Information and Genetics

quote:

---

If this GA truly emulates evolution as you have clearly implied, then why do GAs only permit uphill movement on the fitness terrain?

---

Evolution DOES move uphill on the fitness terrain -- that
doesn't mean that things have to go from simple to complex.It means they have to tend toward the more suitable wrt the
selection criteria.If there was any merit to the line of reasoning you seem to be
implying (that evo. requires movement toward the complex)
then surely there would be no single-celled critters on earth.Haven't amoeba been evolving for billions of years too? (within
the evolutionary framework that is ... not in your opinion)[This message has been edited by Peter, 09-16-2003]

quote:

---

Here is the important question: If this GA truly emulates evolution as you have clearly implied, then why do GAs only permit uphill movement on the fitness terrain?

---

As I mentioned, commercial applications only allow "uphill" movement so that their result only ever gets better - results that go "downhill" are "killed". Of course, this comes at the expense of parallelism. If you want systems that allow the organisms to do whatever they want in competing with each other, we need to start discussing the more research-oriented applications. It's your call.

quote:

---

I never claimed "humans could have done it better'

---

Perhaps I'm misinterpretting you, but you stated:"As I mentioned in my prior post, GAs are really just trial&error experiments and only rarely would provide benefits based on what an intelligent source can glean from findings of the GA. "Naturally, commercial GAs are evoled for a "purpose", with a "goal" that determines their selection - that is what they're there for! However, there is no "gleaning" that needs to be done - you simply take whatever "organism" rated the best when you've run the GA for long enough. You don't have to review it, to pick out pieces of information or, anything. When we train our ANNs (neural nets, without backtracing (which itself is really just an additional layer on top of a standard GA)) here, you *don't need to do anything* apart from commit your results to CVS. Simply by coding the selective factor that you want to optimize, successive random changes in the net optimize it for that.

quote:

---

You know, GAs really boil down to souped up trial&error experimentation."

---

You know, evolution really boils down to souped up trial&error experimentation.Amazing how that works.

quote:

---

After the specified length of time has been reached, do engineers "take whatever it has at that point"?

---

YES. How many times do I have to tell you this? Let's try five more. YES. YES. YES. YES. YES. You don't need to "glean" a d*mn thing. Because it only ever gets better, you merely need to run it for a while, and you have a good result. If you run it for longer, you have a better result. That is because they lack different niches, all have the same selection factor, etc - as was stated, that is what commercial applications do.Look, if you don't believe someone who *does this in her place of work*, I don't know what will convince you. Would you like me to direct you to an ANN package that you can train yourself? Would that do it for you?

quote:

---

you are committed to naturalism so much you cannot see that GAs do not emulate what occurs in nature

---

Commercial GAs emulate *precisely* what evolutions accept as occuring in nature, with the following exceptions.1) There is no randomness in the selective factors. This is done to speed up the process.
2) All "organisms" are subject to the exact same selective factors. For obvious reasons.
3) Algorithms may or may not lack specific features sometimes found in life forms, such as the ability to "breed" organisms, to shift or replicate sections of "DNA", etc. This depends on what the algorithm is trying to accomplish.None of these apply to research systems. Research systems typically proceed through the exact same mechanisms that scientists attribute to evolution. The only thing that varies in the world is the "phyiscs" - i.e., instead of gravity, magnetic forces, quantum effects, etc, the laws of the universe may involve varying numbers of dimensions of polygonal blocks, "executable statements", or all sorts of other exotic possibilities.------------------
"Illuminant light,
illuminate me."

Sorry I'm just responding, but duty called and I couldn't find time to post.

quote:

---

Natural selection selects for fitness for the environment, not for better or bigger or fancier or more complex. That's why an evolutionist will always tell you that evolution doesn't require that things move "uphill", as you describe it.
But quite clearly the history of evolution is one of increasing complexity, and this is because genomes are accumulative. Much of the genetic history of an organism remains as a storehouse of knowledge upon which to draw. Depending upon the type of mutation, new mutations add to old, and the genomes can grow ever more complex with time.
And it is also because of increasing competition and the changing environment. Sometime increased complexity isn't called for, as when moths change color (back and forth, apparently) to take advantage of the predominate background shade. Sometimes increased complexity *is* called for, such as when a faster predator moves into an area thereby causing selection pressures for greater quickness or evasiveness among prey species, which often requires innovation which in turn often requires increased complexity.

