Thread: Are Evolutionary Algorithms Realistic?

I often read that Evolutionary Algorithms demonstrate the capability of mutation and selection to generate new molecular level function. When I look at them in detail the working algorithms seem to have imbeded within the code either a target or background information about the search space that is used to alow the algorithm to succeed where it otherwise would not. Evolutionary theory claimes that natural selection has no goal or target and is blind to the environment or search space it occupies.

It seems unrealistic to cite these models if they don't conform to the tenets of the theory. In another thread one poster cited these as validation of the cabability of Darwinian processes to generate new functional systems. I don't see how they are unless the models are modified to fit the tenents of the theory.

Originally Posted by cypress

I often read that Evolutionary Algorithms demonstrate the capability of mutation and selection to generate new molecular level function.

do you happen to have any examples of what you mean by "evolutionary algorithms" ?

Originally Posted by cypress

Evolutionary theory claimes that natural selection has no goal or target and is blind to the environment or search space it occupies

That's kind of silly. How are you distinguishing selection criteria from this "target" notion of yours, and how do you envision a living being, with selection criteria operating on its offspring, as having no embodied information about its environment?

A model is never realistic. It is a model.

The function of a model is that it has explanatory value of a realistic situation, where the model has more value when the explanatory value is greater.

Everybody is quite aware that not all parameters are incorporated in a model. If this wouldn't be the case, it wouldn't be a model.

The value lies in a simplification of reality through a model which has relevant explanatory power.

So your opening question is stupid.

So Cypress, a big fan of trueorigin.org are we?

Originally Posted by iceaura

Originally Posted by cypress

Evolutionary theory claimes that natural selection has no goal or target and is blind to the environment or search space it occupies

That's kind of silly. How are you distinguishing selection criteria from this "target" notion of yours, and how do you envision a living being, with selection criteria operating on its offspring, as having no embodied information about its environment?

I think you understanding the difficulty of the problem that faces a modeler. Selection criteria that does not display any bias to a particular solution and does not artificially reward convergence when advantage is not apparent, would not be considered a targeted search. I don't think evolutionary processes require that the organisms lack any information about their environment.

Originally Posted by marnixR

Originally Posted by cypress

I often read that Evolutionary Algorithms demonstrate the capability of mutation and selection to generate new molecular level function.

do you happen to have any examples of what you mean by "evolutionary algorithms" ?

Here is an example of one I looked at. They call it EV.

T.D. Schneider, 2000, “Evolution of Biological Information,” Nucleic Acids Research, Vol. 28, No 14, pp.2794-2799

Originally Posted by spuriousmonkey

A model is never realistic. It is a model.

So you take the position that those who claim evolutionary algorithms demonstrate the capability of evolutionary processes are simply wrong. OK, I can accept that.

Originally Posted by KALSTER

So Cypress, a big fan of trueorigin.org are we?

No, I don't know of the site. Forgive me for not looking at it.

Ah... you must be using creationwiki, then.

Typical inow, and KALSTER. Avoid the substance and attack credibility instead.

Originally Posted by cypress

Originally Posted by spuriousmonkey

A model is never realistic. It is a model.

So you take the position that those who claim evolutionary algorithms demonstrate the capability of evolutionary processes are simply wrong. OK, I can accept that.

I don't take the position at all. I just take the position that you made a question without any merit to it.

Originally Posted by cypress

Typical inow, and KALSTER. Avoid the substance and attack credibility instead.

No substance and no credibility, I'm afraid. Your egregious misrepresentation of the post above regarding the limitations of models ('Oh so you're saying they're wrong') is simply further evidence that you're trolling. See my longer post, in the other thread you hijacked for your hobby-horse, for more detail as to why you no longer deserve substantive responses but meta-posts merely about your trolling style. You've brought it on yourself.

Originally Posted by sunshinewarrior

Originally Posted by cypress

Typical inow, and KALSTER. Avoid the substance and attack credibility instead.

No substance and no credibility, I'm afraid. Your egregious misrepresentation of the post above regarding the limitations of models ('Oh so you're saying they're wrong') is simply further evidence that you're trolling. See my longer post, in the other thread you hijacked for your hobby-horse, for more detail as to why you no longer deserve substantive responses but meta-posts merely about your trolling style. You've brought it on yourself.

