Subjects: Evolutionary computation. Information technology–Mathematics.
… the authors establish that their mathematical analysis of search applies to models of evolution.
I have all sorts of fancy stuff to say about the new book by Marks, Dembski, and Ewert. But I wonder whether I should say anything fancy at all. There is a ginormous flaw in evolutionary informatics, quite easy to see when it’s pointed out to you. The authors develop mathematical analysis of apples, and then apply it to oranges. You need not know what apples and oranges are to see that the authors have got some explaining to do. When applying the analysis to an orange, they must identify their assumptions about apples, and show that the assumptions hold also for the orange. Otherwise the results are meaningless.
The authors have proved that there is “conservation of information” in search for a solution to a problem. I have simplified, generalized, and trivialized their results. I have also explained that their measure of “information” is actually a measure of performance. But I see now that the technical points really do not matter. What matters is that the authors have never identified, let alone justified, the assumptions of the math in their studies of evolutionary models.a They have measured “information” in models, and made a big deal of it because “information” is conserved in search for a solution to a problem. What does search for a solution to a problem have to do with modeling of evolution? Search me. In the absence of a demonstration that their “conservation of information” math applies to a model of evolution, their measurement of “information” means nothing. It especially does not mean that the evolutionary process in the model is intelligently designed by the modeler.1
I was going to post an explanation of why the analysis of search does not apply to modeling of evolution. But I realized that it would give the impression that the burden is on me to show that the authors have misapplied the analysis.2 As soon as I raise objections, the “Charles Ingram of active information” will try to turn the issue into what I have said. The issue is what he and his coauthors have never bothered to say, from 2009 to the present. As I indicated above, they must start by stating the assumptions of the math. Then they must establish that the assumptions hold for a particular model that they address. Every one of you recognizes this as a correct description of how mathematical analysis works. I suspect that the authors recognize that they cannot deliver. In the book, they work hard at fostering the misconception that an evolutionary model is essentially the same as an evolutionary search. As I explained in a sidebar to the Evo-Info series, the two are definitely not the same. Most readers will swallow the false conflation, however, and consequently will be incapable of conceiving that analysis of an evolutionary model as search needs justification.
The premise of evolutionary informatics is that evolution requires information. Until the authors demonstrate that the “conservation of information” results for search apply to models of evolution, Introduction to Evolutionary Informatics will be worthless.
1 Joe Felsenstein came up with a striking demonstration that design is not required for “information.” In his GUC Bug model (presented in a post coauthored by me), genotypes are randomly associated with fitnesses. There obviously is no design in the fitness landscape, and yet we measured a substantial quantity of “information” in the model. The “Charles Ingram of active information” twice feigned a response, first ignoring our model entirely, and then silently changing both our model and his measure of active information.
2 Actually, I have already explained why the “conservation of information” math does not apply to models of evolution, including Joe’s GUC Bug. I recently wrote a much shorter and much sweeter explanation, to be posted in my own sweet time.
a ETA: Marks et al. measure the “information” of models developed by others. Basically, they claim to show that evolutionary processes succeed in solving problems only because the modelers supply the processes with information. In Chapter 1, freely available online, they write, “Our work was initially motivated by attempts of others to describe Darwinian evolution by computer simulation or mathematical models. The authors of these papers purport that their work relates to biological evolution. We show repeatedly that the proposed models all require inclusion of significant knowledge about the problem being solved. If a goal of a model is specified in advance, that’s not Darwinian evolution: it’s intelligent design. So ironically, these models of evolution purported to demonstrate Darwinian evolution necessitate an intelligent designer. The programmer’s contribution to success, dubbed active information, is measured in bits.” If you wonder Success at what? then you are on the right track.
