Good guest post at Uncommon Descent by Aurelio Smith,
SIGNAL TO NOISE: A CRITICAL ANALYSIS OF ACTIVE INFORMATION
For those who prefer to comment here, this is your thread!
For me, the argument by Ewert Dembski and Marks reminds me of poor old Zeno and his paradox. They’ve over-thought the problem and come to a conclusion that appears mathematically valid, but actually makes no sense. Trying to figure out just the manner in which it makes no sense isn’t that easy, though I don’t think we need to invent the equivalent of differential calculus to solve it in this case. I think it’s a simple case of picking the wrong model. Evolution is not a search for anything, and information is not the same as [im]probability, whether you take log2 of it or not. Which means that you don’t need to add Active Information to an Evolutionary Search in order to find a Target, because there’s no Target, no search, and the Active Information is simply the increased probability of solving a problem if you have some sort of feedback for each attempt, and partial solutions are moderately similar to better ones.
Enjoy!
One confusing point about the UD discussion is whether or not the definition of a “search” which involves fitnesses itself includes the fitness surface.
It depends on who’s doing the defining. Most discussions of optimization will involve a search algorithm that looks nearby on a fitness surface, perhaps also using information about past events. For example, matrix update methods that gradually build up an estimate of the local curvature of the surface. Then we can ask how these search algorithms behave on different shape surfaces. It is therefore clear that in these discussions the fitness surface is not included in the definition of the search algorithm.
Dembski, Ewert, and Marks define a “search” as a probability distribution of outcomes. They thus incorporate the fitness surface into the “search”, if the “search” results from a process that considers the fitnesses.
Thus yes, a “search” (per DEM) incorporates the fitness surface, and no, a search algorithm (per most researchers on optimization) does not incorporate the fitness surface.
Thanks for the clarification, Joe!
I added a title. That way, people seeing the summary on the front page have a link they can click to get to full post with comments.
I’m so out of the TSZ loop I forgot there was a title field!
Nice to see you guys 🙂
It doesn’t matter Lizzie, because we have already leaned this week that the way to conclude there is design on Earth is to simply look at it and come to the conclusion that it looks designed, just like keiths and Alan taught us when looking at carvings.
So really they don’t need anything any more complicated than this.
No, that’s not right.
You look at it, and form a tentative opinion. But now you must try to get beyond that tentativity, and that requires more evidence.
Like Mung, phoodoo has no confidence in his ability to defend his beliefs.
It’s understandable. None of us have any confidence in him either.
Welcome back Lizzie!
On the substantive matter of Lizzie’s post, I’m wondering if we shouldn’t have two threads. One that is dedicated to discussing the DEM papers and the article at Pandas Thumb by Joe Felsenstein and Tom English. And keep this one for welcoming Lizzie back. Joe’s signal comment seems to have got lost in the noise.
OK, looks like I have to remember how to move posts….
ETA: OK, remembered!
I moved some comments that seemed better suited to Lizzie’s “Hi Everyone” post.
Thanks, Alan!
We already have a thread here already for discussing the DEM papers in the context of the PT thread.
(Not sure which is my “signal comment”).
Let me make a few quick points about the relation of DEM’s theorems to the No Free Lunch theorem.
1. At the UD thread there were some loud dismissals of models that had genotypes and a fitness surface. It was declared that these genetic algorithms weren’t models of evolution. Actually DEM called such models “evolutionary search”, so they don’t seem to agree with the ID supporters in the UD thread.
2. At the UD thread it was also declared that the NFL theorem applied to models of evolutionary search. The NFL theorem works by considering a space of (say) genotypes and not one fitness surface, but all possible ways that the same set of fitnesses can be associated with the genotypes. It then makes a generalization about the average behavior, averaged over these. A typical (randomly selected) association of fitnesses with genotypes is a “white noise” surface. Real fitness surfaces are smoother than that “because physics”. This point has been made repeatedly since 2002 (originally by Richard Wein, by Jason Rosenhouse, and by Elsberry and Shallit, subsequently by several other people). It is explained carefully in my 2007 article. The UD commenters have not understood that the NFL theorem does not establish that evolution on a real fitness surface will fail to improve fitness.
3. The DEM paper has theorems saying that a randomly chosen “search” (as they define it) will not do better than random choice of a single genotype. The NFL theorem says that evolution on a randomly chosen fitness surface does as well, on average, as finding the best of m randomly chosen genotypes. That is much better than in DEM’s theorems. Why? Because DEM’s “searches” include all sorts of crazy searches that, when presented with m genotypes, don’t even choose the best of these. These “searches” might even choose the worst of them!