---

The above is a good illustration of why evolution is not a theory. A theory should be testable and falsifiable. Evolution theory however is set up to handle all situations, accomodate all types of evidence.

quote:

---

I don't believe either Rei or myself are implying that GAs emulate biological evolution.

---

No, Rei has made it quite clear he beleives GAs emulate biological evolution.

quote:

---

It you're looking for accurate modeling of evolution in nature then you would do best to look to the field of biology, not to design engineering where GA's are but a design tool.

---

Thank you! Isn't this precisely what I have been saying?!

quote:

---

It's important not to lose the original point, that GA's are an excellent example of how the application of the principles behind evolution can provide unique and original solutions to complex problems.

---

The principiles being used are no different than ellaborate trial&error expirments scientists/engineers have been doing for centuries.

quote:

---

Fred: My claim has always been that GAs can only produce useful information within the presence of already-existing information, ie an intelligent sender.
Percy: And the result has always been that you've been unable to support this point.

---

Unable to support this point? I have supported it over and over again. All that is required is ONE example where a GA produced useful information without intelligent source to realize it, ie put it to use. If you can produce this, then you should publish your work and win a Nobel Prize!

quote:

---

If GA's already contain preconceived solutions then they aren't trial and error programs.

---

You can have a pre-conceived solution (loosely defined) and use trial & error to assist in reaching it. Heck, look at the very article Rei referred to regarding the Boeing engine: "Engeneous coded each design factor as a "digital chromosome," then mixed these chromosomes together to form trial designs."

quote:

---

Fred: I will ask you again. After the specified length of time has been reached, do engineers, "take whatever it has at that point"? Yes or no. If no, why? If yes, seek help soon.
Rei: YES. How many times do I have to tell you this? Let's try five more. YES. YES. YES. YES. YES.

---

Rei, this is so bizzare I cannot put it into words. I hope you are simply misunderstanding my argument, so I will try one last time. When engineers at GE set up the commercial GA to aid in producing a more efficient compressor, were they determined to use the result no matter what it produced? Yes or No.

quote:

---

Commercial GAs emulate *precisely* what evolutions accept as occuring in nature, with the following exceptions.
1) There is no randomness in the selective factors. This is done to speed up the process.
2) All "organisms" are subject to the exact same selective factors. For obvious reasons.
3) Algorithms may or may not lack specific features sometimes found in life forms, such as the ability to "breed" organisms, to shift or replicate sections of "DNA", etc. This depends on what the algorithm is trying to accomplish.

---

This is like saying, "John Madden Playstation emulates football *precisely*, except in the passing and running".
One of the two fundamental mechanisms of evolution is natural selection. GAs clearly do not emulate this mechanism becuase they employ truncation selection (where animals are ranked by fitness and selected at some threshold ). Truncation selection does not occur in nature, and no serious evolutionist will claim otherwise. So if it makes you happy feel free to continue to claim "it *precisely* matches evolution, except, uh, for 50% of the mechanisms involved."More fun examples:"Uh, an apple *precisely* matches a baseball, except in its content""You know, a pistol *precisely* matches a rifle, except in its length""I think my point *precisely* supports my argument, except for the half that doesn't"

quote:

---

When engineers at GE set up the commercial GA to aid in producing a more efficient compressor, were they determined to use the result no matter what it produced? Yes or No.

---

I was not there, but if they were sane, yes. I'm not understanding what is so hard about this to grasp for you. If you were playing blackjack, and you got your hand - and you had the option to trade in your hand as many times as you liked, with the guarantee that the hand that you got back would never be more than 21, and would always be at least as good as the hand that you traded in - would itA) be reasonable to accept whatever hand you've got after a number of tries, or
B) intensely scrutinize each hand to determine whether you like it or not?The new result of most commercial GAs is *always* as good or better than the previous version - as a consequence, there is no reason that you wouldn't like the later version more than an earlier version, and thus, no reason not to just let it run.

quote:

---

John Madden Playstation emulates football *precisely*, except in the passing and running

---

If that statement were true, it would demonstrate kicking 100% successfully. Commercial GAs demonstrate one part. Research GAs demonstrate all parts. How many times do I have to say this to you?Notice that you keep refusing to comment on research GAs. Could it be that you don't want to touch them?

quote:

---

GAs clearly do not emulate this mechanism because they employ truncation selection (where animals are ranked by fitness and selected at some threshold

---

Incorrect. Most commercial GAs use it because it is faster than non-truncation selection. They don't have to, to function - and research GAs generally don't use it.

quote:

---

No, Rei has made it quite clear he beleives GAs emulate biological evolution.