The Series
Evo-Info review: Do not buy the book until…
Evo-Info 1: Engineering analysis construed as metaphysics
Evo-Info 2: Teaser for algorithmic specified complexity
Evo-Info sidebar: Conservation of performance in search
Evo-Info 3: Evolution is not search
Evo-Info 4: Non-conservation of algorithmic specified complexity
Evo-Info 4 addendum
They’d need to turn to the evidence in order to show that their “analysis” does–or does not–apply. Unfortunately for them, the evidence is the problem, since it indicates that evolutionary processes did occur, while intelligence did not step in to transcend the limitations of evolutionary processes in any meaningful way.
They’re not trying to model what happens, but to show that it didn’t. It’s very easy to fail to model what happens, so they manage to do that every time.
Glen Davidson
So Tom. Where can we find mathematical analysis of models of evolution? Or is such analysis simply not needed.
You and I both know there is a long history of ‘information theory’ in biology quite independent of DEM. Just a passing fad of the times I suppose. An attempt to make biology sound mathematical and scientific.
GlenDavidson,
Sorry not to have made it clear that they analyze the models of others. See footnote a. They address a kind of model that specifies an evolutionary process along with an event that tends to occur in the process. The specification of the event is what they regard as a problem solved by the evolutionary process. See a thing or two wrong with that?
Only the same excruciating mistake as always.
But what are they supposed to do, understand evolution?
Glen Davidson
And you commented on Joe Felsenstein’s “Wright, Fisher, and the Weasel.” The thread was all about verifying that mathematical analysis of a special case of the Wright-Fisher model held approximately for the Weasel model.
I honestly don’t understand why you’re bringing this up.
Marks, Dembski, and Ewert have developed mathematical analysis of search for a solution to a given problem. That’s engineering analysis of a problem-solving technique. So the question is whether the assumptions of the analysis hold for models that specify an evolutionary process along with an event that tends to occur in the process. In other words, the modeler is supposed to have devised an evolutionary process in order to generate a solution to a given problem. Does that make sense to you?
Their theorems show that on average over all possible “evolutionary searches” their performance is no better than drawing a single random genotype. The problem in applying that to models of evolution is that their space of “evolutionary searches” is way too broad. It includes “searches” that prefer genotypes of worse fitness, and also “searches” that are themselves totally uninformed by fitness. The average performance of these is, of course, pathetic.
What we showed by our “GUC Bug” example was that once there are genotypes that have fitnesses, and given that those fitnesses affect the survival and reproduction of the organisms, the model performs far better than choosing a random genotype. Now those are pretty minimal conditions for calling something an evolutionary model.
Conclusion: their argument does not establish that models of evolution on average do no better than choosing a random outcome. As Tom argues, this is a huge apples-orange problem. And this argument is absolutely central to their book.
Tom says they are comparing apples and oranges. Joe F. disagrees with Tom. For Joe F. it’s a comparison between ripe fruit and rotten fruit. But he won’t actually say he disagrees with Tom in spite of the rather obvious contradiction evidenced in his post.
Evolutionary algorithms (and genetic algorithms) are search algorithms. They are problem solving algorithms. Engineering. I thought we had finally managed to agree on these points.
Sure, it’s tough on all the anti-ID crowd here at TSZ to eat that crow after all this time. But the truth will out.
So the only question that remains is just what sort of models of evolution are running around out there that are not based on an evolutionary search algorithm?
Where are you getting this shit?
No, they can be used as problem solving algorithms. Some of them are really just attempts at simulating an evolutionary process analogous to the real biological one, to see how it works and what results from it.
What the flying fuck are you blathering about here?
Why is that the only question that remains? I don’t even see why that would be an interesting question in the first place.
The most interesting question, to me, is whether some particular simulation of evolution is similar enough to the real thing to draw real-world conclusions from it.
Does Mung mean “evolutionary algorithms” as employed by engineers to solve problems? Or models of evolutionary processes, as employed by biologists to understand them better?
The former can be thought of as searches. The latter are rarely formulated that way, as you can verify by reading chapters from my book on theoretical population genetics.
Implicit (but of course not explicit) in Mung’s jibe is the assumption that if it can be established that models of evolutionary processes are “evolutionary searches” as defined by Marks-Dembski-Ewert, that they then have theorems showing that evolution cannot succeed in improving adaptation. Their theorems show no such thing, as Tom and I have carefully shown (as noted above and also here).
