Design Challenge Results: "Evolution is Smarter than You Are"

Yeah, an evolutionary biologist working in intelligent design.... This problem was quite easily to do with an excel spread sheet, and realizing that the problem can be reduced to half a problem as the starting point is symmetric. It would become rapidly impossible to solve it in such a way when the symmetry is broken, or when the number of line elements is increasing.

Mr. Thomas!

Kudos to you! This series of posts, for me, was spellbinding. I'll admit that I await, with baited breath, the response by Salvador. I hereby predict that Salvador will attempt to dismantle your code and show it doesn't work when a particular part is removed. This, of course, will be his evidence that you frontloaded an answer. This avoids him actually having to prove (show) where the answer is and demonstrating that you have frontloaded the code.

He has already stated that the code wouldn't exist without you, therefore the natural process that it is designed to demonstrate must be intelligently designed. Even to a rookie like me, this argument is ridiculous. Your code attempts to copy an observed natural phenomena (RM +NS). According to his logic any program that simulates a observed natural phenomena, like the law of gravity, demonstrates the law of gravity is intelligently designed.

Well done!

This problem was quite easily to do with an excel spread sheet, and realizing that the problem can be reduced to half a problem as the starting point is symmetric.

NNNEEERRRDDDSSS!!!!!!!

I am not following. Isn't the exact Steiner solution asymmetric?

— ofro

Isn't the exact Steiner solution asymmetric?

— ofro

Well, I don't know if the true Steiner Tree is symmetric (though it's probably a safe assumption), but the symmetry (or lack thereof) of a given solution is (I think) a big clue as to whether or not that solution was designed by a person, as opposed to a GA.

Excellent posting. Sal is often quick to make assertions but often very slow to support them in any manner, exhibiting the typical scientifical vacuity of Intelligent Design.
As for Wells... Sigh, it seems his latest book is not much better than his ill fated Icons...

Well, I don't know if the true Steiner Tree is symmetric (though it's probably a safe assumption), but the symmetry (or lack thereof) of a given solution is (I think) a big clue as to whether or not that solution was designed by a person, as opposed to a GA.

I am not following. Isn't the exact Steiner solution asymmetric?

— ofro

Thanks, Dave, for having provided such an amusing way for wasting part of my weekend! :)

If anyone is curious, the random search program I wrote is at:

http://vote.grumpybumpers.com/tornado.txt

And a second script for turning its results into animated GIFs is at:

http://vote.grumpybumpers.com/gif-tornado.txt

I figured that as long as we were having a contest between evolution and design, it would be fun to submit a solution on behalf of the proverbial "Tornado in a Junkyard" which serves as a straw man of evolution in so many creationist arguments. So I wrote a program which was as close as I could figure to the "random search" algorithm which, in Dembski's writings about the "no free lunch" theorems, we are given the impression that evolution is incapable of performing better than at solving problems. The program repeatedly generates entirely random attempts at a Steiner solution using a method of construction and "fitness function" similar to Thomas' FORTRAN evolution program, keeping a running record of the best solution it has found so far as it goes. The program is also written in Perl, which itself tends to resemble a tornado in a junkyard.

You can see the progress of the program in this animated GIF, which has one frame for every time the random search found a new minimal proposed solution:

http://vote.grumpybumpers.com/tgif2.gif

The final frame shows the best proposed steiner solution the tornado program ever found. (I've still got it running, actually, but it never did find a solution better than that one.) This "solution" is 1674.3 units long, which incidentally is poorer performance than Thomas' evolutionary algorithm achieved either overall or for the specific graph topology that the tornado solution represents-- even though Thomas' genetic program only ran for about 200 generations to find that solution, and the tornado search has been running for over a full day. This should of course be no surprise to anyone with the slightest amount of familiarity with algorithms. Whether it would be a surprise to a reader of Dembski's "no free lunch" is another question :)

Dave, are you going to release the new code?

Also, do we get a chance to see soap film solutions?

