Dawkins blind programmer

[npetreley]

http://www.rae.org/MutationProgram.htm

We found that if you remove these false assumptions and allow random mutations to occur freely whether they match the goal or not, you can run the simulation for millions of generations without ever getting close to the goal.

...

But the most amazing revelation we had when we were doing this simulation is that if Dawkins’ save function were true, it would prevent the organism from ever evolving again, not to mention that implicitly if you try to match a goal, you are introducing intelligent design into the equation.

 

[LiveFreeOrDie]

We found that if you remove these false assumptions and allow random mutations to occur freely whether they match the goal or not, you can run the simulation for millions of generations without ever getting close to the goal.

Translation: If we remove any behavior analogous to natural selection, evolution does not occur.

My response: Duh.

But the most amazing revelation we had when we were doing this simulation is that if Dawkins’ save function were true, it would prevent the organism from ever evolving again,

Unless somehow the goal changes.

not to mention that implicitly if you try to match a goal, you are introducing intelligent design into the equation.

So when a plant grows toward a light (goal) that is an example of intelligence? I don't think so.

 

[npetreley]

Originally posted by LiveFreeOrDie
Translation: If we remove any behavior analogous to natural selection, evolution does not occur.

You are being disingenuous. The articlea actually pointed out that the algorithm fails to produce results when you DO introduce behavior analogous to real mutations and naturual selection.

Originally posted by LiveFreeOrDie
My response: Duh.[/B]

A very appropriate response. The saved letters made no sense until the phrase was complete. In other words, his algorithm had intelligent design and foreknowledge. Natural selection does not know in advance what will be useful and then prepare and save mutations for that future use.

Originally posted by LiveFreeOrDie
Unless somehow the goal changes.

Again, you betray your ignorance of natural selection. It has no goals.

Originally posted by LiveFreeOrDie
So when a plant grows toward a light (goal) that is an example of intelligence? I don't think so.

Whether or not the plant is "intelligent" is not the issue. The designer is in view here, not the living creature. The fact that a plant moves toward light, or that a plant "twirls" until it hits an object and then winds around that object (vines) are good examples of intelligent design and therefore point to an intelligent designer.

 

[Morat]

We recreated this simulation, and introduced a number of other factors that Dawkins probably did not consider. First of all, we make an assumption that when a letter (or gene) matches a desired goal, there is something that prevents it from further mutation, even though there is no biochemical evidence or process to back up this assumption. This way we give the process the benefit of the doubt way beyond what the evidence calls for. Second, if we presume that assumption to be true, then once the organism achieved the “goal” it is forever prevented from further evolution. His thesis (and therefore evolution) is refuted both ways, for if we take away this “mutation save” function, our simulation goes on for millions of generations without achieving the goal. Without the source code to Dawkins’ program, we can only run a simulation that gives the benefit of the doubt far beyond what is reasonable, and establish limits within which evolution can occur

  No, Nick, they didn't accurately simulate Natural Selection. Note the first bolded section. They added behavior that does not occur in evolution, and even admitted it.
   The second bolded section is also untrue, but rather pointless to the discussion at hand. 


   Their first claim "goes on for millions of generations without achieving the goal" is false. I have written a Weasel program myself, and get the result quickly. If they're not getting the result, then their fitness function sucks.

  And, yes it does!

In addition to testing the results both with and without the save function, our computer simulation adds into the mixture several more factors involved in mutations that serve to increase the number of generations.  Our simulation saves the results in a table if the results match a specified number of letters in the goal. When we ran the simulation without the save function for more that 7 million generations, we could only produce 6 matches on 9 letters out of 28.

   This is not a proper fitness function, and does not mimick natural selection (heck, it doesn't even mimick reproduction well).

    My weasel program used 26 strings, initially formated to null. Every "generation" each of these strings has a chance to recieve a simple point mutation.

   After the strings have been mutated, they are compared to the phrase "Methinks it is like a weasel".

   Now selection occurs: The 13 strings least like "Methinks it is like a weasel" are discarded. Now reproduction occurs: The remaining 13 are cut apart and recombined to form 26 new strings.

    Fitness is checked again. Here's a generic GA:

   1) Check population against fitness function.

   2) Cull least fit.

    3) Repopulate using most fit (reproduction).

    4) Mutate.

    5) Check population against fitness function.

   Nick, they didn't write a genetic algorithm. Not even close. They wrote a program where they randomly changed letters in a string. That's not a genetic algorithm, and only someone utterly ignorant of the subject would think so. There was no reproduction, no population, and their fitness function was useless.