---

*Research* GAs do. Commercial GAs show applications for it, but employ speedups because we don't want to wait 4.5 billion years. What is so hard about this for you to grasp?

quote:

---

The principiles being used are no different than ellaborate trial&error expirments scientists/engineers have been doing for centuries.

---

You are not grasping the difference. Commercial GAs work on the previously best functioning "trials". In this manner, evolution - as evolutionists view it - IS trial and error.

quote:

---

Unable to support this point? I have supported it over and over again. All that is required is ONE example where a GA produced useful information without intelligent source to realize it, ie put it to use.

---

So, you don't think that learning how to trace out sections of the brain (what we do here) is useful? What do you think we have to do with them, talk to them really nicely to get them to run, or manipulate them? We run the program, giving it the ANN and the brain to trim. End of story. It produced something useful. All we had to do was give it the selection rules (in this case, letting there be a wide variety of sample brains that it is randomly assigned to trim from, and evaluating the accuracy of its trim).Now, before we go any further, you need to decide: Do you want to discuss commercial GAs, or research GAs? If you want to discuss commercial GAs (I.e., producing useful things for humanity), you need to accept the fact that engineers do use speedups (such as "truncation selection") for precisely that purpose - speeding things up. If you want to discuss research GAs which do not use this method of selection (and, as a consequence, usually aren't used in commercial applications because of their lower speed, but often produce fascinating results), then say so. Research GAs have even been used in studies of the spread of genes in viral and bacterial populations.------------------
"Illuminant light,
illuminate me."

quote:

---

--------------------------------------------------------------------------------
When engineers at GE set up the commercial GA to aid in producing a more efficient compressor, were they determined to use the result no matter what it produced? Yes or No.
--------------------------------------------------------------------------------

​

I was not there, but if they were sane, yes.

---

Rei, if you made such a statement in a design meeting at 99.999% of the engineering companies in the world, you would immediately be put on a short leash, if not canned. I strongly suspect you know this, and because of this intellectual dishonesty I cannot see any value in debating this with you any further.The reason folks that Rei refuses to give the sane answer of ‘No’ to my question above is because he knows it paints his position into a corner. By saying ‘No’ he would be admitting that intelligence is required to determine the usefulness of the commercial GA that GE employed. Perhaps Rei feared that evolution would crumble on this one point, and thus why he stubbornly refused to grant such a simple, straightforward fact. While it proved my point that the GA used by GE requires intelligence for the information it produces to be realized (they also require intelligence to design them and set them up, but that is beside the point), it doesn’t by itself refute evolution, since the evolutionist can still respond that the particular GA isn’t how evolution works (as Percy did, kind of). This would then lead the debate to research GAs, but I see no value whatsoever hashing this out with Rei since he is unable to even grant the most obvious, simplest of truths regarding commercial GAs.If some think I am being hard on Rei, just imagine if I was debating Randy on the global flood and Randy was trying to formulate a particular argument that was based on the truth statement that water runs downhill. If for some reason I refused to grant that water runs downhill (perhaps seeing that it aids the particular point Randy is seeking) then Randy would be completely justified in putting me on his ignore list, because it would be a big waste of his time to carry the debate any further.For the record, my position on research GAs is:1) Like commercial GAs, they cannot produce useful information outside the presence of already existing information (ie intelligence).
2) They do not emulate evolution in nature because they typically invoke truncation selection (the GAs I have looked at are Tierra and Avida)
3) They assume positive mutations that increase the genetic information such that the wild-type is more viable than the parent type in a normal environment (little if any real-life examples of this type of mutation exist)
4) The assumed positive mutation rate is unrealistically too high

THere is a lot I could take issue with in your post, but lets start with a simple objective issue.What do you mean when you say that Avida uses "truncation selection" and what is your evidence that this is the case ?

quote:

---

The reason folks that Rei refuses to give the sane answer of 'No' to my question is that he knows it paints his position into a corner

---

The reason that Rei refuses to give the ridiculous answer of 'No' to your question is that she knows that the answer is 'Yes'. Did you even look at the analogy that I presented? If you had such a choice in blackjack, would you not pass your cards in and let it shuffle them around for quite a while using those rules, and accept whatever is passed back? Because if you wouldn't, you're crazy.