It should be noted that I submitted both symmetric and asymmetric solutions. Symmetry is not a reliable indicator of intelligence in this case.

Dave, thanks for an intriguing and illuminating puzzle.

I found the tenth MacGyver early on when attempting to solve the problem by reduction - two four node trees. As a designer, I realized quickly that removing just one of the connections to either center point would result in a significant reduction in line length. I would expect evolution to find this solution earlier on, as it has no built in bias towards symmetry, as many of us humans do. Of course with pressure to reduce total line length, it would also be discarded soon after in favor of better answers. It did in fact lead me to the first MacGyver solution and ultimately to the actual answer.

Regarding symmetry, the actual Steiner solution can be flipped vertically about the horizontal centerline with no change in total line length. The diagram above seems to indicate differently. Likewise, the first MacGyver can be flipped about either the horizontal centerline or the vertical centerline yielding 4 equivalent solutions. I wonder if in Dave's genetic code they could interbreed or would they be an example of speciation?

It is point symmetry, also resultng in rotational symmetry. What is important, is that if you know that single point (in the middle between the mid points), you only have to calculate the distances of one halve to get to the answer.

Dave, are you going to release the new code?

Also, do we get a chance to see soap film solutions?

It should be noted that I submitted both symmetric and asymmetric solutions. Symmetry is not a reliable indicator of intelligence in this case.

Dave,

Congratulations! An absolutely superb and challenging experiment. And funny too. ;-)

Marshall

Dave Thomas: This was well done, indeed! I encourage making comparisons between GA and RA(Random Algorithm) by the number of generations multiplied by the number of individuals versus the number of random attempts for a solution. Not only does this seem 'fair', but it is enough to demonstration how quickly GA develops a 'good enough' solution in comparison to RA's stumbling in the dark...

Dave, would you please provide a very careful, concise and clear list of exactly which point or points in the evolution/creation debate this demonstration was created to address?

I was browsing the UD weblog and from comments I saw there you run the very real danger that your GA will run into the same misinterpretation/ misrepresentation problems as Dawkins' "Methinks it is Like a Weasel" algorithm: namely, that people will attempt to claim that it is supposed to prove points, or simulate situations, that you never intended.

The problem with this is:

1) It is easy for opponents to obscure the points you are making by bringing in irrelevant scenarios.
2) If an opponent can show that your GA is simplistic or unrealistic for their artificial scenario, then they can dismiss all of your demonstration without addressing the specific points you make (essentially using your demonstration as their straw man).

I am not sure I understand why it is fallacious to build the "Answer" into the fitness function. If you're level of "fitness" is defined as the extant at which you match the phrase "WILLIAM DEMBSKI IS A BOOB", then isn't it appropriate to save that phrase as a constant somewhere and compare strings at each generation in order to judge your level of fitness?

Frank, the idea is to show that a dumb algorithm can come up with a specific, hard-to-find solution. IDers say that Dawkins's WEASEL program is cheating because the answer is built into the program, which of course it is. Dave's program addresses that complaint.

Weka, Frank, this post is a continuation of a thread started earlier this summer, Target? TARGET? We don't need no stinkin' Target!

The purpose of the "Target" post was simply to show that Genetic Algorithms can produce elaborate designs without anything like the actual form of those designs being fed into the fitness test.

As explained there, IDers and creationists alike have all latched onto Dawkins' "Weasel" algorithm as evidence that all genetic algorithms need to have the final answer snuck into the fitness test. You'll find several references of exactly that happening in July's "Target" post. It's not just creationists, but ID stalwarts like Meyer and Dembski too.

So the "strawman" I'm going after is the undeserved "guilt by association" linkage of any and all GA's with target-specific simulations like "Weasel."

The second major aim of this work is to directly address the ID claims that any novelty produced in a Genetic Algorithm is "frontloaded" in the algorithm itself. If such "frontloading" was being done in my Steiner GA, then anyone looking at the source code should be able to determine where the algorithm will go. Not one person from the ID community stepped up to demonstrate this.

