This is just an effort to help keep Joe’s thread focused and to help keep it from being derailed. People can use it or not. I hope they will.
CharlieM: Can someone explain to me, why is all of this not just a model of directed evolution? Surely it is set up to be directed towards a target?
Allan Miller: It gets towards the target by means of variation and selection of genotypes in the current population. The programmer does not direct it towards the target. Indeed, there would be no point in writing GAs for problem solution if it were simply a matter of specifying a target and directing the program to find it.
Consider a small modification – instead of distance from target, evaluate fitness by adding up the ASCII bits. Those with the greater sum are fitter than those with a lesser. There is no mention of a distant target – although it is clear that the program will converge on a string of all Zs, the program doesn’t know this. It is not drawn by, nor directed towards, that target. It is simply doing generational evaluation of fitter and less fit genotypes in the current population.
Discuss.
But if the best is unique? No, yes?
Maybe, but the travelling salesman GA is not guaranteed to find the best. My understanding is that given a large number of nodes, there is no way of knowing what the best is.
There are competitions involving 10,000 node problems. I think it is possible to calculate the minimum possible route length, and to know when you have reached an acceptable approximation. (I’m not sure about that).
Biological evolution has a concept known as Dollo’s Law, which is something like a directional arrow. My lack of math sophistication will show here, but my understanding is that from any “starting point” in biology, there are many possible non-fatal mutations, only one of which reverses the last mutation. The chances of evolving in reverse run into your law of big numbers. It doesn’t happen.
Trees don’t evolve into humans because every change in the genome constrains the universe of possible non-fatal mutations. Drunkard’s walk, and all that. The direction is not forced or predictable, but it is constrained.
Thomas Ray’s TIerra rewards only continued existence. It demonstrated that parasitism and hyperparasitism could evolve using a subset of known evolutionary mechanisms.
Allan Miller,
I don’t understand Allan, why would I need to WRITE a GA? nature didn’t write evolution right? It just happened by accident, because things are imperfect.
So, all I need is a faulty computer and just turn it on and let it make mistakes. Eventually it will make something good right? Ok, it might take a while to make something as good as a cheetah, but it can probably at least make a pterodactyl or write a three part tragic comedy.
I continue to be amazed at how many experts on GAs refuse to code one and play with it. I only know of one UD regular who actually experimented with GAs, and he quit posting there shortly thereafter.
What if we had 10 different target phrases and measured fitness as the lowest Hamming distance (or the highest
in Joe’s model)?
Answer, you would have no way of predicting which one it would converge to (starting at a random initial string) it would likely converge to the closest one.
And what if we picked an initial string that’s at the same Hamming distance from all target strings? The SAME algo would converge on random targets.
Same algo, different fitness landscape, unpredictable and random results. How is fitness landscape in any way relevant to the algo and whether it’s “directed” or not? Well, it’s not IMO
stcordova,
You point to a place where you mention you wrote a GA, but the dog ate it.
I am, nonetheless, quite prepared to believe that you wrote something. If it was a GA, I retract.
The idea of directedness is movement toward something despite encountering obstacles. When we have auto-pilot guidance systems in an airplane and wind gusts hit the airplane, the guidance system sets it aright, it directs it toward the goal.
WEASEL is like that. It selects toward a trait that doesn’t even exist. It can select toward non-existent irreducibly complex Rube Goldbergesque designs that don’t exist yet in a cumulative fashion. It is really conceptual cumulative selection, it isn’t natural selection.
As I said before, Darwin and Dawkins falsely advertised conceptual selection to be what happens in nature and called it natural selection. It’s a rhetorical misdirection, it isn’t what really happens naturally. If they said “creatures evolved by UN-natural selection directed toward non-existent conceptual traits” that would be more accurate description of their hypothesis, but if they said it that way, the absurdity of their claims would be readily apparent.
phoodoo,
If you think that’s a good reason for not writing a GA, I am hardly equipped to persuade you otherwise.
stcordova,
It is curious that an environment that kills more of one type than another results in an increase in the frequency of the one that is killed less often, though. It’s not that absurd, is it?
stcordova,
So how does it work in nature? When an experiment is set up so that certain viruses are put in strong selection pressure for infectivity, and after a while they evolve better infectivity, was God helping scientists get their results?
Certainly not like this where “VIRUS ” or “DONKEY” or “FAWCET” or “CAMERO” or “ROCKET” etc. becomes
“WEASEL”.
No that’s what I call macro-evolution! 🙂
Question were raised as to what was meant by directed evolution , I provided my suggested definition.
I have a program that has about 100,000 words as targets. The selection engine scores genomes according to how many characters match any part of any word.
The direction is unpredictable, and it often converges on a hybrid that looks like a word but is not in the dictionary.
stcordova,
So convergence is directed, nothing else?
It would be a simple matter to have a second target phrase, and randomly evaluate against either at any given decision point. I’d suggest THIS TARGET SUCKS AND BLOWS. If it finds one of them, is it directed? How ’bout 3? 4? Any ideas on a cutoff for directed evolution? Any at all?
Sounds great. Do you have it published somewhere please?
Actually it would be really cool to see how drift makes it explore different target phrases instead of getting stuck in a local maxima. I would add a printout of the current closest target for clarity. Could try tweaking your algo myself. I can hardly remember any C but that should be nothing some googling couldn’t fix
If I coded a GA it would show I could code a GA. So what? Also, how would you know I had not just found it published somewhere and copied it?
It’s for your benefit, not ours.
It was intended as a game, but I never finished the competition mode.
http://itatsi.com
Just click Start for a demo.
You don’t need a DriftWeasel to do that.