I read spurious to be saying that models can't possibly be realistic and therefore it is pointless to cite them as evidence for a theory, and that it was stupid to even discuss it.

I don't mind if you don't post substantive responses. I do find it odd that you would blame me for your poor behavior.

Honestly, how can you be so thick? A model is model. A representation of reality that necessarily is limited to a reduced amount of parameters that determine reality.

This restriction is actually where the power lies in a model. You can't calculate reality. You can't isolate reality as you can with a model.

The model is a powerful tool.

The choice how to model a real process is however of utmost importance. And that can be under discussion.

And it is in the literature. And it isn't a problem, because science isn't the bible, but instead a means to discover more about the principles that underlie our world. And by putting our heads together, by making mistakes, by correcting false assumptions, by scrutinizing each other's work we make headway in understanding the world.

And no religious belief can substitute for this fabulous process because religious doctrine is determined a a small group of limited people with nothing more in mind then powergames.

Yes spurious you said that earlier and I agreed. I recognize the utility of models to simulate known conditions and processes. I use physical and statistical and empirical models ever single day in my job keep tens of billions of dollars worth of process equipment running optimally.

However several posters have attempted to argue that these models are evidence that evolutionary processes can accomplish even small steps that have not been observed. this is a very different use of modeling. Whereas models are typically used to simulate conditions known to occur, these people want to use a model to simulate process to produce unverified effects.

I am surprised you all continuously want to bring religion into this discussion unless you mean those who treat science as there religion. I certainly have not mentioned religion here.

Originally Posted by iceaura

Originally Posted by cypress

Evolutionary theory claimes that natural selection has no goal or target and is blind to the environment or search space it occupies

That's kind of silly. How are you distinguishing selection criteria from this "target" notion of yours, and how do you envision a living being, with selection criteria operating on its offspring, as having no embodied information about its environment?

I think this may be the fundamental issue with use of these search models. Most use some variation of a hamming model to compare fitness to a target fitness. Many use a oracle routine to determine fitness but they inadvertently insert information about the search space which can artificially speed up convergence to the target patterns.

If a target is used in a hamming model to determine selection, then the model will converge on the target, whether or not the target was explicit or not. I don't think many of us believe that fitness involves targets. How about a system that uses multiple targets and a more realistic criteria for predicting fitness? Would this kind of algorithm more closely model the theory?

Originally Posted by cypress

That's kind of silly. How are you distinguishing selection criteria from this "target" notion of yours, and how do you envision a living being, with selection criteria operating on its offspring, as having no embodied information about its environment?


I think this may be the fundamental issue with use of these search models.

It's an issue with your criticism of them, first. What are you talking about, when you use the word "target"? Especially, when you claim these "targets" are hidden or implicit?

Originally Posted by cypress

I don't think many of us believe that fitness involves targets.

Most of us believe that selection involves criteria, though. How do you tell the difference, in these models, between an implicit target and the likely products of selection criteria?

Are you labeling the common hydrodynamic shapes of pelagic fish the "targets" of the Darwinian selection process?

I think they're technically called genetic algorithms. One may flaw they are constrained by their programming (whereas organisms can mutate there programming not just some value). Also they are not just models for evolution, you can use them to find an approximation to a formula that you have very little hope of figuring out on your own. They have many practical applications.

However several posters have attempted to argue that these models are evidence that evolutionary processes can accomplish even small steps that have not been observed. this is a very different use of modeling. Whereas models are typically used to simulate conditions known to occur, these people want to use a model to simulate process to produce unverified effects.

Depends on the model. Some can indeed accomplish small steps. Why should it not be possible?

Originally Posted by spuriousmonkey

However several posters have attempted to argue that these models are evidence that evolutionary processes can accomplish even small steps that have not been observed. this is a very different use of modeling. Whereas models are typically used to simulate conditions known to occur, these people want to use a model to simulate process to produce unverified effects.

Depends on the model. Some can indeed accomplish small steps. Why should it not be possible?