So the NFL theorem already presumes an algorithm that, presented with two or more genotypes, ends up with a better-than-average one of these. DEM argue that when we always have a search that does this, it has a fair amount of “Active Information”. Just having organisms that reproduce, and have fitnesses that affect that reproduction, is enough to get that Active Information. And “because physics” those fitnesses are even smoother, and the evolutionary search will do even better, than a the choice of the best of m points on a white noise fitness surface.
Let me qualify that. The NFL theorem establishes a bound. It says that m steps of an optimization algorithm, on average over all possible mappings of fitnesses to genotypes, cannot do better than choosing the best of m randomly chosen genotypes.
It does not presume that the algorithm will achieve this bound, it says that no algorithm can do better than that. The best possible algorithm does as well as that. DEM’s “searches” include lots of algorithms that do horribly worse than that, and they argue that having that best possible algorithm means that you have been provided with lots of Active Information. So having that algorithm together with a smoother-than-white noise fitness surface means that you have been provided with even more Active Information.
Indeed. But As Lizzie picked the topic of Aurelio Smith’s guest post at UD, naturally enough there were comments welcoming her back which seemed to drown your comment a bit.
See above. “Signal” was a lame attempt at punning on the UD post title.
DEM have never used active information in substantive analysis of a “search.” What they actually do is to determine what contributes to the average number of “queries” (e.g., fitness evaluations) required to “hit the target.” However, they create the appearance of doing more. They divide the average number of queries, E[Q], by the endogenous information, –log p, where
p = |T| / |Omega|
is the fraction of the search space Omega that is in the target. Because the target T is fixed in each analysis, the endogenous information is a constant, That is,
SENSIBLE = E[Q] / –log p queries per bit
is the average number of queries required to hit the target, normalized by a constant indicating the hardness of the problem. Investigating what contributes to the SENSIBLE measure of performance of the “search,” the normalization is irrelevant. Everything comes down to E[Q]. Evidently, SENSIBLE doesn’t feature information sufficiently for propaganda purposes. DEM flip it on its head:
AAIPQ = 1 / SENSIBLE = –log p / E[Q].
They call this the average active information per query, when it’s endogenous information per average number of queries. It’s definitely not some sort of average of active information, which looks like this:
AI = log q — log p.
The Takeaway. Their “conservation of information” theorem is for active information, and does not apply to the analyses of ev, Avida, the Weasel program, etc., which involve a measure that is quite different from active information.
Intrepid leader Barry Arrington has joined the fray! He seems very interested in the mechanisms of emergence. I wonder if he will be demonstrating the mechanisms of ID?
Richardthughes,
No, he’s interested only in getting Lizzie to admit to the implications of materialism. Getting her to flit between Darwinism and materialism is a score for him. The story being told in churches across the United States is that the scientific evidence is strongly against Darwinism, and that it is only a commitment to a materialistic worldview that keeps scientists from admitting to the truth.
Tom English,
We see it in Joe Gallien and Phoodoo here too, questioning the beliefs of others but a complete inability to give an account or explanation of their own views. I suppose religion feels it has been attacked by science in some regards so I understand it, but if it was concerned with the truth rather than being perceived as right we likely wouldn’t have this friction .
Of course BA is not willing (or not capable) of restraining himself from being pig-rude, but in this thread he is also pig-ignorant:
Barry is delusional. Just imagine thinking that you’re making any valid point in a scientific discussion by reference to a 2300-years-dead philosopher (who claimed that human males have more teeth than human females do).
And molecules have the potency “iPhone” under certain conditions.
There seems to be a problem at UD that some people seem to think that because the premise of “materialism” is flawed, any argument any “materialist” poses against any ID argument must also be flawed – and so can be countered by any anti-materialist argument!
My objection to ID (as it used to be anyway) is that I think that the inferential argument is flawed, not that the conclusion is incorrect. Similarly, the arguments made by Provine and Dawkins that the evidence indicates no Designer is also flawed in my view. We can falsify specific Designer hypotheses, but we cannot rule out a Designer in principle, nor rule one in. Materialism is no more falsifiable than Theism is. So science is a threat to neither.
But you wouldn’t think that, looking at the discussion. I blame Lewontin.
Elizabeth,
This is why I’ve always thought that Deism is more likely that Theism, because it lacks the falsified truth claims. No matter what you choose at some point you must assert something with no explanation.
☺
I tried to follow the UD thread but it keeps getting interrupted by KF’s paranoid rants and accustoms against Elizabeth for some past imaginary indiscretion. What I don’t understand is why Barry doesn’t intervene. Accepting this behaviour just makes UD look foolish
Pay no attention to the man behind the curtain.
If Barry intervened, it would be to ban Elizabeth. He would not ban KF.
