A More Realistic Computer Simulation of Biological Evolution
| June 9, 2009 | Posted by GilDodgen under Intelligent Design |
In another thread a fellow who goes by Legendary made some rather derisive comments about a suggestion I once made, concerning making computer programs that purport to model biological evolution more realistic. The suggestion was half serious and half tongue-in-cheek, since it would be impractical.
My argument was as follows: Computer programs that purport to model biological evolution invariably isolate the effects of “mutations” to only those aspects of the “organism” that have a chance of helping the organism approach the desired goal (EQU in the case of Avida, for example). But this ignores an extremely important aspect of modeling living systems.
Random mutations, if they are truly random, will affect, and potentially damage, any aspect of the organism, including its ability to survive and reproduce. The computer program, OS, and hardware represent the features of the simulation that keep the organism alive and allow it to reproduce, but this is artificially isolated from the effects of mutations.
Thus, a realistic simulation would allow the program, OS, and hardware to be affected in a random fashion, just as a real organism’s ability to survive and reproduce would be affected randomly by mutational interference. A mutation might cause an enzyme to malfunction and the organism would suffer an early demise, or it might be rendered sterile, and the beneficial mutations would never be passed on.
Of course, this would not be practical, and each “organism” would require its own computer, but the point should be clear: A simulation can’t just arbitrarily ignore aspects of the reality it purports to simulate, because taking them into account would be likely to result in an undesirable outcome.
As a footnote, I highly recommend reading Eric Anderson’s piece on Avida here.
55 Responses to A More Realistic Computer Simulation of Biological Evolution
Leave a Reply
You must be logged in to post a comment.
Mr. Nakashima
An improvement need not be innovative to qualify for a patent. I looked over several of the award winners and saw just that – improvements, not inventions. It confirms what you had said earlier,
These are designed constructs that employ technology to do some heavy lifting and search out a number of possible solutions for the ideal one. They love referring to “evolution,” but their simulations never do what evolution claims to – invent something new.
Nakashima-san,
Outstanding comments at 26. There are now many engineering problems that humans know how to make amenable to solution by evolutionary computation, but do not know how to solve analytically. The two types of knowledge are fundamentally different in kind.
Humans indeed bias and constrain the evolutionary search for good solutions to engineering problems. The prior knowledge we use to do so is often generic and, in a sense that is intuitively clear but difficult to formalize, inexpensive to obtain. Good solutions are typically high in value relative to the cost of running evolutionary algorithms on computers for long periods of time.
For example, I know very little about the dynamics of solar weather, yet I applied an evolutionary algorithm to obtain what was, at the time, by far the best predictor of annual sunspots counts. My algorithm selection reflected my knowledge of certain modeling principles, my belief that the sequence of counts was chaotic, my belief that a class of models was good for chaotic time series prediction, and my knowledge of what works and what does not in evolution of models in that class. After consuming several weeks of computer time in execution of the algorithm, I had an outstanding predictor of sunspots counts. And no person knows, or is likely to figure out, what the model, a combination of more than 20 thousand smaller models, “knows” about sunspots. The information I added to the evolutionary search is of a type that I can use, with slight and rapid adaptation, for many problems. The information that came out of the search is of a type I would never have obtained with my unaided intellect.
The distinction in kind and value of the information used in algorithm selection and the information obtained through algorithm application is missing from all information-theoretic analyses of search. Not all bits are created equal. There is a sort of “free lunch” when valuable information is purchased by supplying cheap information to a cheap computational process.
That was Nakashima at 28, not 26.
Mr ScottAndrews,
They love referring to “evolution,” but their simulations never do what evolution claims to – invent something new.
Evolution claims to change allele frquencies over time. If you don’t think what is happening in EC is invention, please explain.
Nakashima:
ASnd that is what I meant by equivocation and evolution
Mr Joseph,
I am being the opposite of equivocal. I am being explicit about which of several possible meanings I am using. Not that it matters in this case, as none of the definitions you refer to in your blog entry match the claim that “evolution claims to invent”.
Mr. Nakashima:
I’m sure there are hairs to split over the meaning of the word “invention.”