Indeed models deliberately designed to demonstrate a concept will accomplish what it was designed to to. I would also think it possible to create a more realistic model of evolutionary processes. Whether a realistic model will accomplish anymore than what has been observed by experimental biology is an open question.

Originally Posted by iceaura

Originally Posted by cypress

I think this may be the fundamental issue with use of these search models.

It's an issue with your criticism of them, first. What are you talking about, when you use the word "target"? Especially, when you claim these "targets" are hidden or implicit?

A targeted search compares each search attempt against a target configuration and generally assigns fitness relative to the distance to the target. Dawkins used a simple targeted search in his book The Blind Watchmaker to illustrate the concept of natural selection as a fitness search. He started with a random series of letters and "mutated them to the target phrase, "Me thinks it is like a weasel".

In many evolutionary algorithms also called genetic algorithms use of a target is not as explicit as in the weasel example. Often the targets are imbedded in the fitness function.

Originally Posted by cypress

I don't think many of us believe that fitness involves targets.

Most of us believe that selection involves criteria, though. How do you tell the difference, in these models, between an implicit target and the likely products of selection criteria?

One way would be to look at how the selection criteria determines fitness. If it compares the relative fitness of the permutations to each other without reference to any absolute standard of fitness then it may not contain an implicit target. If the criteria uses fixed standards of fitness to compare each permutation, those are targets. There are numerous other methods that inject goals and artificial help (like extra information about the search landscape) into the algorithms to cause them to be more successful.

Are you labeling the common hydrodynamic shapes of pelagic fish the "targets" of the Darwinian selection process?

No I don't think so, physics dictates that certain shapes provide advantages over others. Ao long as fitness is measured relative to each other, I would generally agree it fits the theory.

I often read that Evolutionary Algorithms demonstrate the capability of mutation and selection to generate new molecular level function. When I look at them in detail the working algorithms seem to have imbeded within the code either a target or background information about the search space that is used to alow the algorithm to succeed where it otherwise would not.

The analogues of these are the existing genetic framework of an organism (background information) and the adaptation that would allow the organism to be able to survive the particular environmental challenge long enough to be able to reproduce (target). Agreed?

Evolutionary theory claimes that natural selection has no goal or target and is blind to the environment or search space it occupies.

I don't think it does. Mutation is blind, but how the environment selects for these mutations is not blind.

Originally Posted by KALSTER

I often read that Evolutionary Algorithms demonstrate the capability of mutation and selection to generate new molecular level function. When I look at them in detail the working algorithms seem to have imbeded within the code either a target or background information about the search space that is used to alow the algorithm to succeed where it otherwise would not.

The analogues of these are the existing genetic framework of an organism (background information) and the adaptation that would allow the organism to be able to survive the particular environmental challenge long enough to be able to reproduce (target). Agreed?

Interesting. If background information were extracted from and stored back into the configuration of the virtual organism that would be different from encoding background information into the selection oracle. The challenge then would be how to generate the background information in the first place.

I don't understand your target analogy though. If you are suggesting that natural selection and adaptation makes a comparison to existing fitness in a relative sense as opposed to a target which is absolute, then I think we are on the same page. Is this what you mean?

Evolutionary theory claimes that natural selection has no goal or target and is blind to the environment or search space it occupies.

I don't think it does. Mutation is blind, but how the environment selects for these mutations is not blind.

Explain this more, when I hear people talk of selection being blind I understand it to mean that it has no goal in mind and that it is not seeking anything. I don't take that to mean it doesn't favor the outcomes that in the present allows the species to be more competitive. If all you mean by not blind is not random, then I don't see that as very informative. Are we saying the same thing?

Evolution does not have "goals," it has "outcomes."

Looks like the consensus here is that Genetic or Evolutionary models are not realistic. While they might make good models to simulate postulated behavior, they are inappropriate to cite as evidence for the capability of mutation and selection

There is no such consensus, Cypress.

I don't see anbody providing a scientifically valid argument that GA's provide evidence for capability of evolutionary processes to generate observed diversity. Do you?