He’ll ban me anyway, I expect, shortly.
If he does, it will be done silently. He will then accuse you of running away and, if asked, deny that he banned you.
The standard line of criticism that’s been perfected at UD goes like this: materialism is self-refuting, so it’s not worth taking seriously. And why is materialism self-refuting? On this point they give us a lot of quotes from Darwin, Haldane, C. S. Lewis, and Plantinga.
However, they don’t notice or care that there are plenty of responses to these arguments. Since they don’t notice or care that there are responses, they have no idea whether the responses are good or not. Perhaps if they took any time at all to look at the responses to that argument, they would decide that the responses are bad and that the original argument is good. But they don’t even do that. As a result they are in no position to evaluate whether the “materialism is self-refuting” argument is really a good argument or not. They want it to be a good argument, but they are too lazy to put any work into determining whether or not it’s a good argument. They’d rather recycle quotes than have to actually think about any of the really hard issues.
Never mind that no one knows what they mean by “materialism”, though Alex Rosenberg is often trotted out as a good example of a materialist. They don’t know (or want to know) that Dennett and Kitcher are severely critical of Rosenberg, and they don’t want to know, because they need Rosenberg to be The Materialist.
It is worth a moment to notice a parallel here: the UD crowd has the same attitude towards “soft naturalists” (Dennett or Kitcher today) that the New Atheists have towards religious liberals (think here perhaps of Marcus Borg or Karen Armstrong). On both sides — both the most extreme of the New Atheists and the most extreme of theocratic reactionaries — one finds the attitude that anything less that the most extreme view is to be rejected as waffling, inconsistency, or hypocrisy, and the most extreme view is to be eliminated at all costs. I do not think I am exaggerating when I suggest that such an attitude is implicitly fascistic.
Phoodoo, that is an excellent description of most ID comments at UD.
The tone of a blog is only as good as the moderator. And the best way to do that is to lead by example. And Barry’s example, on a good day, is deplorable. We can’t blame people like KF, Mapou and Joe, who are just following Barry’s lead.
Moved phoodoo’s post to Hi Everyone
They need Rosenberg. He’s their kind of materialist, in that he says the things that they want to criticize.
I tend to think of Rosenberg as taking an extreme position and trying to defend it, as his way of doing philosophy.
That’s surely right. My point was that they seem to have no interest in various naturalists who disagree with Rosenberg. If they were to do so, they might have to abandon their commitment to the tenets that naturalism entails nihilism and/or that naturalism is self-refuting.
Kantian Naturalist,
The whole ID program seems to be aimed at rubes in pews. Its aim is to keep them in the pews. It’s a kind of Potemkin village.
They are all as mad as a whole guild of hatters. I posted the following on Aurelio Smith’s thread:
Result? No reply, but my “Don Pedro” alias was silently banned soon afterwards (presumably by Big Bazza himself: he has now appeared in person, with more rude inanities addressed to Lizzie). Gordo is a sacred cow and his paranoia mustn’t be mentioned.
Barry once advised me (here) to retain legal council. I’d like to now advise him to retain his toys in his pram.
Well folks. This is all really good news.
It seems that the DEM arguments are THAT good.
The last line of defense seems to be ‘well, they are playing ball at the wrong park after all’.
……thank our chancy stars we dont have to keep walking down the ‘active’ path. The sandwalk has prettier bulbs anyway.
Gooooood job DEM!!!!!
Now, just how to transfer the money in Joe’s grant box….into DEM’s grant box.
After all, that’s where the TRUE runs are scored.
I guess no AVIDA search will do. It has no sense of greenback. It takes AI to smell the green.
Steve:
Earth to Steve…
ETA: To see just how bad their arguments are, see the comments in this thread.
Barry’s choice of words / actions is a dead giveaway of religious-based petty dictatorship. Of course, they can and do schism with each other (even unto the point of genocide when they have actual worldly power) but when they’re not in the midst of some doctrinal schism, they’re compelled to ignore their co-religionist’s faults in order to attacki the “common enemy”.
And being Right-Wing Authoritarians, even if they weren’t already driven mad by religion, they would certainly be paranoid and in constant search of “enemies” to blame for everything that goes wrong in their own lives. After all, if you’re an RWA, what’s wrong can’t be that you’ve personally chosen the hateful bullying lifestyle; it can’t be that you’re disliked because you’re a racist sexist homophobic asshole; it can’t be that decent humans rightfully resist you because you’re trying to force bronze-age clan moralities onto a world which (finally, thank god!) embraces inclusion and cooperation. No, no, no, it MUST BE SOMEONE ELSE’s fault.