And the third aim was to provide a Test for the detection of "real" versus "apparent" CSI. If Intelligent Design Theorists can indeed identify when something has or has not been "designed," as they claim they can, then they should get cracking to decide which of the two networks shown above (just below the fold) was produced by a human Intelligent Designer (a college computer sci prof, by the way), and which was produced by simulated mutation and natural selection on a population of strings over many generations.

I hope that helps.

Dave

Well done, Dave.

But I think the IDer's complaint that the Steiner solution is prebuilt is to some extent valid. There exists a unique global minimum (but not necessarily a unique configuration of the tree) for the problem. Thus, the GA will always tend towards that solution.

May I propose a defeater to the "front-loading" argument? Allow your metric to vary subtly per generation of your GA. One simple way to extend it is to vary the exponent of your distance metric according to, say, a Gaussian distribution, centered about 2. In other words |x-y| = sum (xi-yi)^p, where p is a random variable with mean of 2. You can further complicate the problem by introducing uncertainty in the locations of the terminal points.

Now, there is no fixed landscape for which a unique fitness value describes all global minima. Would the GA still tend towards the exact SMT? I suspect it would... or perhaps additional behaviors might manifest, such as the GA alternating between configurations that tend to be robust for extremal cases.

Whoo! I found the first MacGyver, but I never sent it in because I wasn't able to figure out the length exactly.

Dave, I am a software engineer and am very interested in playing with the code. You posted SteinerGAdlg.cpp code, but without the header files and resources it is not easy to recreate, and being the lazy guy that I am, I am not willing to spend the few days it would take to reverse engineer the code. Is there any way for me to get the complete code? Either way, thanks for listening.

Dave, I am a software engineer and am very interested in playing with the code. You posted SteinerGAdlg.cpp code, but without the header files and resources it is not easy to recreate, and being the lazy guy that I am, I am not willing to spend the few days it would take to reverse engineer the code. Is there any way for me to get the complete code? Either way, thanks for listening.

Please forgive me. I just finished reading the "Target? TARGET? We don't need no stinkin' target!" post, including the explanations you quoted from John Bracht, and while I appreciate that you are going through all the trouble to jump through the hoops that the ID'ists are throwing at you (I'm truly impressed by all of this), I still cannot understand what their complaint is.

It seems to me like they want to remove anything resembling a fitness function from the program. But if you do that, then all you have is randomness, without anything simulating natural selection. Am I erecting a straw man here, or are they that stupid? Of course, in order to simulate Evolution, you MUST have some function that is capable of SELECTING those organisms which are the "fittest." Complaining about the 5-point Steiner program because it, "[selects] for networks that (1) connect all five points, and (2) have shortest path-length" seems, to me, like complaining about Alaska because it selects for organisms that (1) are white, and (2) have fur and layers of fat to keep them warm.

I don't even fully understand why targeted GA's are a problem, even after Dawkin's explanation of it. I'm not saying they're NOT a problem; just that I can't grasp why.

Hmmm.
No trackback, no triumphalist crowing?

Where's the brilliant engineering squad at UD?

Anything to say, Sal? Anything at all? Or are your thoughts no more profound than the sound I hear, which is the proverbial crickets chirping?

I'll bet your pocket protectors are overstuffed. Goddamned nerds.

But I think the IDer's complaint that the Steiner solution is prebuilt is to some extent valid. There exists a unique global minimum (but not necessarily a unique configuration of the tree) for the problem. Thus, the GA will always tend towards that solution.

— Lurker

Dave, this was a thoroughly brilliant investigation, and I learned a lot about a subject altogether foreign to me. My sister is a mathematician, and she's told me a bit about soap films, but I think this whole exercise was very beneficial. Also, I'm surprised that William Dembski didn't participate, maybe he anticipated being beaten by evolution??