  Five simple steps, and they managed to get one right.

   I thought all those Linux junkies had software skills. Guess not!

Again, you betray your ignorance of natural selection. It has no goals. 

 It has a filter: Survival. Which means, "enviromental fitness". Which means the goal of natural selection is to produce fit species.

    Which is what fitness functions do.


 

 

[LiveFreeOrDie]

Originally posted by npetreley
You are being disingenuous. The articlea actually pointed out that the algorithm fails to produce results when you DO introduce behavior analogous to real mutations and naturual selection.

I take issue with your assertion that the authors' modifications somehow more accurately model evolution. Specifically, the authors call the following a false assumption:

"Mutations are saved if it matches the goal"

Without this rule, there is no "natural selection" in the model. All letter combinations are equally "fit". Of course the simulation won't converge.

I also question the authors' claim that their additional rules more closely model evolution.

"Mutations may be inserted, moving the entire string over one gene."

This rule is fine, but they neglected to add the converse: mutations may also delete a gene.

"Mutations favor “hotspots” and rarely occur elsewhere"

While it is true that some areas of the genome do seem to be more prone to mutation than others, to say that they "rarely occur elsewhere" is simply wrong.

"Mutations heavily favor the T gene"

This is simply a repeat of the prior, incorrect rule.

"Mutations may result in unusable or meaningless code, represented by special characters in our simulation"

Irrelevant. They're just adding letters to the alphabet. As long as one population is "more fit" than another, they "more fit" will survive.

"Mutations may kill the organism with a “poison factor” and stop the process"

True, but invalid in the context of this simulation. Remember, populations evolve, not individuals. For every individual that inherits a fatal mutation, there will be tens of others that inherited a neutral or, more rarely, a beneficial one.

Does Dawkins' simulation model evolution perfectly? Of course not. It wasn't intended to. All it was designed to show is that random mutation plus natural selection can produce "information".

A very appropriate response. The saved letters made no sense until the phrase was complete. In other words, his algorithm had intelligent design and foreknowledge. Natural selection does not know in advance what will be useful and then prepare and save mutations for that future use.

The meaning of the letters in the English language has absolutely nothing to do with the validity of the simulation. Dawkins simply chose a very simple and easy to understand "fitness function": similarity to a given sequence of letters. If he instead chose "maximum number of letters the same", it wouldn't change the result of the simulation yet the final sequence would a-priori unknown.

Real-world fitness functions are hideously complex. It would obscure Dawkins' point to try creating one even remotely life-like.

Again, you betray your ignorance of natural selection. It has no goals.

Yes it does. Survival.

Whether or not the plant is "intelligent" is not the issue. The designer is in view here, not the living creature. The fact that a plant moves toward light, or that a plant "twirls" until it hits an object and then winds around that object (vines) are good examples of intelligent design and therefore point to an intelligent designer.

That's a topic for another thread.

 

[LiveFreeOrDie]

Wow, Morat, great minds think alike, eh?

 

[Morat]

  I'd like to echo the "Real world fitness functions are insanely complex". Creating a fitness function isn't easy in the simplest of cases. Even with the "Weasel" program, I still had to think a bit to come up with a fitness function that was workable (point-based. Each letter in the correct spot was one point. High scorers went on to another round).  

 

[DNAunion]

LiveFreeOrDie: Does Dawkins' simulation model evolution perfectly? Of course not. It wasn't intended to.

DNAunion: Indeed. In fact, on the page after the discussion of the WEASEL program in THE BLIND WATCHMAKER Dawkins himself admits that the fixed target he uses does not model natural evolution. He wasn't intending to accurately model evolution - his point was about cumulative selection vs. single-shot selection.

 

[npetreley]

Originally posted by DNAunion
DNAunion: Indeed. In fact, on the page after the discussion of the WEASEL program in THE BLIND WATCHMAKER Dawkins himself admits that the fixed target he uses does not model natural evolution. He wasn't intending to accurately model evolution - his point was about cumulative selection vs. single-shot selection.

That's correct. And in so doing he has misrepresented the whole problem. There are two issues here: the number and types of mutations necessary to produce a feature, and the natural selection of that feature. He confused them into one algorithm, so that the mutations (letters) were treated as if they were the features selected. But that's not how it works at all, even by evolutionary speculation. So his example proved only that he's a blind (inadequate) programmer who can't even figure out how to illustrate his point.

 

[excreationist]

npetreley:
Maybe the program they used was a dud.
Try this Java one:
http://home.pacbell.net/s-max/scott/weasel.html

It only takes a few hundred generations and a few seconds. It uses a population size of 1024 though (in the Java version). The first try with having a population size of 10,000 (in the Java applet) I got the solution in 38 generations. The second time - 54. Then 43. Then 44. It really depends on the population size I think.