Originally Posted by cypress

One way would be to look at how the selection criteria determines fitness. If it compares the relative fitness of the permutations to each other without reference to any absolute standard of fitness then it may not contain an implicit target. If the criteria uses fixed standards of fitness to compare each permutation, those are targets.

So you do claim that Darwinian evolution is targeted, because the absolute standard of reproductive success is implicit in the "selection" phase.

Any realistic model of Darwinian evolution would then incorporate the same target, implicit in the selection process.

In addition, real situations are limited in their circumstances - the open ocean is an environment of a certain kind - so that contingent targets would also be present (implicitly) in the selection process. The limited number of "fitter" hydrodynamic shapes for pelagic fish, for example, would be implicit targets of the selection process.

Any evidence of convergent evolution supplies us with examples of implicit targets in the evolutionary process involved.

And all models of Darwinian evolution would perforce include them as well, to the extent they were reasonably representative of the natural process.

Originally Posted by iceaura

Originally Posted by cypress

One way would be to look at how the selection criteria determines fitness. If it compares the relative fitness of the permutations to each other without reference to any absolute standard of fitness then it may not contain an implicit target. If the criteria uses fixed standards of fitness to compare each permutation, those are targets.

So you do claim that Darwinian evolution is targeted, because the absolute standard of reproductive success is implicit in the "selection" phase.

Any realistic model of Darwinian evolution would then incorporate the same target, implicit in the selection process.

I don't think so. There shouldn't be any absolute measure of fitness. The fitness function should make only relative comparisons. Relative comparisons would not require any target at all.

In addition, real situations are limited in their circumstances - the open ocean is an environment of a certain kind - so that contingent targets would also be present (implicitly) in the selection process. The limited number of "fitter" hydrodynamic shapes for pelagic fish, for example, would be implicit targets of the selection process.

The fitness function would not seem to operate on targets even in this situation. One only needs to compare the variations with reproduction capability relative to each other to determine gene distributions by generation.

The GA's don't do this. Instead they compare each variation to a target configuration.

Any evidence of convergent evolution supplies us with examples of implicit targets in the evolutionary process involved.

Trouble is that convergent evolution is a just so story invented to allow phylogenic bushes and webs to be reshaped into trees. If we were to take your inferences about convergent evolution as true then these would be explicit targets which would indicate the process of novel gene creation is likely a goal directed process.

And all models of Darwinian evolution would perforce include them as well, to the extent they were reasonably representative of the natural process.

Are you suggesting evolution is a goal driven process?

Originally Posted by cypress

I don't think so. There shouldn't be any absolute measure of fitness.

You'll have to complain to Darwin about that. He put reproductive success in his theory, and there it remains.

Originally Posted by cypress

The fitness function should make only relative comparisons. Relative comparisons would not require any target at all.

Perhaps you could give us an example of selection via a "relative comparison" with no implicit target. I'm drawing a blank.

Originally Posted by cypress

Trouble is that convergent evolution is a just so story invented to allow phylogenic bushes and webs to be reshaped into trees.

? Are you saying that, for example, the similarly shaped pelagic vertebrates of the sea belong in a "phylogenic" bush or web, rather than distributed among various trees of "sharks", "tuna", "dolphins", and so forth?

And that "convergent evolution" was invented to prevent that classification scheme?

Originally Posted by iceaura

Originally Posted by cypress

I don't think so. There shouldn't be any absolute measure of fitness.

You'll have to complain to Darwin about that. He put reproductive success in his theory, and there it remains.

One of us is confused. Darwin never claimed that reproductive success is a targeted fitness search. He claimed it was relative. I have not heard anybody claim the theory is based on anything but relative successes.

Originally Posted by cypress

The fitness function should make only relative comparisons. Relative comparisons would not require any target at all.

Perhaps you could give us an example of selection via a "relative comparison" with no implicit target. I'm drawing a blank.

So too must all the writers of GA's. Everyone seems to want to use targets. Yet evolutionary biologists claim Darwinian Evolution is blind and has no target.

Originally Posted by cypress

Trouble is that convergent evolution is a just so story invented to allow phylogenic bushes and webs to be reshaped into trees.