Anyhow, there is one flaw I think in how this was carried out - and that is, what to think of the results. Correct me if I'm wrong, but did Salvador Cordova in any way state an opinion about the fairness of the project? When James Randi debunks charlatans, he gets them to say on record how fair they thought the test was. Most of the time, they are fully confident that it was fair, and then when Randi reveals that they did no better than chance, they are caught. They grumble and make up excuses afterwards, but they have been caught on tape saying that they were 100% sure that it was fair, and sure that they were successful.

So now that the project is over, Salvador Cordova will be spinning like a flagellum, making excuses and denying the meaningfulness of the contest. Did Sal make any admission or statement that could show that he was confident about it, and sure that design would beat out evolution?

Seriously, we need to treat these people like psychics... they will spin and spin until they barf - but they will never admit that any test was fair after they hear the results.

And let's not forget that back in the real world, "solutions" may not need to be optimal. I don't think 2% makes too much of a difference to survival. Afterall, we're all MacGuyvered solutions ourselves.

I posted this series of panda posts on metafilter. Looking back I could have described the post sequence better in the front page description - though the format demands brevity and interestingness so I played upp the challenge, possibly to its detriment. But within the thread I flatter myself that I described the essential basics of the results in four shortparagraphs.

I like the writing thing, and I was following this experiment in all its aspects with interest from it's start - as I said within, it's a very visually descriptive example of how a raft of concepts work, so I can only applaud Mr Thomas for its design and execution. But I am interested in improvement, so if I got any nuances or facts wrong I would like to apologise and be instructed. Suggestions welcome.

"Could you show that the GA will also *structurally* tend towards the formal solution? "

Well, actually you could. You can think of Dave's GA as a local optimization routine coupled with a metaheuristic global optimizer, such as simulated annealing. There are proofs for SA that as the number of trials approaches infinity the probability of achieving the global minimum is 1. All Dave had to do was wait long enough for the space to be sufficiently well sampled.

As for the local optimizer, consider Dave's Transcribe() function was designed with a structure that would permit fairly smooth transitions from parent to child. From my initial scan of the code, each coordinate was integer, given a max of 999, and so the GA was merely crawling along a reduced search space of finite grid points. Now, look at the Mutate() function, and you can see that he jury-rigged the mutation rates on his organism's DNA to minimize mutations in # of steiner points. Further, it is restricted to a preset m_varbnodes variable. How would a GA know a priori to limit m_varbnodes to m_fixdnodes - 2? This is strictly a mathematical result -- a MacGyver heuristic, if you will.

The net effect was that the complexity of the problem was severly restricted from its initial presentation. This design is, I suspect, so that Dave could get an answer within his lifetime, and punish Sal in a timely manner. But it doesn't escape the fact that the GA had a helping hand in a number of ways.

Having said all of that (I feel like I am doing all of Sal's work for him), let me say that I personally do not buy into the jury-rigging/ front-loading argument about Darwinian evolution. Existence proofs of a solution does nothing to demonstrate the actual construction of the solution itself. Design is about the latter, and the fact that Sal failed to construct something a GA could is for me quite telling. To generalize, it may well be that our Universe permits the existence of many mechanical solutions to dilemmas facing organisms on the planet Earth. But the fact that some of these solutions do exist, despite an absence of a handbook that describes the construction of these solutions, requires a person to take a theory like Darwinian evolution seriously.

"Could you show that the GA will also *structurally* tend towards the formal solution? "

Well, actually you could. You can think of Dave's GA as a local optimization routine coupled with a metaheuristic global optimizer, such as simulated annealing. There are proofs for SA that as the number of trials approaches infinity the probability of achieving the global minimum is 1. All Dave had to do was wait long enough for the space to be sufficiently well sampled.

Wouldn't the fact that the formal solution is in a significant minority of achieved results disprove that? I mean, you could think of it like that, but you'd be wrong.

"Wouldn't the fact that the formal solution is in a significant minority of achieved results disprove that? I mean, you could think of it like that, but you'd be wrong."

How would being a minority prove that the exact SMT would not necessarily be found later?