So the whole effort to show that models of evolutionary processes are “searches” is basically pointless.
Perhaps Mung will clarify for us why he wants to establish this, what establishing it would accomplish.
The relationship I laid out is this: a search for a solution to a problem is a process that
1. samples the space of possible solutions, and
2. outputs the best solution it can find among those sampled.
When the sampling is done by simulation of an evolutionary process, the search is described as evolutionary, even though the solution-seeking component bears no relation to evolution. This decomposition of search comes from the “no free lunch” analytic framework. The framework given by DEM in a technical paper (the one that you and I studied closely) decomposes search into more components. I’ve merely coarsened their decomposition (lumped some of their components into my sampling component, and lumped the remainder of their components into my solution-seeking component).
In their conflation of evolutionary model and evolutionary search, Marks, Dembski, and Ewert are essentially substituting the sampling component of an evolutionary search for the entire search, and referring to it as an evolutionary model because it simulates evolution. (I don’t refer to all simulations of evolution as models of evolution. I’m just explaining what’s going on with them.)
Again, MDE address a kind of model that specifies an evolutionary process along with an event that tends to occur in the process. They take the specification of the event to be a problem that the modeler seeks to solve. They never say so clearly, but it is implicit in their analysis that the modeler devises an evolutionary process in order to generate a solution to the problem — exactly what someone like me does in selecting the sampling component of an evolutionary search.
They dare not say sampling instead of search, because search is the rhetorical fig leaf they wear when referring to the performance of a sampling process (in generating a solution to the problem) as information. Of course, as proponents of the “intelligent design” offshoot of “creation science,” they are committed to accounting for everything in terms of information. Here the practice is obviously ludicrous, at least to the technically literate. A sampling process is statistically independent of the data associated with the sample. When the data are “fitnesses” associated with possible solutions to a problem, the sampling process (perhaps a simulation of an evolutionary process) does not gain information by processing the fitnesses. Samplers are distinguished by their biases, not their information. Marks, Dembski, and Ewert have a lot to say about NFL, and yet are oblivious to the fact that there’s not much more to NFL than statistical independence of the sample (e.g., genotypes) and the data (e.g., fitnesses).
I hope it’s becoming clear to you, if no one else, why I am big on the decomposition of search into a sampling component and a solution-seeking component. It’s not just that it allows for an explanation of their conflation of evolutionary modeling and evolutionary search. It also gets at how wrong they are in equating information with the performance of a sampling (perhaps evolutionary) process in generating a solution to a problem.
Obviously, I decided to suppress most of this in a general-audience response to the book. Now I’m blurting it out. Hopefully it’s of some use. Back to what’s really important…
The problem.
They have no justification for turning the modeler’s specification of an event into a problem. The math assumes that the problem is determined independently of the solution-generating process. That is, the choice of a search depends on the problem, but the problem does not depend on the choice of a search. The modeler jointly specifies an evolutionary process and an event that tends to occur in the process. You made that plain as day in the GUC Bug. The (singleton) event specified in the model is whichever genotype happens to be fittest. That violates the assumption of the math. The event is not determined independently of the evolutionary process. I cannot imagine any case in which a modeler specifies the event independently of the evolutionary process. The dependence of the two is pretty much the point of the model: the event is likely to occur under some circumstances and not others. But I really don’t want to get into that, because the model is what the modeler says it is. There are no debates to be had. The default assumption obviously is not independence, because that’s a very strong condition. If the modeler does not say that the event is specified independently of the evolutionary process, then there is no justification for saying what the modeler does not say. Considering whom we’re dealing with, I should emphasize that it’s generally impossible to decide the matter by reading the code of a computational model.
I do apologize for going on so long. It just started coming, and I just let go with it.
