Uncommon Descent Serving The Intelligent Design Community

Dawkins Weasel vs. Blind Search — simplified illustration of No Free Lunch theorems

Share
Facebook
Twitter
LinkedIn
Flipboard
Print
Email

I once offered to donate $100 to Darwinist Dave Thomas’ favorite Darwinist organization if he could write an genetic algorithm to solve a password. I wrote a 40-character password on paper and stored it in safe place. To get the $100, his genetic algorithm would have to figure out what the password was. I was even willing to let him have more than a few shots at it. That is, he could write an algorithm which would propose a password, it would connect to my computer, and my computer that had a copy of the password would simply say “pass or fail”. My computer wouldn’t say “you’re getting closer or farther” from the solution it would merely say “pass or fail”. But he wasn’t willing even to go that far. He declined my generous offer. 🙂

Dave Thomas, like Richard Dawkins, advertise the supposed mighty power of genetic algorithms, but when pressed to solve the sort of problems that are relevant to evolution, they are no where to be seen.

Complex functional proteins for a system are notoriously difficult to construct. They are like passwords that must be matched to the right login.

Evolving a functional protein in one context to become functional in another context is not so easy. This is akin to taking a functional password for one account and presuming we could evolve it in steps to become a functional password for another account. Thankfully this doesn’t happen, otherwise thieves could be evolving their bank account passwords to be able to compromise your bank account passwords!

In general, transitionals from one functional protein to another are not selectively favored so as to coax evolution toward a new functional target. If each attempt to evolve a new protein is met with “pass or fail” versus “you’re getting closer or farther”, the search is effectively blind as a random search. The evolutionary search for new functional proteins fails for the same reasons thieves cannot evolve their functional passwords into your functional passwords.

The fact that Dave Thomas declined my offer indicates deep down he understands the fallacious claims of Darwinism and that Dawkins Weasel is a misleading picture of how natural selection in the wild really works when trying to solve problems like protein evolution. He knew he couldn’t take his passwords and evolve them into mine.

Despite this, we hear evolutionists proudly proclaim “evolution doesn’t evolve proteins from scratch, it evolves them from existing ones”. This claim may look promising on the surface, but let me pose this rhetorical question to the readers. You have a functioning password that works for your account, it may even share extreme similarities to other passwords that people have for their accounts. Does that fact give you a better chance of solving their passwords over blind luck? No. But evolutionary biologist are effectively saying that when then say “evolution evolves one protein from another.” So if Darwinian evolution will not evolve proteins what will? Surprise, there is a New mechanism of evolution — POOF….

But these considerations do not hinder Dawkins from advertising Weasel as the way evolution works. In contrast, as reported at UD, real evolution destroys complexity over time. The average of all reported real-time or near-real time lab and field observations is that most adaptive evolution is loss of function, not acquisition of function — Behe’s rule. In fact real evolution is worse than blind search, it can’t even hold on to the complexity that already exists, much less create it. The Blindwatchmaker would dispose of lunches even if they were free.

No Free Lunch theorems are the formalization that shows that Darwinian search is no better than blind search for cases like solving passwords. No Free Lunch would assert that if Dave Thomas’ genetic algorithm solved my password, he likely was privy to specialized information which a random search algorithm didn’t have. By way of analogy, in the case of Dawkins Weasel, if we viewed the phrase: “METHINKS IT IS LIKE A WEASEL” as the target password, then Dawkins pretty much front loaded the desired password to begin with. But Dawkins and Thomas will have no such luck if they don’t have the desired password up front.

But I didn’t give Dave the target answer, hence there was no free lunch ($100 worth) for Dave Thomas.