But I think the IDer's complaint that the Steiner solution is prebuilt is to some extent valid. There exists a unique global minimum (but not necessarily a unique configuration of the tree) for the problem. Thus, the GA will always tend towards that solution.

The highest possible poker hand, a royal straight, never appeared. If "the solution" was hard coded into the selection routine, this should have been it, but it never appeared in all the times I ran the simulation. This makes sense, of course. Building pairs allows step by step improvement, but in poker a four-card straight isn't particularly better than a three-card straight. You either have a full five-card straight or you don't.

— Michael Suttkus, II

I'd think it was because the pathways found would preclude other solutions. Once the shorter fitness function has done its work you are left with a subset of naturally selectable alternatives. With "naturally selectable" being equivalent to the fitness function for the purposes of the model. MacGyvers won't mutate into "better"/shorter Macgyvers.

I feel this very strongly - though admittedly I am not able to say categorically until the code is released.

until the code is released

You can think of Dave's GA as a local optimization routine coupled with a metaheuristic global optimizer, such as simulated annealing.

— Lurker

Now, look at the Mutate() function, and you can see that he jury-rigged the mutation rates on his organism's DNA to minimize mutations in # of steiner points.

Further, it is restricted to a preset m_varbnodes variable.

"Simply stating the optimization criterion --- shortest network, highest hand, or accurate image formation --- isn't sufficient to reach it, there has to be an evolutionary path, as there is for Steiner networks and eyes but not for poker hands."

I agree. But, the existence of an evolutionary path depends on the designed landscape used in the GA. Consider if in searching for the royal flush, you constrain the problem such that all hands initially start with K,Q,J of the same suit. The search problem is considerably simpler than if it started with some random set of hands. Also, note what Suttkus wrote,

"but it never appeared in all the times I ran the simulation."

This is also something else I pointed out. A solution exists. Dave had to constrain the problem so that he could achieve the SMT within his lifetime. Combinatorial optimization problems are typically finite (perhaps inpractically so, but that is a different problem altogether).

"If the optimal Steiner network is implicit in Dave Thomas's GA, then the eye is implicit in biological evolution."

This is most definitely false. DA's GA is explicitly teleological. It aimed to solve a problem, quite well posed by the metric and the search mechanism. The notion of biological organisms solving "problems" is on the other hand quite problematic, for the simple reason that there is no single specified problem an organism has to solve. Problems change all the time in biological evolution. This is why I proposed a stochastic modification to DA's GA as a defeater to the "front-loading" argument.

Dennett's notion of cranes and skyhooks can be applied here. There are all sorts of systems that would be more effective in terms of the "produce viable offspring" global optimization function (top-down skyhook), but never occur because they can't be reached via bottom-up cranes. A global optimization function does not guarantee that an optimal solution will be reached, and there is no meaningful sense in which the solution is implicit in the function, any more than the solution to the three-body problem is implicit in its statement, or the proof of Fermat's Last Theorem is implicit in its statement. One can say that the Steiner network solution is implicit in the program as a whole, but only if one can say that extant biological systems are implicit in the mechanisms of biological evolution.

Let me suggest something to think about: Everyone here is so focused on the _instance_ of a solution to the SMT problem. Try focusing instead on the emergence of higher order strategies developed by GAs, instead. For instance, can a GA evolve the heuristic of minimizing trees by parts? Could a GA evolve robustness?

"A global optimization function does not guarantee that an optimal solution will be reached, and there is no meaningful sense in which the solution is implicit in the function, any more than the solution to the three-body problem is implicit in its statement, or the proof of Fermat's Last Theorem is implicit in its statement."

A global optimization strategy can be demonstrated to locate a global minimum, provided it exists. A simple naive search is a global optimization strategy. What you are talking about is the notion of tractability. I have never argued that SMT problems need be solvable in a given amount of time. In fact, I believe the problem is known to be NP-complete.