There is a target located in a search space” METHINKS IT IS LIKE A WEASEL
Why is a GA the only way to find the target phrase, or even the best search algorithm to use in this case?
petrushka,
Beat me to it. ‘You’re only cheating yourself’, as my mum used to say.
Mung,
It might not be. That’s not its purpose – it’s a teaching aid.
You think if I write a GA that it will cause me to leave UD? How would that benefit me?
If I wanted to teach someone to write a GA I sure wouldn’t use the Weasel program as my example.
I think you would learn something.
Why would you leave UD?
Most of us did not voluntarily leave.
Mung,
I’ve never seen anyone wriggle so much.
Mung,
And that was not its purpose either. It wasn’t intended to teach GA writing, but something about evolution. Namely cumulative selection.
That question answers itself.
Are you saying that a purely selective Weasel can realistically get out of local maxima? How?
dazz,
Mutation … or crossover, in sexual versions.
Allan, to Mung:
It’s amusing, isn’t it? Writing a GA would be win-win for Mung’s education, but it’s lose-lose for Mung’s ego.
If he writes it correctly, his own code will demonstrate that he’s wrong to claim this:
Ouch. And if he doesn’t write it correctly, he’ll be revealing his ignorance of GAs. Ouch.
He won’t be able to hide behind obfuscation, because his code will be there for anyone to inspect. Ouch.
Any way you look at it, his ego loses. Do it for the knowledge, Mung!
Mung isn’t the only one.
I’m not hearing much from the genetic meltdown crowd.
It’s pretty hard to demonstrate that meltdown is inevitable when a GA is parameterized.
Fair enough. But I guess without recombination, it would take a very flat landscape for point mutations alone to be enough. In a weasel algo that would mean having a crap load of target phrases, right?
🙂
Weasel is like digging a 100 ft hole and falling into it.
Think exploration and exploitation.
In addition to coping with nonlinearity, the genetic algorithm helps to solve a conundrum that has long bedeviled conventional problem-solving methods: striking a balance between exploration and exploitation.
The genetic algorithm exploits the higher-payoff, or “target,” regions of the solution space, because successive generations of reproduction and crossover produce increasing numbers of strings in those regions.
ouch.
http://www2.econ.iastate.edu/tesfatsi/holland.GAIntro.htm
Nice link.
What do you mean? Do you think anyone is arguing that environments do NOT present target regions? As I recall, these are called “niches” and are explored for, and exploited when found. The higher payoff is survival, facilitated by differential reproductive success.
You can wish all you want but if I think I can put my point across more clearly by using analogies then I’ll continue to do so.
Natural selection acts like natural selection because certain physical features of the organism determines its survivability. Individual moths are counted and finch beaks are measured because these are things that are observed. There is a correlation between individual gene differences and phenotypal variety but the causal relationships are not so easy to figure out. Light moths, dark moths, thick beaked finches, thin beaked finches; all of these existed before trees changed colour or drought took hold. It was these physical features that caused the change in gene frequencies, not the other way round.
The “gene pool” is an abstraction which is not observed, physical features are observed.
Following drought conditions the average beak depth of Galapagos ground finches was greater. There you have it, environmental conditions directing evolution. You have already agreed on this type of directed evolution. Natural selection is just an effect the environment has on its inhabitants.
Giant pandas and three toed sloths can be said to occupy niches, but to say that humans or rats occupy niches is to make a mockery of the meaning of the word.
I propose transporting humans and rats to Anarctica, without technology.
I would disagree. As petrushka is trying to tell you, the environmental niche where humans can live is fairly narrow. So what humans have done is to modify their environments to their own needs. We have learned how to keep our body temperatures within a narrow range with clothing, heating, air conditioning, dwellings that shelter us from extremes, and so on. And rats piggyback on those technological microenvironments, living where people do.
You are surely aware that the hominin bush had quite a few species, perhaps as many as half a dozen concurrently. Except for one, none of these species survived, or had large populations or lived across broad temperature ranges. Yet all of them must be counted among the most intelligent species ever to live, at least as we conceive of intelligence.
Now, I have seen the argument that because people select their environments rather than environments selecting individuals, we humans have short-circuited natural selection, and are evolving through drift alone. I don’t know.
Perhaps you should let the rest of your friends at UD know the next time they talk about WEASEL as if it represents the finest example of a GA known to humanity?
dazz,
Or, I guess, high mutation and/or weak selection. Both only up to a point.
CharlieM,
I am simply offering my opinion, as a consumer of your prose, that your analogies serve more to confuse – not just your reader, but you.
There is a tension between your desire to communicate and anyone else’s willingness to put the effort in to comprehend. When someone says “it’s like …”, and it’s not like that at all, much time can be wasted talking about something which isn’t in fact the thing being discussed. But it’s your dollar.
CharlieM,
So you think ALL evolution by natural selection is ‘directed’ evolution, with environment as director. So all GAs with a selection routine are directed in that sense. So they are implementing Natural Selection. So why are we arguing?
I missed high mutation too, but weak selection wouldn’t be an option since we were talking about a purely selective algo. Appreciate the answers, not so much the sarcasm, I guess :p
dazz,
Sarcasm? Unintentional, I assure you – not sure what you mean.
Directed evolution and “natural selection” are oxymorons in the view of the average materialist. To equate them smells of pantheism or deism (or Calvinism) and we can’t have that now can we.
To the average materialist the idea of the environment directing either directly or as the intermediary force of another being is pure heresy.
To Mike the materialist the outcome of evolution is not proscribed or directed it just happens all on it’s own.
bootstraps and all that
like I always say this debate will in the end come down to the problem of other minds.
Just another drive by comment.
peace