I think Richard Dawkin's book also had software which evolved "biomorphs"

I think these were just pretty shapes where the user guides the evolutionary process.

There are even more realistic evolutionary programs... e.g. Gene Pool for Windows. I highly recommend you try it out and have it run on your computer for a couple hours. So far I've got to time 100,000. (see the stats button) The info button tells you the technical details about the program. Unfortunately I think that is the only way to see that text. (It isn't on the internet)
The other humanoid and frankenstein pools are pretty interesting to look at but they seem to die out after a while.
The creatures mutate quite a lot and in nature the mutation rate would be much lower....
In this program they'd be able to evolve significantly within maybe 100 generations - but in nature there would be millions of generations - remember - millions of years? To simulate that to a certain degree you could just leave the simulation running for a couple weeks.
There would be programs that simulate evolution more accurately - but this is one of the best easy-to-use ones I've come across so far.

Some comments on your link:

1. Mutations may be inserted, moving the entire string over one gene.

That would be good for explaining how the sizes of DNA can increase - e.g. from fish to humans. The programmers probably would think it would be a fairly pointless feature... natural selection would still ensure that the fitter organisms breed more. It would make the program take longer to make, and probably make it run a lot slower. There might be some programs that work like that though.

2. Mutations favor “hotspots” and rarely occur elsewhere.

This would be easy to add into a program...

3. Mutations heavily favor the T gene.

These simulations don't use the four bases that we have in genetics which create amino acids and proteins... individual cells are made up of millions of molecules I think and it might require a supercomputer to somewhat accurately simulate even one cell - so they are simplified. I wonder if the writer thinks the "T gene" is used to make the letter T in that weasel program...

4. Mutations may result in unusable or meaningless code, represented by special characters in our simulation.

This would just slow down a computer simulation. The idea of these simplified programs for everyday laypeople is to let them see things evolve quickly before their eyes... to have huge populations and huge numbers of generations, people would need very powerful computers and lots of time on their hands. Those more realistic simulations are more suited to researchers.

5. Mutations may kill the organism with a “poison factor” and stop the process.

This is seen in programs which search for food to some degree - if they have a serious mutation it would starve to death... this may be the reason why the organism would die in real life - and it could also be because it is missing vital components in its body, etc.

Update:

Framsticks is the most advanced free 3D evolution simulator I've come across... well it is free to see it in wireframe, but not proper textured 3d...

The genomes don't have a set length and genes are probably inserted during mutations. Here is more in-depth information about their genotypes. There would be a potential for extremely complex organisms to evolve. Apparently the organisms in this program can even fight each other... they can also swim.

Here is an example of evolution using Framsticks... as you can see, the length of the genome clearly changes. It doesn't say how many generations were involved.

 

[LiveFreeOrDie]

Originally posted by npetreley
...so that the mutations (letters) were treated as if they were the features selected. But that's not how it works at all...

But that IS exactly how it works. Remember, our DNA is composed of four bases, or "letters": A, C, T, G. Ultimately, all of the inherited features of the organism are determined by these letters. To say that nature selects certain features implies that it is selecting certain DNA letter combinations over others.

 

[npetreley]

Originally posted by LiveFreeOrDie
But that IS exactly how it works. Remember, our DNA is composed of four bases, or "letters": A, C, T, G. Ultimately, all of the inherited features of the organism are determined by these letters. To say that nature selects certain features implies that it is selecting certain DNA letter combinations over others.

But natural selection doesn't preseve letters, it preserves the features that they produce. So a proper analogy would be to mutate until it gets to the working phrase, which wouldn't save individial letters. Give it up LFOD, you're just ignoring the obvious.

 

[LiveFreeOrDie]

Originally posted by npetreley
So a proper analogy would be to mutate until it gets to the working phrase, which wouldn't save individial letters.

I don't understand what you mean. Please illustrate.

 

[npetreley]

Originally posted by LiveFreeOrDie
I don't understand what you mean. Please illustrate.

You said it yourself: "To say that nature selects certain features implies that it is selecting certain DNA letter combinations over others."

In other words, it doesn't preserve letters, it preserves COMBINATIONS. So you can't save letters in a phrase and call that an algorithm that illustrates natural selection. You'd have to mutate the letters without saving any of them until you got the combination that is selected for, in this case, the "methinks" phrase. Then you'd save THAT. But you'd never even get to the phrase that way, so Dawkins scrambled the analogy and cheated.