? Are you saying that, for example, the similarly shaped pelagic vertebrates of the sea belong in a "phylogenic" bush or web, rather than distributed among various trees?

And that "convergent evolution" was invented to prevent that classification scheme?

No I don't think so. I am refering to molecular (genetic) similarities. Are body shapes gene based or developmental or both? What is the molecular process that is responsible for detrmining body shape? How are those traits manifest?

Originally Posted by cypress

I don't see anbody providing a scientifically valid argument that GA's provide evidence for capability of evolutionary processes to generate observed diversity. Do you?

There are thousands of peer reviewed papers out there that do that.

Here is one I like.

Salazar-Ciudad, I., & Jernvall, J. (2002). A gene network model accounting for development and evolution of mammalian teeth. Proceedings of the National Academy of Sciences, USA. 99: 8116–8120.

feel free to discuss its merits and its pitfalls with me.

Originally Posted by cypress

One of us is confused. Darwin never claimed that reproductive success is a targeted fitness search.

And yet it fits your idea of an implicit absolute target, perfectly.

And the world is full of examples of contingent implicit targets, according to your rubric.

Which provides us with a hint as to why all these models of evolution have, in your view, targets - especially, implicit ones.
You quoted, but then ignored, this, btw. Overlooked?

Perhaps you could give us an example of selection via a "relative comparison" with no implicit target. I'm drawing a blank.

Originally Posted by iceaura

Originally Posted by cypress

One of us is confused. Darwin never claimed that reproductive success is a targeted fitness search.

And yet it fits your idea of an implicit absolute target, perfectly.

No it doesn't fit. Reproductive success has no defined ideal configuration. No defined configuration, no target.

And the world is full of examples of contingent implicit targets, according to your rubric.

Which provides us with a hint as to why all these models of evolution have, in your view, targets - especially, implicit ones.

I don't think we can. There is only better or worse suited for today.

You quoted, but then ignored, this, btw. Overlooked?

Perhaps you could give us an example of selection via a "relative comparison" with no implicit target. I'm drawing a blank.

Natural selection works on relative fitness with no defined target. The models generally don't.

Originally Posted by spuriousmonkey

Originally Posted by cypress

I don't see anbody providing a scientifically valid argument that GA's provide evidence for capability of evolutionary processes to generate observed diversity. Do you?

There are thousands of peer reviewed papers out there that do that.

Here is one I like.

Salazar-Ciudad, I., & Jernvall, J. (2002). A gene network model accounting for development and evolution of mammalian teeth. Proceedings of the National Academy of Sciences, USA. 99: 8116–8120.

feel free to discuss its merits and its pitfalls with me.

Tell me a little about the algorithm and how it works. Also give me a summary of the results. I wonder what they think they have accounted for.

Originally Posted by cypress

No it doesn't fit. Reproductive success has no defined ideal configuration. No defined configuration, no target.

So this stuff about"implicit target" was just a confusion, on your part, that you have cleared up?

OK. You still have the various defined ideal configurations produced in various environments.

Originally Posted by cypress

I am refering to molecular (genetic) similarities. Are body shapes gene based or developmental or both?

Body shapes are genetic, mostly - expressed during deveolpment, of course.

The genetics that produce them are not often similar, however - evolution converges on similar shapes from many directions, so to speak. The target shapes are then found in many different phylogenetic tress.

For an even sharper example, consider mimicry of various kinds. There the defined target is simple and obvious, and the evolutionary approach to it even measurable.

Originally Posted by iceaura

Originally Posted by cypress

No it doesn't fit. Reproductive success has no defined ideal configuration. No defined configuration, no target.

So this stuff about"implicit target" was just a confusion, on your part, that you have cleared up?

Not confusion. I used implicit targets to describe the strategy used by some GA or EA models.

Originally Posted by cypress

I am refering to molecular (genetic) similarities. Are body shapes gene based or developmental or both?

Body shapes are genetic, mostly - expressed during deveolpment, of course.

They are both then.

The genetics that produce them are not often similar, however - evolution converges on similar shapes from many directions, so to speak. The target shapes are then found in many different phylogenetic tress.