But I agree with you it is meaningless to suggest that by asking the question, the answer is already implicitly there. That is in the domain of existence proofs. As I said, constructions (i.e. designs) are much more difficult problems. And it is telling that trial-and-error approaches beat out a lot of supposedly "intelligent" designers.

A solution exists.

"If the optimal Steiner network is implicit in Dave Thomas's GA, then the eye is implicit in biological evolution." This is most definitely false. DA's GA is explicitly teleological. It aimed to solve a problem quite well posed by the metric and the search mechanism.

"The only teleological factor is Dave's intent of finding a solution by running the algorithm; the GA algorithm itself is not the sort of thing to have intentions."

Strawman. I explicitly said "DA's GA" is teleological.

Popper's Ghost, if you don't mind, I'm going to cut this short. I'd rather not have any discussions with you, as you are from my experience as abrasive as the Christian fundamentalists we have to deal with routinely. You don't think. You merely react.

It has been --- see http://www.pandasthumb.org/archives/2006/08/desi...

Yes, but http://www.pandasthumb.org/archives/2006/08/design_challeng_1.html#comment-121421

that'll take some time.

That said - my first inclination when Dave posted his original code was to take it and run it through as many generations as my CPU could stand to see if it did end up in a final Steiner solution regardless. It doesn't. The pathways, once traversed, seem somewhat unique - backtracking doesn't occur both because of the limitations imposed on mutable qualities and the fitness function (I've been wrong before so it was good to test it). Such was my guess, but it's appropriate - a penguin will not evolve into a polar bear as a result of mutation, even though their fitness functions are all about survival in the cold (to simplify it horrendously).

It's getting late and I can already see ways I have not been clear. And on preview heavier math theory guns than I are loading up (I have a potato gun). See you in the morning - but I would like to thank Mr Thomas for his fascinating experiment and the excellent commentary it has engendered.

Note that there are simple mechanical theorem provers than can prove any true statement in first order propositional logic. It would be bizarre to say that all these proofs are implicit in such a theorem prover, or that such a theorem solver is "teleological".

"The only teleological factor is Dave's intent of finding a solution by running the algorithm; the GA algorithm itself is not the sort of thing to have intentions." Strawman. I explicitly said "DA's GA" is teleological.

Popper's Ghost, if you don't mind, I'm going to cut this short. I'd rather not have any discussions with you, as you are from my experience as abrasive as the Christian fundamentalists we have to deal with routinely. You don't think. You merely react.

Yes, but http://www.pandasthumb.org/archives/2006/08/desi... that'll take some time.

Such was my guess, but it's appropriate - a penguin will not evolve into a polar bear as a result of mutation, even though their fitness functions are all about survival in the cold (to simplify it horrendously).

The answer is that both were designed.

In both the GA and...uh..."classical geometry" approaches, you had an intelligent designer with some kind of goal in mind. The designer then developed a method that would reach this goal.

Further the "classical math" method is just as "evolutionary" as the GA method. Analytical geometry and calculus did not simply fall out of the sky. They are the product of centuries upon centuries of development of mathematics that goes back to Euclid and before. Without Euclid, Pythagoras, Arabic work in algebra and numerals, and the Cartesian plane, you don't have calculus.

Hence, all you've really done is compared methods of design. Is it better to design a recursive algorithm to solve a problem, or take some direct approach?

If you had designed a bad algorithm, you never would have had your tree emerge. As anyone with experience in this field knows, stopping your "for" loop one iteration short, screwing up an array index, or accidentally overwriting the data a pointer is addressing will destroy the results of an algorithm. You need a non-designed algorithm with no particular goal to really make this the challenge you want it to be. So in this particular case, you have failed to demonstrate anything other than that there is more than one way to design something.

"And the third aim was to provide a Test for the detection of "real" versus "apparent" CSI. If Intelligent Design Theorists can indeed identify when something has or has not been "designed," as they claim they can, then they should get cracking to decide which of the two networks shown above (just below the fold) was produced by a human Intelligent Designer (a college computer sci prof, by the way), and which was produced by simulated mutation and natural selection on a population of strings over many generations."