 

[LiveFreeOrDie]

Originally posted by npetreley
You said it yourself: "To say that nature selects certain features implies that it is selecting certain DNA letter combinations over others."

In other words, it doesn't preserve letters, it preserves COMBINATIONS. So you can't save letters in a phrase and call that an algorithm that illustrates natural selection. You'd have to mutate the letters without saving any of them until you got the combination that is selected for, in this case, the "methinks" phrase. Then you'd save THAT. But you'd never even get to the phrase that way, so Dawkins scrambled the analogy and cheated.

Why do you assume that there is only one "phrase" that is "more fit" than all the others? Are real fitness functions really this discrete? Or are there other combinations between "xxxxxxxx" and "methinks" that have intermediate levels of fitness? Throw a couple of wildcard characters into the fitness test and evolution becomes rather inevitable.

What if our fitness function looked like this:
(doesn't match anything below) = fitness 0
me* = fitness 1
me??i?k? = fitness 2
met?i?k? = fitness 3
met?inks = fitness 4
methinks = fitness 5

Obviously this will take more generations to converge, but it WILL converge.

Now that I think about it, this reminds me of those word puzzles where you need to change one word into another, changing only one letter at a time, and where each intermediate is still a word:

dog
dot
cot
cat

If you think RM + NS can't do THAT, then it's pretty clear who the blind one here is.

 

[npetreley]

Originally posted by LiveFreeOrDie
Why do you assume that there is only one "phrase" that is "more fit" than all the others? Are real fitness functions really this discrete? Or are there other combinations between "xxxxxxxx" and "methinks" that have intermediate levels of fitness? Throw a couple of wildcard characters into the fitness test and evolution becomes rather inevitable.

It really is useless arguing this with you, since you're still confusing the concepts. Even your example assumes that the correct letters would have some reason to remain unmutated until a useful wild card came up and in so doing created some useful combination of letters that could be selected. This does not reflect the hypothesis of mutation and selection (the kind that produces new features, not variation). This hypothesis never suggests the process will maintain a "correct" letter if it isn't yet part of a useful chain. Because until the chain is useful, none of the letters are "correct".

 

[Morat]

It really is useless arguing this with you, since you're still confusing the concepts. Even your example assumes that the correct letters would have some reason to remain unmutated until a useful wild card came up and in so doing created some useful combination of letters that could be selected. This does not reflect the hypothesis of mutation and selection (the kind that produces new features, not variation).

  Jeez, Nick. You really are clueless about evolution. And genetic algorithms. You're the one who doesn't have a clue about this. I'd think a linux junkie would be able to handle something this simple, but....

   The correct letters do have a reason to remain unmutated. They're more fit if they're correct. Yes, the "correct" letters have the same chance to mutate each generation as any other letter. But mutating to an "incorrect" letter lowers the organism's fitness, making it highly unlikely to be selected.

This hypothesis never suggests the process will maintain a "correct" letter if it isn't yet part of a useful chain. Because until the chain is useful, none of the letters are "correct".

   Oh bull, Nick. A correct letter makes an organism more fit for it's enviroment. The whole phrase constitutes maximum possible fitness.

   The difference between "Me*ghtss* and "Methinks" is the difference between a simple eye and a complex one. "Methinks" is the best possible eye. "Mgfgfgfgf" is an eye that conveys even a small advantage (say, light sensitivity).

   The fundamental difference between Genetic Algorithms (in general) and evolution in GA's, we define what counts as "fit", whereas in evolution the enviroment defines it.

 All you've proven here, Nick, is that you don't understand evolution, and you're incapable of understanding simple genetic algorithims as well.

 

[LiveFreeOrDie]

It really is useless arguing this with you, since you're apparently unable to grasp even the simplest concept.

Even your example assumes that the correct letters would have some reason to remain unmutated until a useful wild card came up and in so doing created some useful combination of letters that could be selected.

I assume no such thing. If the difference between two fitness levels is two letters, then I am requiring both letters to appear in the same mutation. Yes, the odds are small (1 : 26^2 to be exact), but they are still well within reason -- especially given large populations and long periods of time.

This does not reflect the hypothesis of mutation and selection (the kind that produces new features, not variation).

There is only one kind of RM+NS, Nick. "Variation" and "New Features" are simply different points on the same scale.

This hypothesis never suggests the process will maintain a "correct" letter if it isn't yet part of a useful chain. Because until the chain is useful, none of the letters are "correct".

There is no teleology in evolution, Nick. As long as each step in the chain is more fit than the prior one, it will be selected for. It's that simple. That's why I don't understand why you don't get it.

