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
And if he’d instead written an equation to describe the system, he’d have been telling the equation which are fitter — right?
Perhaps you meant to say that he was telling the computer. Then I’d respond: if he’d written an equation to describe the system, he’d have been telling mathematics which are fitter — right?
Phoodoo will demand to know how the fitnesses got into the program. You might explain why we can learn about evolution from simulations in which the fitnesses are made up. (Ewert, if not phoodoo, very much needs the explanation. And he’s probably reading this.)
Many people who use the word genotype do not understand that geneticists classify organisms in a population according to (some of) their heritable traits. That is, it hasn’t registered on them that a genotype is in fact a type (class) of organism. An organism belongs to one and only one class (genotype), while a class may contain many organisms. Feel free to say this better than I have. I’m not one to believe that being a computer scientist grants me special insights into biology. I leave such self-mythologizing to people whose “faith” is severely threatened by the theory of evolution.
In the particular case of a windmill evolving GA, yes, the programmers have in some way decided which are fitter. They have decided that windmills that spin faster and generate more energy are fitter. The ability to spin faster and generate more energy (than the descendant windmill) comes from mutations. A numer of windmills are generated, and then randomly mutated. Then they’re “tested” to see how fast they can spin in a physics simulation. A small sample of the population, those that spin the fastest, are copied to the next generation for a new round of mutation.
A windmill evolving GA will not tell us that evolution can invent windmills, it will just tell us that if windmills exist, evolution can improve their function(and fitness).
Not all the windmills will be “fitter”. Some will mutate and become worse. Those are not among the ones that get copied to the next generation.
Neither the program, nor the programmers of the program, know beforehand which particular individual windmills will be fitter. That is up to the physics simulation to determine. But they DO know that the windmills that spin faster and generate more energy, are fitter.
So while they don’t know how exactly the windmill that will have evolved 20 or 100 iterations down the line looks, they do know it will be a windmill that spins much faster and generates much more energy. That’s the whole point of the program.
If the program did not assign higher fitness to a windmill spinning faster, then as Allan says the program would just do random sampling. And then there’d be no reason to expect the program to evolve progressively better windmills any more than it would make progressively worse windmills.
This is why I think the difference between Avida and so many other GAs is important. In Avida the program can be set up to not assign fitness, or reward with resources, any particular behavior. Yet there is still selection and fitness in Avida in direct analogy to the way it happens in real life. Here fitness differences are emergent, because the reproductive capability of organisms in avida isn’t imposed on the organism from “above” in the same way it is for the windmills. In Avida, fitness is measured, not assigned.
Windmills can’t reproduce themselves, so this is done in code. There are some lines of code, made my human programmers, that copy and mutate the windmills that spin the fastest.
But (afaik) in Avida, the reproctive ability of the organisms is part of the functional repertoire of the organism and tied to all the other functions it has. In Avida, the organisms really are organisms. Simulated organisms, that “divide” on their own. And this ability to divide is intricately tied to the other functions the organism has, such as metabolism and ability to gather resources.
And in Avida, new functions can evolve. They can’t in the windmill. The windmill just has blades that spin. And the number of plades and their shapes is what is mutated and then tested in a physic simulation. But in Avida, the organisms really do have something like cellular functions, and these can mutate and recombine in new ways that produce new functions. In this way, evolution in avida can produce novel functions that weren’t there to begin with. The same can’t happen in the windmill GA.
phoodoo,
Which program? Honestly, when you go off on one, you become barely coherent. I don’t write a different program for every fitness landscape. That can be caused at runtime.
In a particular implementation, if there is no differential in fitness, then there is no difference in fitness. I get the syllogism; I don’t know why it causes you to bust a blood vessel and break out the all-caps. The point of an equal-fitness model is to illustrate that one is not telling the program what the differential is. Same for the ‘real windmills’ implementation.
It is impossible for an individual or a genotype not to have a fitness. Zero is a fitness.
phoodoo,
After, before … it makes no difference. In a repeatedly copied population, after is before, and vice versa. The reddest of herrings.
Why should I be able to predict the fittest genotype in advance? I can’t predict which number will come up on a biased roulette wheel either. Or a straight one. But so friggin’ what?
phoodoo,
1 and 0 aren’t fitnesses, they are your binary code for ‘exists’ and ‘doesn’t exist’. ‘Exists’ and ‘doesn’t exist’ aren’t coded into ‘my program’ either. I do wonder if you’ve ever actually written one, and understand the relationship between code and data.
That should read ancestor windmill, not descendant.