quote:

---

GE requires intelligence for the information it produces to be realized

---

Unless you're talking about people taking the plans that it spits back and building them in the real world instead of in a simulation, you're completely wrong.There is something wrong here. I am not understanding at all what part of this you're not understanding. Why is this such a difficult concept to you? GAs produce finished products - in this case, a finished design. What about this do you not understand?I completely see this as you trying to claim that, if water runs downhill, then there has to be an intelligence there for it to be at the bottom when it's been given enough time to run downhill. To put it into your analogy.

quote:

---

1) Like commercial GAs, they cannot produce useful information outside the presence of already existing information

---

Please explain, given the sample GAs that I presented (in the long-long-ago, before you took us off on this incredulous side tangent) are not producing useful information. Inventing the concept of parasitism, developing physical traits on land and on water and using them to defeat/destroy their competitors, and scores apon scores of other things - that's not "useful information" to the organisms? Tell me, is the ability to walk on land useful to you? Is the ability to steal your blood useful to a flea? How many examples do you want?

quote:

---

2) They do not emulate evolution in nature because they typically invoke truncation selection

---

Neither tierra nor avida evoke truncation selection. You must have incredibly poor reading ability. The only "influencing" tierra does to them is, for each organism in a block of memory space, gives it an amount of CPU time proportional to its size (effectively in parallel). The organisms are free to do whatever they want with that CPU. Avida works in a similar manner. BTW, you've only read about one particular type of GA (the virtual-machine GAs)? So, you haven't read about anything like polyworld or Karl Sims' works? Among others? I'm starting to wonder if you've read more than 2 pages about the topic you're arguing about - a topic that you're arguing about with someone who actually uses them at work.

quote:

---

They assume positive mutations that increase the genetic information such that the wild-type is more viable than the parent type in a normal environment

---

What on earth are you talking about? Non-"truncation selection" systems don't "evaluate" the fitness of the organisms. They just let them run on their own.

quote:

---

The assumed positive mutation rate is unrealisticly too high

---

Again, what on earth are you talking about? They don't define positive or negative mutations - they let them change randomly (and in some GAs, interbreed - in fact, in some tierra-like GAs, breeding wasn't defined, but a type of retroviral insertion style of genetic transfer developed on its own ).Research GAs are used in drug resistance studies, population studies... they've even been used to determine whether a type of "greedy" gene could be used to wipe out a particular species of mosquito from the Earth, although this particular case (unlike the others) hasn't been confirmed as accurate by testing (for obvious reasons ). Why is this so hard for you to understand (apart from your obvious lack of knowledge on the subject)?------------------
"Illuminant light,
illuminate me."

Good theories explain the *available* evidence, and evolution explains the *available* evidence. Evolution is most certainly falsifiable because there is all kinds of evidence that could potentially pop up that would falsify evolution, like mammals in Cambrian layers. If you have a complaint it is with nature for not providing you the evidence that would falsify evolution.This particular objection of yours raises a couple of concerns in my mind that perhaps you could help me set to rest. I thought the pattern of debate that emerges in discussions with you was a result of your lack of time, and that the way you returned to arguments that had been rebutted during your last visit was perfectly understandable because you had probably missed them when you had to leave, and that we need only repeat the rebuttal to get the discussion back on course. But I've seen some of the same rebutted arguments, such as you make here, from you many times now, and this explanation is becoming less credible to me, and it makes me wonder if perhaps you aren't taking advantage of your intermittent participation to ignore rebuttals so you can repeat the arguments in a manner and context that makes them appear as if no challenges had ever been made.Now I'm sure there's really no truth to these suspicions, but one must concede that the appearance *is* there, and so I wonder if perhaps we might resolve at least this among your favorite non-sequiturs (so it appears to me) one way or the other. In other words, I wonder if it might be possible for you to stick around long enough to get to the bottom of this one so that in future discussions either we no longer object to it because we know it's true, or you no longer bring it up because you know it's false. This is called progress. Rei has made it quite clear in his rebuttal to this very statement that you're wrong: The GA's we've been discussing are design tools that apply the same evolutionary principles that a research biologist would employ to simulate biological evolution in nature. Is that really "precisely what [you] have been saying?!" This reply simply repeats your premise. In order to actually address the issue you have to explain how GA's do not employ the principles of evolution. You have to show that they do not accurately model, in principle, reproduction, mutation and selection, which are all the components of descent with modification through natural selection. The task is actually so mundane that both Rei (http://www.daughtersoftiresias.org/progs/ev.c) and I (Ring Counter Evolution) have done this. Rei's model isn't even based on information, his genes are merely numerical indicators of fitness, and they change randomly. In my model the nodes of a PLA hold information, and they also change randomly. The people who wrote the software are not the source of change, a random number generator is.Recapitulating, evolution is understood to be an innovative process that can produce novel solutions. You dispute that a random process like mutation can produce anything new, and so evolutionists provide the example of GA's as illustration of just how easily the evolutionary process can innovate. You dispute that GA's are an accurate model of the evolutionary process, but have so far been unable to show in what way GA's fail to do this.--Percy