For an even sharper example, consider mimicry of various kinds. There the defined target is simple and obvious, and the evolutionary approach to it even measurable.

Target shapes and patterns, or ones that function better than others?

Originally Posted by cypress

Not confusion. I used implicit targets to describe the strategy used by some GA or EA models.

And you appear to be confused in that usage. I see no relevant difference between the convergent shapes of evolution and the "implicit targets" of evolution-modeling algorithms - and you have since amended your postulation of "targets" to "clearly described" targets, apparently in intuitive recognition of that difficulty.

Originally Posted by cypress

Target shapes and patterns, or ones that function better than others?

Define "better".

Let's use an example. The chess playing program "Deep Blue" evolves, becomes a better functioning chess playing program through "mutation and selection", also known as "trial and error". Are targets involved, and if so how do they differ from the apparently equivalent entities of Darwinian bacterial evolution?

Originally Posted by iceaura

And you appear to be confused in that usage. I see no relevant difference between the convergent shapes of evolution and the "implicit targets" of evolution-modeling algorithms - and you have since amended your postulation of "targets" to "clearly described" targets, apparently in intuitive recognition of that difficulty.

No, my position is unchanged. targets are goals. They are an end in mind. A finish line, whether explicit or implicit. Is it your position that convergent evolution implies that evolution is goal directed?

Originally Posted by cypress

Target shapes and patterns, or ones that function better than others?

Define "better".

Am I being unclear? Choose any dictionary definition you like.

Let's use an example. The chess playing program "Deep Blue" evolves, becomes a better functioning chess playing program through "mutation and selection", also known as "trial and error". Are targets involved, and if so how do they differ from the apparently equivalent entities of Darwinian bacterial evolution?

Deep Blue is much more than a trial and error alogrithm. It also is aware of the rules. It is also an oracle as it anticipates moves based on designer input. It has a clear definition of the goal (of winning) and looks as far forward into the future as it can to choose the most favorable move each time. I don't think that is a good example for you. Deep Blue makes use of several goal directed targets programmed into it by its designers. Surely you see this as different than Darwinian evolution. You seem to be suggesting that Darwinian evolution may be a goal driven process, is that what you are saying?

Originally Posted by cypress

Originally Posted by spuriousmonkey

Originally Posted by cypress

I don't see anbody providing a scientifically valid argument that GA's provide evidence for capability of evolutionary processes to generate observed diversity. Do you?

There are thousands of peer reviewed papers out there that do that.

Here is one I like.

Salazar-Ciudad, I., & Jernvall, J. (2002). A gene network model accounting for development and evolution of mammalian teeth. Proceedings of the National Academy of Sciences, USA. 99: 8116–8120.

feel free to discuss its merits and its pitfalls with me.

Tell me a little about the algorithm and how it works. Also give me a summary of the results. I wonder what they think they have accounted for.

read the paper.

Originally Posted by spuriousmonkey

Originally Posted by cypress

Originally Posted by spuriousmonkey

Originally Posted by cypress

I don't see anbody providing a scientifically valid argument that GA's provide evidence for capability of evolutionary processes to generate observed diversity. Do you?

There are thousands of peer reviewed papers out there that do that.

Here is one I like.

Salazar-Ciudad, I., & Jernvall, J. (2002). A gene network model accounting for development and evolution of mammalian teeth. Proceedings of the National Academy of Sciences, USA. 99: 8116–8120.

feel free to discuss its merits and its pitfalls with me.

Tell me a little about the algorithm and how it works. Also give me a summary of the results. I wonder what they think they have accounted for.

read the paper.

I did. Other readers would be interested in a summary as well. Forgive me for thinking you actually wanted to have a discussion.

Originally Posted by cypress

Target shapes and patterns, or ones that function better than others?
Define "better".


Am I being unclear? Choose any dictionary definition you like.

All the definitions I know create your proposed "implicit targets", which you also claim do not exist in Darwinian evolutionary processes.

Originally Posted by cypress

Deep Blue is much more than a trial and error alogrithm. It also is aware of the rules. It is also an oracle as it anticipates moves based on designer input. It has a clear definition of the goal (of winning) and looks as far forward into the future as it can to choose the most favorable move each time.