I’m running an experiment. I’ve got a bunch of pea seeds and some blotting paper. I create a bunch of different conditions and want to see which ones cause the most growth. But … I can’t predict which ones will do best! Waaaah! I can’t do science!
because the designer must be able to predict it, so that evolution produces exactly what the designer wants: the pinnacle of evolution: phoodoo
After reading your paper, I have to agree with you.
😀
Mung,
Oh, ouch, I’m sure.
Haha, right.
Right now there are about 7.5 billion people on the planet with a fitness of 1. There are another 62.8 trillion with a fitness of 0. Or could be 886 trillion. Or 45 people. Or 7. We will never know for sure because they don’t exist, so its hard to give an accurate count.
phoodoo,
No there aren’t. Given the information that fitness is defined as number of offspring – when applied to a genotype, mean number of carriers’ offspring reproducing themselves – perhaps you could spot your own error. Or just keep failing to understand what fitness is, one of the two.
Fitness: the genetic contribution of an individual [or genotype] to the next generation’s gene pool relative to the average for the population, usually measured by the number of offspring or close kin that survive to reproductive age.
Allan Miller,
So everyone alive who hasn’t given birth to an offspring yet has a fitness of zero?
phoodoo,
Better to look at genotypes. You aren’t desperately interested in counting one individual’s eventual offspring, in an EA or in evolution. If a genotype renders its bearers sterile whenever it arises, then yes, it has a fitness of zero.
Of course its better for your argument to say look at genotypes, but I didn’t ask you about genotypes, I was discussing individuals. Because after all, every individual is a new genotype.
So according to your definition, that you provided, every individual on the planet has a fitness of zero, unless they decide to have kids, and stays alive long enough to see those kids become old enough to have their own kids. So the woman who has six kids under 12, she still has a fitness of zero.
Allan Miller,
And if the case of your hypothetical wind turbine EA, every individual has a fitness of zero, then 1, then perhaps zero again.
phoodoo,
It’s just standard practice to look at genotypes. Don’t shoot the messenger. If we started to look at every individual in a population, we’d soon get bored. [eta – the ‘genotype’ here is not the full genetic sequence of the individual, but their possession or otherwise of a particular allele]
The fitness of an individual is not crystallised until all its offspring have reached reproductive maturity. Just like, until an individual reaches full height, you don’t know how big they are going to be.
phoodoo,
More gibber. The fitnesses of the individuals is the number of times each is copied before being lost. For some, that will be zero, for others 1, for others … other numbers are available.
Allan Miller,
Don’t shoot the messenger Allan, I used YOUR definition. So all you have done is confirmed that what I said is exactly correct.
I fully understand why you would want to run to a genotype now, like I just said.
The fitness of an individual is 0, unless they decide to have kids, and those kids live long enough to give them a fitness. The woman with the 6 kids remains a zero, until she no longer is.
So back to our EA, the fitness of any of the types is either 0 or 1.
You constantly complain about how creationist just refuse to see the brilliant logic of the materialist, and yet here you are now, being shown that what I said was exactly right, and your defense is not to admit it, but to say, “Well, we scientists, you know, we prefer to look at this, we get bored looking at that…”
phoodoo,
No. No it isn’t. Even if you want to talk about individuals, it isn’t. Dance your little dance of victory if you wish, but you are Just Not Getting It.
Telling everyone else how they just don’t get it is your standard operating mode Allan. Admitting you are wrong is not your strong suit.
Don’t bust any blood vessels.
phoodoo,
If, having been told that fitness equates to number of offspring, you insist it can only have the values 0 or 1, and then I say you don’t get it … do you get it now? I’m sure I’ve got more than 1. Last time I looked. But if I’m wrong, I’m wrong and happy to admit it. Just a sec, I’ll ask the missus.
Allan Miller,
No no, Allan, try reading again. I said in your wind turbine example, all individuals are 0 or 1.
We could go on and on Allan, you would NEVER admit you are wrong. I guess its not in your DNA. I guess its in your genotype. It must be a syndrome. Contrarianism on steroids…Why are you getting so red-faced?
Phoodoo
Please resist the urge with the “have-you-stopped-beating-your-wife” routine. As you are a minority member, I give you some leeway and I request you not to abuse it..
In calculating the expected fitness of a genotype (or estimating it empirically), it is best to start each generation at the newborn stage. Then the fitness is the probability of survival to adulthood, times the expected number of newborn offspring produced by survivors. (Actually, half the expected number in the case of diploids).
Otherwise we can’t keep track of the effects of genotypes, as the genotypes of offspring are often different from those of their parents (see Mendel, Gregor, 1864).