 

[npetreley]

Originally posted by LiveFreeOrDie
If the difference between two fitness levels is two letters, then I am requiring both letters to appear in the same mutation. Yes, the odds are small (1 : 26^2 to be exact), but they are still well within reason -- especially given large populations and long periods of time.

It's not the difference between two fitness levels. It's the difference between having a function and not having one. And there's nothing to select for until it has one. But just for the sake of argument, let's see how silly your assumption is that it's reasonable. Apply your logic that it's reasonable for two letters to stay the same until the whole sequence has a function to this real-life sequence:

CGGCGCCGCGAGCTTCTCCTCTCCTCACGACCGAGGCAGAGCAGTCATTATGGCGAACCTTGGCTGCTGG
ATGCTGGTTCTCTTTGTGGCCACATGGAGTGACCTGGGCCTCTGCAAGAAGCGCCCGAAGCCTGGAGGAT
GGAACACTGGGGGCAGCCGATACCCGGGGCAGGGCAGCCCTGGAGGCAACCGCTACCCACCTCAGGGCGG
TGGTGGCTGGGGGCAGCCTCATGGTGGTGGCTGGGGGCAGCCTCATGGTGGTGGCTGGGGGCAGCCCCAT
GGTGGTGGCTGGGGACAGCCTCATGGTGGTGGCTGGGGTCAAGGAGGTGGCACCCACAGTCAGTGGAACA
AGCCGAGTAAGCCAAAAACCAACATGAAGCACATGGCTGGTGCTGCAGCAGCTGGGGCAGTGGTGGGGGG
CCTTGGCGGCTACATGCTGGGAAGTGCCATGAGCAGGCCCATCATACATTTCGGCAGTGACTATGAGGAC
CGTTACTATCGTGAAAACATGCACCGTTACCCCAACCAAGTGTACTACAGGCCCATGGATGAGTACAGCA
ACCAGAACAACTTTGTGCACGACTGCGTCAATATCACAATCAAGCAGCACACGGTCACCACAACCACCAA
GGGGGAGAACTTCACCGAGACCGACGTTAAGATGATGGAGCGCGTGGTTGAGCAGATGTGTATCACCCAG
TACGAGAGGGAATCTCAGGCCTATTACCAGAGAGGATCGAGCATGGTCCTCTTCTCCTCTCCACCTGTGA
TCCTCCTGATCTCTTTCCTCATCTTCCTGATAGTGGGATGAGGAAGGTCTTCCTGTTTTCACCATCTTTC
TAATCTTTTTCCAGCTTGAGGGAGGCGGTATCCACCTGCAGCCCTTTTAGTGGTGGTGTCTCACTCTTTC
TTCTCTCTTTGTCCCGGATAGGCTAATCAATACCCTTGGCACTGATGGGCACTGGAAAACATAGAGTAGA
CCTGAGATGCTGGTCAAGCCCCCTTTGATTGAGTTCATCATGAGCCGTTGCTAATGCCAGGCCAGTAAAA

Duh.

Originally posted by LiveFreeOrDie
There is only one kind of RM+NS, Nick. "Variation" and "New Features" are simply different points on the same scale.

Congratulations. You just identified where creationists differ with evolutionists.

Originally posted by LiveFreeOrDie
There is no teleology in evolution, Nick. As long as each step in the chain is more fit than the prior one, it will be selected for. It's that simple. That's why I don't understand why you don't get it. [/B]

Because it's wrong. No single mutation in a sequence can be selected for if the resulting sequence doesn't have any function. There's no reason for a letter to remain constant and wait for the others to join it in making sense.

 

[LiveFreeOrDie]

Originally posted by npetreley
It's not the difference between two fitness levels. It's the difference between having a function and not having one. And there's nothing to select for until it has one.

But that's the whole idea if the intermediate fitness levels. The intermediates DO have a function (else why would there be any effect on fitness?)

METHUNKZ doesn't do the job as well as METHINKS, but it still does it better than MEOWMEOW.

But just for the sake of argument, let's see how silly your assumption is that it's reasonable. Apply your logic that it's reasonable for two letters to stay the same until the whole sequence has a function to this real-life sequence:

"My logic" is that there are many intermediate sequences that still provide a fitness benefit to the organism.

Duh.

Because it's wrong. No single mutation in a sequence can be selected for if the resulting sequence doesn't have any function. There's no reason for a letter to remain constant and wait for the others to join it in making sense.

There is if a few correct letters still results in an organism that is better able to survive and reproduce than one with no correct letters.