Someone like Lizzie? Well, sure, she makes a great scapegoat. For now, until they move on to the next 2 minute hate.
When Barry’s control unravels, he shows himself to be every bit as bad – or worse – as KF.
What is amazing is that he has not yet appeared to have banned “Unwilling Participant”, who has berated KF for his behaviour, berated Barry for not doing anything about KF, and called Barry rude and arrogant in his behaviour towards Lizzie.
A brain is a mediocre commodity. Where I come from men operate websites with no more brains than a scarecrow. All you really need is a diploma or two.
Barry is now arguing as “UDEditors”, using the first person plural of power and defacing the adversary’s comments with interlineal retorts:
There’s a nice knock-down argument for you!
Apologies for this very long comment. The points made in Aurelio’s guest-post at UD have remained almost entirely unaddressed in the UD thread’s comments.
The host of the guest post, johnnyb (self-identified as Jon Bartlett), has (here) posted his views on Aurelio’s points. They deserve at least a quick response, so let me try, here.
I was the one who used the word “crazy” to describe most of the searches that DEM include in their universe of searches. They include ones that deliberately try to find worse fitnesses, for example. The point, which Aurelio quoted, was that simply having organisms that have fitnesses excludes all the crazy searches. It thus starts you out with a lot of Active Information.
johnnyb’s points about how one gets major changes (such as new body plans) is a frequent concern at UD, but it has nothing to do with the arguments Aurelio cites. DEM do not make it part of their argument. They are concerned with whether. when an organism does successfully climb a fitness surface, that fitness surface has had Active Information built in. So johnnyb’s argument is not responsive to the issue of Active Information, whether it is useful, and if so, where it comes from.
Again, an issue in some other argument. About Active Information, this point says … what?
Also, I have no idea what a “recursive mutation” is. It sounds scary.
Dembski uses the UPB to rule out “chance” effects such as evolution by pure mutation. In the 2001 version of Dembski’s argument, there was a Law of Conservation of Complex Specified Information which was supposed to prove that evolutionary mechanisms such as natural selection could not put CSI into the genome. The LCCSI accomplished no such thing, and has been quietly abandoned.
DEM’s argument is about Active Information. Yes, it conceded (at least for the sake of argument) that natural selection can put CSI into the genome. DEM do not at all discuss “how much and how far” information can be put into the genome. That may be “the question” in some other argument. But not in DEM’s argument.
The next point johnnyb’s argument is about whether or not a “search formalism” is needed. Aurelio was pointing to our article and whether evolution ought to be modeled as a search for what DEM call a target.
DEM call every process a “search”, even crazy ones. But for all of them their criterion for success is the probability of ending up with a genotype from the predefined Target. Evolutionary theory does not do this. It has fitness surfaces. and there are many papers discussing them. But overwhelmingly the discussion is of whether the fitness is increased, not whether some target is achieved. So no, evolutionary theory does not need a search for a Target.
Slime molds, oak trees, and jellyfish know where they are evolving to? I don’t recognize that as a conclusion of “recent biology”.
I am lost as to what “it” is. Suffice it to say that there has been much analysis of fitness surfaces, but there is no necessity of having a Target.
The poster shows a calculation of the information needed to get to an immunoglobin sequence. It presents a DEM-like discussion of “Active Information” and mentions a “search algorithm”. But in the immunoglobin part of the poster it never mentions any search algorithm other than pure mutation. It does not discuss any fitness surface, or the probability of getting to the immunoglobin on it. So it is mysterious.
Joe Felsenstein,
To get what he was driving at with “active,” rather than “passive,” organisms, think of Shapiro’s Natural Genetic Engineering. That ties in to his references to the SOS response in his abstract and poster.
I guffawed when he referred, ever so confidently, to the “generation of biologists raised on Dawkins.”
People who read mostly popularizations often make that mistake — they think that the most visible popularizers are the most influential biologists.
They don’t know about cosmologists other than Hawking, about particle physicists other than Feynman, or evolutionary biologists other than Dawkins or Gould.
I thought that was very revealing actually. Dawkins is the real bugbear here, I think. ironic that he is (IMO) such so unsound on evolutionary theory. I wonder how many opponents of evolution know that Dawkins isn’t even an evolutionary biologist? Well, apart from the odd paper on digger wasps (and I don’t think he was even first author).
So nobody seems to want to defend Active Information as an ID argument.
Oh well.
KF puts the entire Internet to sleep with five words:
FYI
Ewert has responded (at ENV) to Felsenstein’s and English’s criticisms of the criticisms of the criticisms of the critique of evolution’s searchinesslessness.
Intermission.
Popcorn. Bathroom. Look at trade status on iphone.
Wait.