In more complex life cycles such as overlapping generations, we have to use quantities like the intrinsic rates of natural increase (the “Malthusian parameter”) which are calculated using age-specific birth and death rates. Bringing those into our discussion here seems, fortunately, unnecessary.
phoodoo,
OK, [scroll, scroll] … here’s what you said:
Which it isn’t, even accepting that the flip between individual and ‘type’ might be accidental. A given string in the population – an individual – could be copied 0, 1, 2, 3, 4, 5 […] times before being removed. Its fitness is that number, not just 0 or 1. If you’re really talking about type (genotype), the number can be fractional, being the mean fitness of carriers of that allele. Again, 0 and 1 are not the only possibilities.
Anyway, your attempts to make fitness vague or post hoc don’t really help your contention that a programmer has to tell the program what the fitnesses are.
In evolutionary computation, we simulate an evolutionary process in order to sample the space of possible solutions to a problem. You have substituted a description of how the simulator works for a description of what occurs in the simulated process. This is what Marks et al. do, and I have to tell you (excuse my bluntness) that it’s a severe error.
The process of simulating an evolutionary process is no more an evolutionary process than the process of simulating a spinning windmill is a spinning windmill.
The simulation of physics is a component of a simulation of evolution.
(Forgive me for not writing more. I should write an OP on the matter. However, I’m losing interest in Marks et al., now that it’s clear just how weak their claims are.)
We cannot simulate physics without giving physics a procedural description. We cannot simulate differential reproduction without giving differential reproduction a procedural description.
Tom English,
That is a slightly different issue, as I read it. We are not telling the process how many copies of a genotype to generate. We can cause differential reproduction of genotypes by the way we implement selection. But we could do it (as in my ‘real windmill’ example) completely outside of the program.
I’m prepared to agree to disagree on this, because I don’t think much hinges on it, but I think processing of genotypes in an EA is evolution.
I don’t think I understand what the difference is.
You will have to say more here if I’m to understand the nature of the error I have made.
I agree. But are you telling me this because you got the impression from something I wrote that I thought otherwise?
The simulation of physics is a component of a simulation of evolution.
I’d rather say that simulations of physics make the simulationf of evolution more realistic.
I can think of simulations of evolution with no physics simulations at all. Like Dawkins WEASEL. And other simulations of evolution that do, like BoxCar2D.
I think it makes more sense to say they have an unmeasured fitness.
Joe mentioned the concept of expected number of offspring. You can only expect something if you’ve got historical data of some sort that predicts the thing you now expect. So it only makes sense to speak of their fitness either as a matter of prediction (we predict they will have X number of offspring because they carry these particular alleles, which we have historically measured to be associated with some N level of reproductive success), or after they die (at which point we know how many offspring they managed to produce).
In my experience, Allan has a habit of telling people who really don’t get it, that they don’t. I think he hopes, as I do, that this will motivate them to try harder to get it, rather than feel like they have to say something to save face or to take digs at “materialists”.
In this particular case, no, it really is you who doesn’t get it.
Rumraket,
Oh no! You fell into phoodoo’s trap! “Who does the expecting?”. “You can’t predict … “. I’d help you out, but I might fall in meself, and then we’d both be stuck.
That’s the same thing. lol
Rumraket,
Good catch, I should read the insults more carefully!
Joe Felsenstein,
There’s that diplocentricity again! One could start from the gamete, rather than the diploid genotype. How much use it would be to do so, I don’t know.
LoL! Until you take the actual measurement, how do you know how many offspring they have?
Because as I outlined previously in this thread, that’s the question that phoodoo asked you, which you then pretended to answer.
And Tom accuses me of a non-sequitur. The irony.
I want to know how you measured those fitnesses. It’s looking to me like they have those fitnesses by some axiom, not by some measurement. Yet your claim is that fitness can be measured.
You’re saying that if you already know the fitnesses you can deduce something from them. The question was how did you arrive at the fitnesses.
How? Define the function
get_fitness.I give Rumraket credit for understanding that fitness is relative. It of course follows that counting number of offspring does not provide “the fitness.”
Mung,
It does for an individual. Which, as phoodoo has made clear, is what he wants to focus on. For some reason.
All you’ve done is replace one term with a different term. Define the function
measure_reproductive_success. So far we’ve found out that get_fitness is just an alias. It’s like pulling teeth around here.Why not write the code and show us? Should be a simple thing for you, dazz.
Who said it’s an issue?
Who said it’s circular?
Bullshit. it would have been better had the connection been kept from evolutionists because then it would have prevented the abuse of one domain by another that is completely unrelated.
John Holland wasn’t an evolutionary biologist.
John Henry Holland was an American scientist and Professor of psychology and Professor of electrical engineering and computer science at the University of Michigan, Ann Arbor. He was a pioneer in what became known as genetic algorithms
QED