Tom, while we all wait for Mung to answer the questions, one question for you. If I have an evolutionary process that has a population of genotypes, and
1. they generate a generation of newborns, and
2. each of those has a fitness, and
3. the probability that it survives to adulthood is proportional to that (so we enrich the population for more fit genotypes)
… in your terminology, is #3 the sampling component or the solution-seeking component?
Thanks for the prompt. It’s very important to note that differential (survival and) reproduction is not search. The evolutionary process is entirely distinct from the solution-seeking component. The solution-seeking component essentially monitors the evolutionary process. It’s like aliens watching from outer space, and swooping in to harvest a tasty human that emerges (or perhaps one with an enticing anus — not that I have any idea what sort of anus an alien seeks to probe). I should mention for Mung’s sake that, although the implementations of the two components are ordinarily interleaved, the components are logically distinct, and amenable to implementation as distinct computational processes.
That leads me to remark on the danger of reading simulator code, with limited knowledge of evolutionary biology, and mistaking what occurs in the simulation for what occurs in the simulated process. MDE see fitnesses processed as data. It looks as though the fitnesses are informing the process what to do. They see fitness-proportionate selection in the simulation. But what occurs in the simulation is not what actually occurs in the simulated process. Typically, there are many more newborns than will survive to adulthood. You’re using fitness here as a description of the propensity of (presumably a type of) an organism to reach the age where it can reproduce. But the typical simulator will not generate the newborns that do not reach adulthood. It will generate, for the sake of computational efficiency, just those that survive to adulthood. So what one sees in reading the code of the simulator is quite unlike what occurs in the simulated process.
I suspect that MDE are mistaking the simulator for the simuland. They emphasize computer and program and programmer constantly, when they’re actually irrelevant. That’s why I am religious in referring to the modeler and the model and the evolutionary process.
I’m not using my example to full effect. The evolutionary process does not “know” what the aliens are seeking. The fitness of a human genotype — let’s say the expected number of offspring for an individual of that type — is not information guiding the species to evolve anuses of a kind that aliens seek to probe. The search is on the part of the aliens, not evolution.
Expected by whom exactly? By you? By Joe Felsenstein?
Until you can define what fitness is, fitness is a meaningless concept in evolution, and therefore serves no role whatsoever-and it is why models of evolution are equally meaningless. Whoever decides what fitness is decides what the program searches for.
Flawed bases of logic equals flawed attempts at modeling, Tom. You will never overcome that.
Since there is no workable definition of fitness in evolution (Unless we are to accept the ridiculous Tom’s notion that what “someone” predicts will reproduce well is fitness, when the someone can’t be defined, thus it can be what anyone predicts) then no possible simulation can replicate something that can’t be defined.
So I guess Mung is not really blathering, rather he is pointing out the rather obvious, until you tell a computer simulation what is “good” it does nothing. You tell it what is good, and THAT is what it SEARCHES for! Simulations are just searches for what the programmer tells it to look for. But real world evolution (that imaginary concept that materialists have come up with) doesn’t know what to search for, so you can’t model it.
So you can’t tell when one person has more children than another? When one plant bears more seeds than another?
Wow, what dramatic and sweeping conclusions you have reached! Wow, what complete and utter nonsense!
phoodoo,
Yep. Weather programs are searches for what the weather will be. The programmer tells them to go and look for the weather.
Not until AFTER they have already had the children Joe, THAT is the whole point. Amazing you still don’t get that.
Oh, you want to talk about ANY simulations, like Disney movies perhaps, I see. I thought we were talking about EVOLUTION simulations silly me. If you want to discuss what Mickey Mouse does, feel free.
phoodoo,
Chortle. Funnily enough, evolutionary algorithms are used to evolve realistic movement in animations and games. Sorry, to search for realistic movement.
I wonder if their definition of realistic is whatever is realistic. Or better still for Tom and Joe, whatever they predict will be realistic before its realistic.
phoodoo,
Your paraphrasing skills are on a par with those of Erik and Joe G. Keep up the good work!