Sounds like a good little evolving entity to me. It's not "aware" of the rules, for example - the rules are just a set of impossible actions, much as suddenly acquiring weightlessness is impossible for insects. All of that anticipation and stuff is evolved capability. And the definition of "winning" is simply a selection criterion. If you agree that "targets" are thereby created, then you agree.

But you claim that selection criteria do not create implicit targets, when they exist as part of a Darwinian evolutionary sequence? Is that right?

Or is it that unless input from humans, they are not "criteria" at all - that humans cannot model evolution because everything humans do is essentially different, for reasons of essential difference?

Originally Posted by cypress

Originally Posted by spuriousmonkey

Originally Posted by cypress

Originally Posted by spuriousmonkey

Originally Posted by cypress

I don't see anbody providing a scientifically valid argument that GA's provide evidence for capability of evolutionary processes to generate observed diversity. Do you?

There are thousands of peer reviewed papers out there that do that.

Here is one I like.

Salazar-Ciudad, I., & Jernvall, J. (2002). A gene network model accounting for development and evolution of mammalian teeth. Proceedings of the National Academy of Sciences, USA. 99: 8116–8120.

feel free to discuss its merits and its pitfalls with me.

Tell me a little about the algorithm and how it works. Also give me a summary of the results. I wonder what they think they have accounted for.

read the paper.

I did. Other readers would be interested in a summary as well. Forgive me for thinking you actually wanted to have a discussion.

Nobody is interested in primary scientific literature except scientists. Otherwise people on forums would quote primary literature instead of websites.


Summary: an inhibitor-activator model can give rise to cusp patterns that are reiterated in the fossil record and extant species (. By comparing actual inhibitor activator patterns in extant species the authors found their coincided with the pattern generated in the model. The model also generates developmental series that coincide with actual developmental series of the cusp pattern.

Originally Posted by spuriousmonkey

Nobody is interested in primary scientific literature except scientists. Otherwise people on forums would quote primary literature instead of websites.

I cite primary literature more often than websites. Does that make me a scientist? 8)

I have to amend the categories.

Currently we have non-scientists and scientists. Let's add a small subset of scientific minded non-scientists.

But in reality there is probably a continuous distribution.

Originally Posted by spuriousmonkey

Nobody is interested in primary scientific literature except scientists. Otherwise people on forums would quote primary literature instead of websites.


Summary: an inhibitor-activator model can give rise to cusp patterns that are reiterated in the fossil record and extant species (. By comparing actual inhibitor activator patterns in extant species the authors found their coincided with the pattern generated in the model. The model also generates developmental series that coincide with actual developmental series of the cusp pattern.

Hmm... that does not sound very similar to the primary context of this discussion of evolutionary algorithms. A model yes, but is it an GA? If so please describe the algorithm and how it works. How are the intermediate steps validated? Furthermore it is difficult to understand how modifications of traits that define cusp patterns in molars is an example of novel form or function. How is this any different than what we already know about modification of traits by genetic drift and confirmed by animal breeding?

In contrast to spuriousmonkey's spurious and irrelevant article, here is a relevant example of a Genetic or Evolutionary Algorithm and the claim that it is evidence for Darwinian evolutionary processes of mutation and selection.

“The Evolutionary Origin of Complex Features.” by Richard Lenski, Charles Ofria, Robert Pennock, and Christoph Adami, NATURE (vol 423, pp 139-144)

here is a link to an explanation of the code.

http://authors.library.caltech.edu/13705/1/OFRal04.pdf

The abstract:


A long-standing challenge to evolutionary theory has been whether it can explain the origin of complex organismal features. We examined this issue using digital organisms—computer programs that self-replicate, mutate, compete and evolve. Populations of digital organisms often evolved the ability to perform complex logic functions requiring the coordinated execution of many genomic instructions. Complex functions evolved by building on simpler functions that had evolved earlier, provided that these were also selectively favoured. However, no particular intermediate stage was essential for evolving complex functions. The first genotypes able to perform complex functions differed from their non-performing parents by only one or two mutations, but differed from the ancestor by many mutations that were also crucial to the new functions. In some cases, mutations that were deleterious when they appeared served as stepping-stones in the evolution of complex features. These findings show how complex functions can originate by random mutation and natural selection.