A simulation based on mutation and selection might improve upon an antenna, but it won’t figure out how to manipulate and receive radio waves to send information.
Perhaps that’s a way to falsify ID. Generate irreducible complexity or CSI in a simulation of random changes and selection. I’d be surprised and disappointed if no one is already trying.
Mr ScottAndrews,
Perhaps that’s a way to falsify ID. Generate irreducible complexity or CSI in a simulation of random changes and selection.
That is not how it is going to happen, I think. Generating CSI is just attributed to the active information content of the EC system. While active information has been defined, we don’t know how to account for it, how to distinguish the relative importance of the fitness function, the clock, the history, and the random number generator.
Dave W.: …the researchers that Eric Anderson freely insulted only purported to be working on simulating natural selection and genetic mutations, not on the entirety of “biological evolution.”
Oh really? Check again. The “researchers” begin their paper with obligatory genuflecting to Darwin (“one of the greatest scientific achievements of all time”), and the entire purpose of the exercise was to demonstrate that the Darwinian mechanism in biology can produce the irreducibly complex systems identified and elucidated by Behe.
Furthermore, Anderson did not insult anyone; he simply pointed out that the entire project was contrived from the beginning to produce a desired result, and assumed in advance what it attempted to demonstrate.
[32] T M English:
In using your EC to find a predictor for sunspot numbers, did you, or did you not, use the actual observations of sunspots in the past in evaluating the efficiency of your EC? I suspect your answer is that you did, indeed, use such observations. That then became your ‘target’, and you simply waited until the program reached the ‘target’. But evolutionists insist that evolution is ‘undirected’, i.e., that evolution is NOT looking for any particular ‘target’ beforehand. Thus, while you might call your algortithm ‘evolutionary’, it does not, in fact, imitate evolutionary processes. It is no more than a blind search through a search space that is amenable to such a search. Your wrote, after all, that you “consumed” several weeks of computer time in finding the desired predictor.
Mr PaV,
Evolution happens whether the selection is guided or unguided. I think Darwin in OoS spends some time talking about breeding animals for this reason.
In this particular case, using the data as part of the fitness function does ot make the data set a target, in the direct sense that most people have from discusing WEASEL. The organisms in the population are models, sets of parameters or actual programs and parameters. In some paradigms, some of the data might be available to a model, while the rest is held out to test fitness. In other paradigms the models never see any of the data.
At no point is the search blind. It always has some of the history of previous samples available.
Mr Dodgen:
For one thing, the researchers made a true statement: Darwin’s theory “is widely regarded as one of the greatest scientific achievements of all time.” If you can present evidence that it is not “widely” regarded as such, regardless of your own feelings about it, please do so.
Second, your mimicking of Anderson’s ad hominem attack on the researchers is a waste of my time and yours.
Exactly correct, which is why I stated (also correctly) that they did not purport to be simulating all of “biological evolution.” If you think that “biological evolution” is synonymous with “random mutation and natural selection,” then you need to learn more biology.
You invoked the same ad hominem attack as Anderson, but Anderson went further, suggesting that the reason the paper was published in Nature was because of the authors’ “genuflecting” to Darwin. This sort of argumentation is massively insulting, because it is clear that Anderson must have thought his readers to be morons for him to consider that a good point to make. Anderson further compounds the insults by asserting motives for the authors for which there is no evidence, and by mocking the real hypothesis testing that was being accomplished.
But really, the greatest insult provided by Anderson to all his readers (you included, Mr. Dodgen) is that he never criticizes the aspects of evolution which were actually being tested, but then in the conclusion he flatly claims that Avida failed those tests. He must think us all very stupid, indeed, to think that anyone would be fooled by such an empty argument.
Mr. Nakashima,
“Here we have to distinguish between evolution as an abstract concept and the simulation of actual biology. In the spirit of some other comments on other recent threads, a GA is not simulating evolution, it _is_ evolution.
I’ll assuming GA means genome analysis? Why would evo biologists be developing a type of evolution that has nothing to do with biology? If they were interested in mirroring what is known of biology to gain understanding, wouldn’t they be striving for a program that allows for no macro evolution, and only small changes to existing structures, as well as larger micro evolution but only in brackets, while at the same time their computer genome would start out ordered and move towards chaos? Also wouldn’t they need to input for reducing phyla and an inverted tree of life?