This is a good challenge test of design detection. I think the design detection angle of ID is lacking, so far. But guys like me, who are agnostic about a designer, are open to ID for a different reason: MET/NDE along with OOL hypotheses have not sufficiently demonstrated the ability to generate the entities that actually exist. Your tests have been no help. I would like to see a test where the fitness algorithms themselves are randomly generated, and capable of generating interesting virtual "entities".

You need a non-designed algorithm with no particular goal to really make this the challenge you want it to be.

"The answer is that both were designed."
"there is more than one way to design something."

Bingo. Design need not be the philosophically-laden term that it is. In other words, there is no need to follow the IDists into making the illogical leap from design to some amorphous, immaterial mind.

If anything, Dave set himself up for failure with the second challenge. It was already a poignant demonstration that "intelligence" operates no more efficiently than trial-and-error (and indeed, one may even say that intelligence _operates_ on trial-and-error strategies). To go further is to sink into the philosophical quagmire where IDists want to live.

So in this particular case, you have failed to demonstrate anything other than that there is more than one way to design something.

Having though about this a bit, your challenge shows that an "entity" could be designed or not, and that the detection of design would be impossible. This should be pretty obvious, and requires no computer to demonstrate. I could chuck a piece of wood into a chipper and have it produce small wood chunks. Then I could whittle a wood chunk to look similar to one of the randomly carved, chipper 'generated' ones, in such a way that nobody could tell which one was designed and which one wasn't.

How does your challenge relate to the bacteria flagellum, for example, with regards to the question of design? Nothing, as far as I can tell, since nobody knows if the flagellum was designed or not. (Duh.) In other words, your challenge doesn't help to answer whether a particular entity IS designed or not. If you set up a test where the fitness algorithms were randomly generated as well as the mutations, and could generate something on the order of a flagellum, that would be impressive.

If anything, Dave set himself up for failure with the second challenge.

How does your challenge relate to the bacteria flagellum, for example, with regards to the question of design? Nothing, as far as I can tell, since nobody knows if the flagellum was designed or not. (Duh.) In other words, your challenge doesn't help to answer whether a particular entity IS designed or not.

"If you set up a test where the fitness algorithms were randomly generated as well as the mutations, and could generate something on the order of a flagellum, that would be impressive."

And if we made the demand on IDers to produce God, would you similarly comply?

Well, I don't know if the true Steiner Tree is symmetric (though it's probably a safe assumption), but the symmetry (or lack thereof) of a given solution is (I think) a big clue as to whether or not that solution was designed by a person, as opposed to a GA.

How does your challenge relate to the bacteria flagellum, for example, with regards to the question of design? Nothing, as far as I can tell, since nobody knows if the flagellum was designed or not. (Duh.) In other words, your challenge doesn't help to answer whether a particular entity IS designed or not. If you set up a test where the fitness algorithms were randomly generated as well as the mutations, and could generate something on the order of a flagellum, that would be impressive.

— mike1962

It seems to me like they want to remove anything resembling a fitness function from the program. But if you do that, then all you have is randomness, without anything simulating natural selection. Am I erecting a straw man here, or are they that stupid? Of course, in order to simulate Evolution, you MUST have some function that is capable of SELECTING those organisms which are the "fittest." Complaining about the 5-point Steiner program because it, "[selects] for networks that (1) connect all five points, and (2) have shortest path-length" seems, to me, like complaining about Alaska because it selects for organisms that (1) are white, and (2) have fur and layers of fat to keep them warm.

For instance, can a GA evolve the heuristic of minimizing trees by parts? Could a GA evolve robustness?

The pathways, once traversed, seem somewhat unique - backtracking doesn't occur both because of the limitations imposed on mutable qualities and the fitness function (I've been wrong before so it was good to test it). Such was my guess, but it's appropriate - a penguin will not evolve into a polar bear as a result of mutation, even though their fitness functions are all about survival in the cold (to simplify it horrendously).