The problem of course is as I mentioned in the opening post. In the case of Avida which uses random selection, insertion, deletion, etc. of software instructions to generate code that performs a targeted search to generate the function that performs NXOR logic. The authors refer to it as EQU since the output of NXOR is 1 if both inputs are 1 or both are 0. Thus its target is to discover the EQU function.

Avida uses specific key information inserted into the algorithm that certain combinations of the universal logic function NAND when combined in certain ways generates all other logic functions. It specifically rewards instruction sets that contain these simpler logic functions that are building blocks for NXOR. The Avida program will not converge on its goal in any reasonable time without scoring and rewarding certain logic functions that are incorporated into the evolving instruction set. While it makes sense to do this, it would also make sense to reward any group of instructions that perform any useful function even if that useful function is not related to generating the NXOR target. When this is done, the Avida program never reaches its target.

It is also critical that certain instructions are included in the pool of available instruction lined substitutions. Certain instructions that Avida uses are deleterious while others are mandatory. The designers of Avida were aware of this and ensured that these operations were included in the list of available instructions.

Originally Posted by cypress

While it makes sense to do this, it would also make sense to reward any group of instructions that perform any useful function even if that useful function is not related to generating the NXOR target. When this is done, the Avida program never reaches its target.

In your version, so what if it doesn't reach its target? It evolves useful functions, and complex structures to perform them, right?

Because that was your selection criterion - useful performance.

Originally Posted by iceaura

Originally Posted by cypress

While it makes sense to do this, it would also make sense to reward any group of instructions that perform any useful function even if that useful function is not related to generating the NXOR target. When this is done, the Avida program never reaches its target.

In your version, so what if it doesn't reach its target? It evolves useful functions, and complex structures to perform them, right?

Because that was your selection criterion - useful performance.

actually no it just meanders around never accumulating any discernible increase in useful performance just swapping one function for another.

Originally Posted by cypress

actually no it just meanders around never accumulating any discernible increase in useful performance just swapping one function for another

So your selection criterion was not "useful performance", then, as claimed.

Unless you built in some kind of equivalence between all the functions, so that they were all interchangeable only and without mutual effects. That of course would not model evolutionary development very well.

Originally Posted by iceaura

Originally Posted by cypress

actually no it just meanders around never accumulating any discernible increase in useful performance just swapping one function for another

So your selection criterion was not "useful performance", then, as claimed.

No, it was. scores were assigned in the same manner assigned for logic functions. Simple solutions were given more credit.

Unless you built in some kind of equivalence between all the functions, so that they were all interchangeable only and without mutual effects. That of course would not model evolutionary development very well.

No, they were not made equivalent.

More likely, the addition of alternatives provided a mechanism that altered fitness landscape so that the pathways to significant alterations were cut off. But either your view or mine leads to the same conclusion. These algorithms do not accurately model evolutionary behavior. since they don't, they should not be used as evidence that over time, evolutionary processes are capable of generating new form and function Given that the models do what they are designed to do rather than mimic evolutionary processes, they do not demonstrate the power of evolution.

Originally Posted by cypress

actually no it just meanders around never accumulating any discernible increase in useful performance just swapping one function for another
So your selection criterion was not "useful performance", then, as claimed.


No, it was. scores were assigned in the same manner assigned for logic functions. Simple solutions were given more credit.

Look: either you are objecting to criteria of useful performance being used in selection, or you are objecting to useful performance not being favored in selection. Which?

Originally Posted by cypress

Unless you built in some kind of equivalence between all the functions, so that they were all interchangeable only and without mutual effects. That of course would not model evolutionary development very well.


No, they were not made equivalent.

More likely, the addition of alternatives provided a mechanism that altered fitness landscape

Most likely of all, you do not understand the logical structure of evolutionary theory, and do not recognize the employment or modeling of it.

I can't figure out what your objection to these models, even given the structure of them alleged by you, actually is.