Comments
Let's make it a physical "problem". You set the password as the combination to your briefcase, and you have something of value in the briefcase. How might evolution "solve the problem"? Maybe it will invent a hacksaw. Or maybe it will invent some acid that disolves the lock on the briefcase. Evolution is not a search. Confining it to a fixed formal setup is unrealistic.Neil Rickert
November 23, 2013
November
11
Nov
23
23
2013
03:42 PM
3
03
42
PM
PDT
Oops messed up HTML
So evolutionary processes are incapable of finding very specified, singular ‘solutions’?
Exactly, because as I said, evolutionary processes are not searches.Alan Fox
November 23, 2013
November
11
Nov
23
23
2013
03:37 PM
3
03
37
PM
PDT
So evolutionary processes are incapable of finding very specified, singular ‘solutions’? Exactly, because as I said, evolutionary processes are not searches.
Alan Fox
November 23, 2013
November
11
Nov
23
23
2013
03:35 PM
3
03
35
PM
PDT
Regarding Sal's comment #1, it's unclear to me how AI can increase information at all, since an algorithm is essentially a necessity mechanism -- a law -- where sufficient conditions produce a reliable, repeatable result given identical input from an initial starting state s0. This automaton is fascinating: see Jaquet-Droz The Writer, and Video: The Writer. It produces a handwritten message with ink and quill on paper, and it's a product of clock-making genius. However it doesn't actually produce any information, rather it's programmed with cams and cam followers to transfer preexisting information from one form/medium to another. At best it seems that an algorithmic system might incorporate some sort of true randomness, however that would merely introduce arbitrary inputs at prespecified decision points, it doesn't change the necessary mapping of inputs to outputs, so F(A) → B, regardless of whether A is random or not, and we still have a necessity mechanism -- a law contrived from a contingent arrangement of matter. Certainly such an algorithm could be programmed to rewire itself on the fly, so that F(A) → C instead of B at some point in the execution, but that itself would also be the result of a necessity mechanism, or at best a random occurrence, which would limit the new information it could plausibly generate to somewhere around the universal probability bound. (I'm attributing law-like properties to algorithms and physical automata because in all cases it would seem that sufficient antecedent conditions produce a reliable consequent.)Chance Ratcliff
November 23, 2013
November
11
Nov
23
23
2013
03:32 PM
3
03
32
PM
PDT
So evolutionary processes explain anything they are proposed to explain.
Nope. What an odd question! Evolutionary processes are not searches. Solutions lie in wait and trap unsuspecting populations into new niches. It's so unfair!Alan Fox
November 23, 2013
November
11
Nov
23
23
2013
03:32 PM
3
03
32
PM
PDT
But even when writing password login pairs the designer has access to both parts (the login and password). In the case of proteins, the architecture of the necessary component is unknown. For example, a living creature may need insulin. The requirement of what the insulin protein need to do are clear, actually constructing it from scratch is pretty difficult. Even we humans with all our technology cannot resolve the protein folding problems in sufficient detail. It's combinatorially prohibitive, like passwords, it doesn't lend itself to simple laws. So add that complication to the two cases in my previous comment, and that is why evolutionary algorithms (be they biological or Weasels in Dawkins computer), cannot resolve them in geological time.scordova
November 23, 2013
November
11
Nov
23
23
2013
03:31 PM
3
03
31
PM
PDT
Nope, they don’t even find those, they throw them away (Behe’s rule).
Pin your colours to that mast if you want Sal. Disappointment awaits. Richard Lenski's work shows the power of evolutionary processes in an asexual environment. Given eukaryotes and sexual reproduction, the sky's the limit!Alan Fox
November 23, 2013
November
11
Nov
23
23
2013
03:29 PM
3
03
29
PM
PDT
But evolutionary processes don’t find particular solutions like passwords. They find any solution!
So evolutionary processes explain anything they are proposed to explain.Upright BiPed
November 23, 2013
November
11
Nov
23
23
2013
03:28 PM
3
03
28
PM
PDT
But evoplutionary pcesses don’t find particular solutions like passwords. they find any solution!
Nope, they don't even find those, they throw them away (Behe's rule). Two cases: 1. protein doesn't exist, but the rest of the system already exists, but it needs the protein to function 2. protein exists already, and we build a system to utilize it The difficulty of case #1 is easy to see the analogy to the password login. Case #2 is actually worse. You form a password first, then you have to build a computer system to utilize it! Just because there are an infinite number of solutions to a problem doesn't mean the probability of solving that problem is close to zero. There are an infinite number of ways to write solutions to Einsteins field equations, doesn't mean you'll find all of them. :-)scordova