Funnily enough, the response of Marks, Dembski, and Ewert is that the evolutionary program wouldn’t have worked unless the programmer had provided it with a source of knowledge of physics. (What you’ve described is analogous to something they address, evolution of designs of bent-wire antennae, using a simulation package to predict how well antennae will perform when implemented according to the evolved designs.) This is an example of how they mistake the simulation for the simuland.
We can evaluate the claim that “information” is provided to the outcome by the programmer using the laws of physics by considering a simulation of some physical process, such as Brownian motion, erosion, formation of solar systems, or landslides. Simulations of any of these need to take the laws of physics properly into account.
Does that mean that dispersion of molecules, formation of valleys, having planets all circle the sun in the same direction in a plane, and rocks falling down a hill cannot be explained without proposing an input of “information” by a Designer?
If someone answers that the laws of physics do provide that information, then I’d say fine. Any explanation of evolution we offer does depend on the laws of physics, and if that’s all that is needed, then that is no evidence that further “information” needs to be added by a Designer once the Universe gets started.
Joe Felsenstein,
The Kindle version is handy. 😉 Section 5.8, “The Search for the Search,” opens (emphasis added):
And if we don’t view evolution as search, then what?
This is the sort of stuff that Zoners love to argue about. You’ve known many more scientists than I have. But I’ve known quite a few. Scientists, apart from cosmologists, are generally uninterested in speculations on how nature came to be what it is. They want to account for nature, the way it is. A fair number are interested in speculations on the future of humankind. I suppose I’m saying that there’s something quite different going on with folks who argue how nature came to be the way it is than with folks who dedicate their lives to accounting for how nature works.
This pertains to the passage I just quoted. When something seems like a miracle by present scientific account, you and I say, “There must be something we don’t understand.” The “faithful” exclaim, “Praise God!”
(I put “faithful” in scare quotes because religious arguments about science seem like extreme lack of faith to me. I’ve found that many Christians agree with me on this point.)
I say that if evolution is searching, then it is searching for a species better than one that includes YECs flying under cover of ID. But I am no eugenicist. I live in faith that whatever it is that evolution seeks, evolution will find.
Date: May 23, 2017
Source: University of Illinois College of Engineering
https://www.sciencedaily.com/releases/2017/05/170523152540.htm
Mung,
Parsing the language that a grad student (in physics — all of the authors are in physics) used in explaining the research to the author of a press release? The first three parts of the article are quite readable, and furthermore read rather differently than the press release does. The discussion section is somewhat more technical, but you can still glean plenty from it. For instance,
So what do you suppose the lead author actually meant to say to the author of the press release? Do you suspect he was trying to put things simply, with no idea that someone would read teleology into his words?
I recommend (to everyone, not just to you) making what sense you can of primary sources before recommending journalistic coverage. And never, ever give any credence to an advocacy journalist’s spin on a journalist’s rehash of a press release on a publication in the scientific literature.
Mung,
Mung, How dare you read the language of a science article. Of course the words don’t mean what you think they do, don’t you realize it’s a science article? You think that just because they quoted someone, that that is what the person really meant to say?
I want to remind you of the point of this OP. The point is Tom English doesn’t like Robert Marks, Winston Ewert and William Dembski. He doesn’t like that they do research, he doesn’t like that they write papers. THAT IS THE POINT OF THIS OP! Ok, do you get it?
Do you get the point, he doesn’t like them. Ok, fine fine, call it professional jealousy if you wish, call it Tom’s personal insecurity if you must. But let’s not distract the point of this OP with meaningless science articles, and facts, and details, for crying out loud. Tom doesn’t like someone, why can’t you get that?
Can we please stick to that for once? Otherwise Tom is going to have to start 7 more OP’s talking how much he doesn’t like someone. What science articles are going to save you then?
You shouldn’t believe anything you read in the News.
I already do that, its called being a skeptic. I think some skeptics should give it a try.
phoodoo,
The capital N in News is not gratuitous. And anything is not the same as everything.
