM
PDT}

Let’s assume that a person believes there is practically zero chance that other parts of the universe have intelligent life (as we have seen on this site, there is very good reason to believe this). If that person were to receive this signal he would have to revise his conclusion, because the signal is obviously designed.
I am sorry but I wouldn't jump to the conclusion it was designed. If I was virtually certain there was no possibility of intelligent life I would examine the possibility of another cause rather seriously. Clearly we have something rather extraordinary happening but then we have already accepted that intelligent life would be extraordinary. One line of thought might be a natural mechanism that generates a stream of increasing numbers but is filtered by another mechanism that eliminates any number that is a multiple of previous numbers in the sequence. Sure I have no idea how such a mechanism could work but then if I don't accept the possibility of intelligent life I have no idea how it could be designed either. One of the big advantages of Bayesian thinking is that makes you estimate the probability in a structured way rather than just saying "it's obvious". The world is full of examples of people jumping to the wrong "obvious" conclusion because they have not adopted Bayesian thinking (including many famous legal cases which you will no doubt be aware of).Mark Frank_{July 15, 2013
July
07
Jul
15
15
2013
09:12 AM
9
09
12
AM
PDT}

First of all, you might like to look at my later response to Timaeus in the same thread, here. Second, I wonder if you could link to my original response to your question. Third, I utterly reject this characterisation of my "problem":
Dr. Liddle’s problem can be summarized as follows: 1. Denying the design inference based on the prime number sequence is not an option. The inference is so glaringly obvious that to deny it would be absurd. Even arch-atheist Carl Sagan admitted this signal was obviously designed (when he used it as the basis of his book “Contact”). Therefore, were Dr. Liddle to deny the obvious design inference she would instantly lose all credibility. 2. So she asks herself: “How can I admit the design inference while continuing to deny the methods of ID proponents?” 3. Her solution: “I know. I’ll admit the design inference but cover up my admission with Bayesian fuzziness, and that will obscure the fact that I used the methods of the ID proponents while I continue to denounce those very methods.”
I am utterly uninterested in "continuing to deny the methods of ID proponents" and thus utterly unmotivated to cover up any putative "admission" whatsoever. This assumption regarding my integrity (or rather lack of it) is extremely offensive. Or would be if I took it seriously, which I do not. So that's all fine. Harrumph. So I will attempt to respond to your OP: You conclude design based on the following reasoning steps:
1. We observe an event (i.e. a radio signal specifying the prime numbers between 1 and 100). 2. Is it highly contingent? Yes. We can exclude mechanical necessity. 3. Is it highly complex and specified? Yes. We can exclude chance. 4. The best explanation for the data: Design.
I use a different reasoning method, and come to the same conclusion. Here is mine, in full (as I don't currently have access to my previous response I will do it again from scratch, and show my working): What I want to know is the probability of Intelligence, given the Signal. I will write that as: P(I|S). using Bayes rule, I get: P(I|S) = P(I)*P(S|I)/P(S) And I can expand the denominator to: = P(I)*P(S|I)+P(¬I)*P(S|¬I) So let me take a stab at some priors: Apparently I said then (and I would say again now):
My priors concerning the probability that other parts of the universe host intelligent life forms capable of sending radio signals (high) My priors concerning the probability that a non-intelligent process might generate such a signal (low).
So let me set my prior on the probability of intelligent life somewhere, quite high, say .01. And let me set my prior on the probability intelligent life being able to send radio signals to earth as lower, let's say .00001 And let me set my prior on the probability of non-intelligent mechanisms sending such a signal even lower, let's say, .00000001 So P(I)=.01 and P(S|I)=.00001 Substituting above, we get, for the numerator: .01 * .00001 Now, for the denominator, we need P(¬I), which is 1-.01 = .99 and P(S|¬I), which I have set low, at .00000001 Which gives us, for the denominator .01 * .00001 + (1-.02)*.00000001 Which, unless Excel is lying to me, gives me a posterior probability of an intelligent source, given the signal, i.e. of P(I|S) of 0.91 Which is clearly high enough to take to the bank. Note that I can tinker with these priors as I want, but as long as I put the probability of intelligent life, and the probability of intelligent life being capable of sending a radio signal as substantially more than the probability of a non-intelligent mechanism emitting a signal, then I can safely conclude design. But at no point have I used CSI. I set my probability of the signal given a non intelligent source at way lower than 500 bits - only 27. So even with a much more lenient prior for non-intelligent origin than Dembski's I can still conclude "intelligent source" with considerable confidence. This is not "fuzziness" - it is a way of extracting an estimate of the confidence that one can place in a conclusion, given one's prior beliefs about what is likely to be the case. And because my prior belief in the likely existence of intelligent designers capable of sending a signal is substantially greater than my prior belief that any non-intelligent mechanism is likely to generate such a signal, I end up with a variant on the "inference to best explanation" method, with a tad more quantitative rigor. And the beauty of the Bayesian approach is that it allows us to revise our estimate as more information comes available. Should we discover some crystal lattice structure that leaves voids in a prime number pattern, and can act as a radio signal filter of some kind (I'm speculating wildly here, deliberately), then we'd have to revise our priors substantially, and might, like Jocelyn Bell, have to abandon the LGM for something less exotic but possibly no less interesting. I entirely agree that my method does not "add" to yours. But nor does it require yours. It does not, as I said, require me to compute the CSI of the signal. Which is just as well, because I do not think that CSI is computable! And nor, in fact, do I think your first step works either - it isn't really separable from the second, which is why Dembski in the end rolled them both up into one, by requiring that for the P(T|H) parameter in the CSI calculation, the H should be "the chance hypothesis that takes into account Darwinian and other material mechanisms". In other words, if you do that,you don't need Step 1, because it has already been done in your CSI computation for Step 2. Except that my case is that it isn't possible to do, for anything other than a very tightly constrained null (for instance, the null that a coin is fair). Look Barry, as you will see from my later response to Timaeus: I honestly don't have a dog in this hunt. Or not much of a dog anyway, maybe a small theological ferret. I'd be truly delighted to see a well-founded approach to design detection in the domain of living things. I even suggested some possible avenues that I would consider potentially fruitful, or at least, not beset by the extremely grave problems I see in CSI and its derivatives. Especially as I rather belatedly discovered today,that Dembski's 500 bit university probility bound, far from being "conservative" (although more conservative than Seth Lloyd's) was actually based on an estimate of the number of particles in the observable universe which is extremely unlikely to be coterminous with the entire universe! So, in effect, we do not have an upper bound on probabilistic resources of the universe anyway. But as I've said, many times, and as this post exemplifies: I would be perfectly happy to conclude "Design" on a much more lenient cutoff criterion than any UPB, given a better founded rationale. Obviously you will differ. But I rest my case (is that the right expression?)Elizabeth B Liddle_{July 15, 2013
July
07
Jul
15
15
2013
09:04 AM
9
09
04
AM
PDT}

Prev 1 2 3 4

You must be logged in to post a comment.

Leave a Reply