IE. aren’t Neodarwinian evo biologists designing their simulations backwards?
The bottom line is that the Darwinian mechanism — which is based on complete and utter 19th-century ignorance of the underlying information-based nature of biological systems — is perfectly inadequate to explain what is observed, beyond the trivial and obvious.
It’s equivalent to trying to turn lead into gold through chemistry. It doesn’t work that way. It is a categorically erroneous explanatory approach.
Neo-Darwinists are latter-day alchemists who have been mysteriously immunized against following the evidence where it leads. What an embarrassment, when the science is screaming design from every corner.
The bottom line, Mr. Dodgen, is that neither you nor Mr. Andserson have critiqued the mechanisms under test in the simulations. You instead fault the simulations for failing to address aspects of evolution and/or biology that they were never intended to address. And when that point is made, you dismiss what you fail to critique as inadequate, trivial or obvious.
Mr. Behe stated that IC was impossible for mutation and selection to accomplish. The EQU test was set up to test just those aspects, and when it succeeded, Mr. Anderson argued not one bit against either the mutation or selection processes, but instead invented a raft of transparent objections which all missed the point, and then had the gall to claim that Avida failed. Or that it accomplished something trivial which Mr. Behe claimed to be impossible. Yes, according to Anderson, what Avida did was trivial and successful and impossible, all at the same time.
Hi Mr Lamarck,
Sorry I was unclear, GA means Genetic Algorithms. The field is based on taking the basic insights of evolution and applying them to optimization problems and other uses in computers.
Dr John Holland is one of the pioneers of the field, and in his book Adaptation in Natural and Artificial Systems he proves what is called the Schema Theorem. The Schema Theorem explains why, in the simplified population genetics of a computer simulation, the fitness of the population will increase over time.
Mr Dodgen,
On the contrary, it is a fairly sophisticated hiding of implementation details. Evolution works, no matter if you implement selection with natural or artificial selection. Evolution works, no matter if you implement variation with epigenetics or DNA mutations.
In your last sentence, I think you might mean “when the evidence is screaming design” not “the science”.
Nakashima:
You’ve compared the success of natural selection to that of artificial selection. What would you cite as the single most impressive product of evolution via artificial selection?
You’re obviously reading into things that aren’t there. Go back and read what he wrote:
All he said was that it works; he said nothing about the size or amount of the results of success.
RDK @48:
I never thought or said that Nakashima overstated his claim.
If I were to split hairs, I might point out this statement: statement
Evolution is change, and we’ve known that such change occurs for thousands of years. To say that it works, regardless of the mechanism, does not require any science. (The sun rises, whether it’s the earth rotating or it gets pulled by a chariot, or something else.)
Do we know which mechanism caused generations of pawed mammals to slowly develop hoofs, or do we name several possibilities and say that it must have been one of them?
PaV (40):
The key Darwinian principle is that variety, heredity, and fecundity yield demography. Many evolutionary algorithms more or less reduce demography to fitness, but the one I applied does not. See Nonlinear Combination of Neural Networks from a Diverse and Evolving Population [warning: PDF renders slowly] for a concise description of the actual approach.
What is relevant to your comment is the success in prediction of a chaotic time series generated by a process that (pseudo-)randomly changes from one unobservable state to another. The dynamics in the two states are radically different, and it is quite unlikely that any one model would capture the dynamics of both states. I would argue that there is, at least as a practical matter, no clear-cut target for the evolutionary algorithm. Pay special attention to the concept of niche implicit in the ranking of models. A model with relatively high error can be selected as a parent because its errors are dissimilar from those of other models. The evolutionary algorithm is not “trying” to make all of the models in the population fit the data as well as possible.
The missing link.
Mr ScottAndrews,
You’ve compared the success of natural selection to that of artificial selection. What would you cite as the single most impressive product of evolution via artificial selection?
Corn.
It seemed like a great question at the time. But it really wasn’t.
Mr ScottAndrews,
Don’t feel bad. Some of the best proofs of God involve picnics, sweet corn, fresh butter, salt and pepper!