[Seriously, you’d learn a lot about 0’Leary by clicking on the links she provided in her smear job, and comparing what you actually see to what she indicated you would see. She took full advantage of the leeway journalists have under U.S. law. If Barry Arrington had posted the same — he knows better — I’d have sued him.]
I am skeptical, I don’t believe anything you write. I think its all gratuitous!
It’s not SCIENCE, it’s ENGINEERING!
Tom is right. We are talking problem solving and performance. These are engineering problems. Computer Science is a complete misnomer. Computer Engineering is more like it!
Just call me a computer science engineer pioneer.
Like everyone else here at TSZ, I pull that shit out of my ass.
Pretend that I’m a complete IDiot.
They can be used as problem solving algorithms because they are actually designed to be problem solving algorithms, or they can be used as problem solving algorithms because they are not designed to be problem solving algorithms.
The most interesting question to me is whether some particular simulation of evolution is similar enough to the real thing to draw real-world conclusions from it.
Great! Evolutionary algorithms are employed by (designed by) engineers to solve problems.
I should have chimed in earlier. Evolutionary computation isn’t always used in search of solutions to problems. For instance, when an evolutionary algorithm is used for adaptive online control of a system that runs indefinitely, the objective is not to generate a solution to a problem — not in the ordinary sense of the term. But I don’t see the exceptions to what Mung is saying as a very big deal. He’s basically got the right idea.
Again, there are two components of a search, one of which generates a sample of possible solutions to a problem, and the other of which outputs the best solution it can find in the sample. The sampling component may simulate an evolutionary process, associating “fitnesses” with possible solutions. The solution-seeking component bears no relation to biological evolution. So an evolutionary search is a search that samples by simulating an evolutionary process. The simulation of an evolutionary process is not itself a search.
Biologists are most definitely not saying that evolutionary processes seek to maximize fitness. In various models of population genetics, including Wright-Fisher, the evolutionary process settles into statistical equilibrium (converges to an equilibrium distribution). Sooner or later, I’m going to post animations of evolutionary processes settling into equilibrium. When I tune the parameters of a particular model to minimize the expected wait for maximum fitness to occur (Marks et al. would say that I maximize the active information per mean query, or some such), maximum fitness occurs very rarely over the long term. There’s a tradeoff of factors, and the choice of evolutionary process in which maximum fitness occurs rapidly is also an evolutionary process in which maximum fitness occurs rarely.
I guess I’m trying to say that investigators of evolutionary computation are not saying, or should not be saying, that evolution seeks to maximize fitness, because the role of evolution in evolutionary search is to sample, not to search.
ETA: Animation that Mung has seen before. The orange process is the one that results from tuning the model to minimize the expected wait for maximum fitness. The blue process differs from the orange only in initialization — the population is filled with maximally fit individuals. It converges to the same equilibrium distribution on fitness that the orange process does. This is the strongest illustration I’ve managed to produce that evolution is not search.
Moved a comment to guano.
The comment can be found HERE
It was a trick question. One person doesn’t have any children. It takes two. 🙂
If someone could direct me to where Tom had some questions for me it would be much appreciated. Thank you in advance.
I am curious as to your response to my comments above. You presumably have a copy of the book, so here’s a question: Where do the authors explain why the measurements of (several forms of) active “information” in Chapter 6 establish that the modelers designed the evolutionary processes in order to generate solutions to problems? Thank you in advance (and not sarcastically — I genuinely want to know what you have to say).
Hi Tom,
I don’t think they’ve been shy at all in saying exactly which models they have been examining. Weasel, ev, Avida.
http://www.evoinfo.org/index/
I would think it rather obvious that each of those had problem solving in mind, but if that’s in dispute we could certainly look further.
Interesting question. I’ll take a closer look at Ch 6.
Happy to oblige. The questions were asked by me, not Tom.
They are in this comment, above.
So to understand Tom’s question in context we need to look at what the authors say about evolutionary models and active information.
Ah. My fault then. I thought you were referring to some questions Tom had asked. Thanks!