November 23, 2013
November
11
Nov
23
23
2013
03:25 PM
3
03
25
PM
PDT
But evolutionary processes don’t find particular solutions like passwords. They find any solution!
So evolutionary processes are incapable of finding very specified, singular 'solutions'?nullasalus
November 23, 2013
November
11
Nov
23
23
2013
03:23 PM
3
03
23
PM
PDT
Lets try that again with spell-check: But evolutionary processes don’t find particular solutions like passwords. They find any solution!Alan Fox
November 23, 2013
November
11
Nov
23
23
2013
03:20 PM
3
03
20
PM
PDT
But evoplutionary pcesses don't find particular solutions like passwords. they find any solution!Alan Fox
November 23, 2013
November
11
Nov
23
23
2013
03:18 PM
3
03
18
PM
PDT
What? How on earth is this relevant to evolution?
If genetic algorithms can't solve passwords, why should they be expected to solve complex proteins. It's the same class of problems except maybe my password problem was easier!scordova
November 23, 2013
November
11
Nov
23
23
2013
03:12 PM
3
03
12
PM
PDT
"How on earth is this relevant to evolution?" Evolution Vs. Functional Proteins - Where Did The Information Come From? - Doug Axe - Stephen Meyer - video http://www.metacafe.com/watch/4018222/ Stephen Meyer - Proteins by Design - Doing The Math - video http://www.metacafe.com/watch/6332250/ "Biologist Douglas Axe on Evolution's (non) Ability to Produce New (Protein) Functions " - video Quote: It turns out once you get above the number six [changes in amino acids] -- and even at lower numbers actually -- but once you get above the number six you can pretty decisively rule out an evolutionary transition because it would take far more time than there is on planet Earth and larger populations than there are on planet Earth. http://intelligentdesign.podomatic.com/entry/2012-10-15T16_05_14-07_00 Doug Axe PhD. on the Rarity and 'non-Evolvability' of Functional Proteins - video (notes in video description) http://www.metacafe.com/watch/9243592/bornagain77
November 23, 2013
November
11
Nov
23
23
2013
02:58 PM
2
02
58
PM
PDT
How on earth is this relevant to evolution?
It's not as such or only peripherally. Sal's after Winston Ewert's job or better! Ewert didn't cope too well with Lizzie Liddle Sal thinks he can do better!Alan Fox
November 23, 2013
November
11
Nov
23
23
2013
02:51 PM
2
02
51
PM
PDT
I once offered to donate $100 to Darwinist Dave Thomas’ favorite Darwinist organization if he could write an genetic algorithm to solve a password. I wrote a 40-character password on paper and stored it in safe place. To get the $100, his genetic algorithm would have to figure out what the password was ... Dave Thomas, like Richard Dawkins, advertise the supposed mighty power of genetic algorithms, but when pressed to solve the sort of problems that are relevant to evolution, they are no where to be seen. What? How on earth is this relevant to evolution?wd400
November 23, 2013
November
11
Nov
23
23
2013
02:31 PM
2
02
31
PM
PDT
...towards a certain outcome.CentralScrutinizer
November 23, 2013
November
11
Nov
23
23
2013
01:10 PM
1
01
10
PM
PDT
WEASEL is a front loaded program with a guaranteed outcome. But to be fair to Dawkins, himself acknowledged that WEASEL is not actually an analog of Darwinian evolution but that it is merely an illustration of how small changes might accumulate by whatever means over time in a system.CentralScrutinizer
November 23, 2013
November
11
Nov
23
23
2013
01:09 PM
1
01
09
PM
PDT
SC: As in isolated islands of function in a vast sea of non-functional configs. Where also the available time and resources are such that only a negligibly small fraction of possibilities can be blindly searched through chance and/or necessity. For 500 bits, and solar system scale resources [10^57 atoms, ~ 10^17 s, one search per atom per 10^-14 s, effectively fastest Chem rate] my back of envelope estimate is search to space is as one straw sized sample to a cubical haystack 1,000 light years across, about as thick as our galaxy at its thickest. Thus, Chi_500 = i*s - 500, functionally specific bits beyond the threshold. The predictable reaction is stout resistance and refusal to take the issue seriously -- this is essentially what we find in NFL. Mix in singleton proteins aplenty, the OOL challenge, missing body plan evo links, lack of observed capacity of mechanisms to do the job and there is a serious challenge to answer to. Hotly denied of course. KFkairosfocus
November 23, 2013
November
11
Nov
23
23
2013
01:04 PM
1
01
04
PM
PDT
Very interesting post. I need to spend some time to digest it because it touches on things that are dear to my heart such as the limits of artificial intelligence.Mapou
November 23, 2013
November
11
Nov
23
23
2013
01:01 PM
1
01
01
PM
PDT