quote:

---

Rei, this is so bizzare I cannot put it into words. I hope you are simply misunderstanding my argument, so I will try one last time. When engineers at GE set up the commercial GA to aid in producing a more efficient compressor, were they determined to use the result no matter what it produced? Yes or No.

---

Do you actually think that a design team of any kind will be "determined to use the result no matter what"? I have preformed a lot of software design and we frequently throw out the result of the design efffort when we realize it isn't going to be suitable.Why do engineers produce prototypes? To find out what is wrong with a design and correct it for one thing.Do GA's produce usable designs? YES.

I am trying to sort out what Fred is having trouble with here. Fred, excuse me if I get this wrong but I am going to try to word your arguement for you. If I inadvertently setup a strawman while doing so please correct me.Fred is saying that there has to be intelligence put into (or built into ) the GA use for it to work. Exactly where this is he hasn't made clear. So let me try.For a GA to be run there has to be computer hardware and software that is build by an intelligence. In addtion, the GA itself is designed by someone. I think he is saying that this is where the intelligence is needed. Do I have this right Fred? Is there anywhere else that you think it is neeed?The analogy here is that to "run" evolutionary processes we need a physical universe and it's laws. For evolution to take place we have to have replicating things with selection taking place. These requirements are, to me analogous to the "intelligent" input to the GAs. If this is the case, Fred is agrguing against the GAs, by analogy, by arguing against the big bang and abiogenesis I think.Do I have this at all right, Fred.(edited to correct one of the spelling errors -- the others are left as an exercise for the reader)[This message has been edited by NosyNed, 09-28-2003][This message has been edited by NosyNed, 09-28-2003]

quote:

---

Do you actually think that a design team of any kind will be "determined to use the result no matter what"? I have preformed a lot of software design and we frequently throw out the result of the design efffort when we realize it isn't going to be suitable.

​

Why do engineers produce prototypes? To find out what is wrong with a design and correct it for one thing.

---

Perhaps there is a difference here because I work with GAs whose final products are software, not hardware. If the hardware people have properly designed their fitness algorithm, then it will work perfectly no matter when you take it. Any changes you do to it will just make it work. If I understand what you're suggesting, you're suggesting that the fitness algorithm doesn't function completely accurately - correct? You see, in software GAs like I work with, that's not a problem - perhaps it is in hardware, that you can't model everything completely and need to do "guesswork".Of course if the fitness algorithm isn't perfect, you're going to get an imperfect result. But, if your fitness algorithm is, then it is very unlikely that you can make the results of the GA better by hand-tweaking.------------------
"Illuminant light,
illuminate me."

If Fred is arguing against the intelligence that sets up the environment in which the GA runs, then he is arguing against Physics in the real-world counterpart. Because, you're basically coding a universe when you write a GA - you code the equivalent of what would be physical laws.It seemed, however, that he was arguing that there is some intelligence needed at the finished product. However, the GAs do complete with finished products; in our case with ANNs, we simply take the ann file that is produced and use it as input to the tracing program.------------------
"Illuminant light,
illuminate me."

You are right, I was thinking too conventionally. With the GAs used this way, of course, almost by definition, the output would be used.I don't know enough about them to have a reliable opinion I suppose. Is it not possible that after a solution is reached it may still be rejected? For example you might produce software that, while "better" than you might have designed, may be so complex (as evolved things in nature tend to be) that you might not want to maintain it?

Yeah, if you evolve anything in software, it is usually almost impossible to understand. If readability is a goal, then you probably shouldn't touch a GA with a 10 foot pole ------------------
"Illuminant light,
illuminate me."