NOTE #3 In the case of my challenge, a 40-character case-sensitive alphanumeric password has 62^40 possible configurations, which yields a Shannon uncertainty (entropy, information) of: log2(62^40) = 238 bits If I cheated myself by giving Dave Thomas the password up front, he could easily write a Weasel-like program because he knows the target in advance. But in such case, there would really be no uncertainty, and if there is no uncertainty, his algorithm really would not have more insight than the information that was front loaded. If I gave him the information up front, there really would be only one final answer (the correct answer) by his algorithm, and thus the Shannon uncertainty in such case would be: log2(1) = 0 bitsscordova
November 23, 2013
November
11
Nov
23
23
2013
12:44 PM
12
12
44
PM
PDT
NOTE #2 Evolutionist will argue that evolution doesn't work with explicit targets like Weasel. What No Free Lunch illustrates is that knowledge of the structure of the search space reduces uncertainty about the final answer, hence even if the target is not explicitly stated, there is no ultimate uncertainty about the final answer in certain cases. The algorithm effectively only used what was known, it did not somehow create knowledge out of a vacuum (doing so would be making free lunches). What is an illustration of having knowledge of the target, but not explicitly stating it? For example if I added the integers 1 to 1000: 1 + 2 .... 1000 = 500,500 I could write a Weasel-like program that will converge on the answer of "500,500" (in the way Dawkins mutated letters), or I could write a more clever Weasel-like program that uses Gauss' formula such that nowhere will the phrase "500,500" be found explicitly in the algorithm, but still the algorithm will converge on the right answer. I wrote just such an algorithm, but unfortunately, the original source code is lost. All we have left on the net are traces of the original debate about what I wrote: Dave Thomas says Cordova's Algorithm is Remarkable. The point however, is that nowhere in my algorithm did I explicitly use the phrase "500,500", my algorithm implicitly described the target, but I was able to do so because I had a priori knowledge of the search space that effectively reduced the uncertainty over the final answer to zero.scordova
November 23, 2013
November
11
Nov
23
23
2013
12:38 PM
12
12
38
PM
PDT
NOTE #1: There has been some debate over whether an evolutionary system (be it bacteria or robots) can create CSI that wasn't pre-existing. In the case of the password challenge above, it is clear, an algorithm or robot cannot reduce the uncertainty about the password, no matter what strategy is used. That uncertainty is pretty much immutable, and the consequences of the No Free Lunch (NFL) theorems are blatantly in evidence. No algorithm, Darwinian or otherwise, will get a free lunch in solving the password without sneaking in some specialized knowledge. The password challenge is a closed system, and clearly there can be no CSI increase in such closed systems. But what about open systems? There has been recent debate at UD whether there can be information increase in scenarios outside of blind search, like say a Robot studying and environment and then constructing Rube Goldberg machines. I support the NFL theorem's conclusions with respect to blind search (a closed system), but I'm not quite so enthusiastic to apply NFL theorems outside of blind search (open systems). What do I mean? I've given illustrations where I think CSI can arguably be increased in an open system by AI, namely robots building Rube Goldberg machines. Whereas for a closed system, like the password problems, it is clear CSI cannot be increased by any AI system (a Darwinian algorithm is an instance of an AI system). I don't think the debate will have resolution because, unlike closed systems, the measures of information for open systems are quite ambiguous. I tried to illustrate the information measurement problem in: The Paradox in Calculating CSI Numbers for 2000 coins. The conflicting answers in that discussion illustrate the difficulty in applying NFL to open systems. I know several of my UD colleagues will disagree with me on this issue, and I don't think it has resolution because of the paradoxes that arise in measuring information in physical artifacts. Even though we might say biology is in an open system, it does not mean evolution will not be challenged by the problem NFL poses for closed systems. The protein evolution problem can be made analogous to a closed system password problem, and hence Darwinian evolution is precluded as being able to evolve sufficiently complex proteins. I'm quite enthusiastic to apply NFL in that context. Also, I pointed out, even supposing CSI can increase in an open system, there is no empirical evidence real selection in the wild will do so. See: How Darwinists confuse Extravagant with Essential.scordova
November 23, 2013
November
11
Nov
23
23
2013
12:37 PM
12
12
37
PM
PDT
1 2

Leave a Reply