I would like to see a test where the fitness algorithms themselves are randomly generated, and capable of generating interesting virtual "entities".

If you set up a test where the fitness algorithms were randomly generated as well as the mutations, and could generate something on the order of a flagellum, that would be impressive.

"And if we made the demand on IDers to produce God, would you similarly comply?"

I'm not an IDer.

RBH: I was referring to Avida fitness functions, not the Challenge. Sorry I didn't make that clear.

Nevertheless, do you think an IDer should produce God at the demand of their critics?

"The burden is on the IDists to prove that it was (intelligently) designed, since they make they claim that it must be (intelligently) designed."

I agree.

"The only burden on evolutionists is to show that it need not have been (intelligently) designed."

If by "evolutionist" you mean a pure non-designed evolution (MET/NDE), then I also agree. However, this hasn't been done. It is encumbent on anyone making positive claims to justify those claims. Neither side has done so, IMO. And maybe nobody cares what I think, but I sure care what I think.

Dave, you can't really use this example to attack CSI - the problem is that there is a single correct mathematically-defined solution. That both a human and a GA come up with pretty much the correct solution is hardly surprising.

Nevertheless, do you think an IDer should produce God at the demand of their critics?

Nevertheless, do you think an IDer should produce God at the demand of their critics?

"Of course, in order to simulate Evolution, you MUST have some function that is capable of SELECTING those organisms which are the "fittest." Complaining about the 5-point Steiner program because it, "[selects] for networks that (1) connect all five points, and (2) have shortest path-length" seems, to me, like complaining about Alaska because it selects for organisms that (1) are white, and (2) have fur and layers of fat to keep them warm."

Which is kind of the point, for guys like me: what is the origin of fitness/selection criteria that let to the bio-entities we actually see on this planet? Designed or not? Particularly with regards to the origin of life.

I don't think Dave's Challenge demonstrates anything useful in determining that, nor does it help us understand how things like the vaunted flagellum arose, etc. Didn't everyone already know that GAs can find goals theyre intended to find? I've been using GAs for 20 years to do that.

When somebody can find some randomly produced fitness algorithms that use random inputs to generate some complex virtual machines like the flagellum, please let me know. The war will be over.

"The only burden on evolutionists is to show that it need not have been (intelligently) designed." If by "evolutionist" you mean a pure non-designed evolution (MET/NDE), then I also agree. However, this hasn't been done.

"Of course it has been shown that it need not have been designed --- Dave Thomas did it here in a small domain, and the theory of evolution does it for biological diversity --- "a pure non-designed evolution" is a possible explanation of biological diversity."

I disagree. Nobody has ever demonstrated that it's even possible for a flagellum to have been assembled the way MET asserts. For that to be true, at least one precise (down the molecule) developmental pathway would have to be demonstrated. For all we know, the laws of nature might actually *forbid* any putative pathway that could be devised due to various chemical interactions. It's an open question.

what is the origin of fitness/selection criteria that let to the bio-entities we actually see on this planet? Designed or not?

— mike1962

Particularly with regards to the origin of life.

Which is kind of the point, for guys like me: what is the origin of fitness/selection criteria that let to the bio-entities we actually see on this planet? Designed or not?

Didn't everyone already know that GAs can find goals theyre intended to find?

When somebody can find some randomly produced fitness algorithms that use random inputs to generate some complex virtual machines like the flagellum, please let me know. The war will be over.

"When somebody can find some randomly produced fitness algorithms that use random inputs to generate some complex virtual machines like the flagellum, please let me know. The war will be over."

Mike, the problem is that the type of system you are talking about is a system that reproduces successfully depending on it's ability to reproduce (as opposed to systems like the challenge, which reproduce successfully based on their ability to solve a problem).

Such systems can (and have) been created. We see virtual life that develop a whole host of reproductive strategies such as viral infection of hosts, producing multiple offsping in one go, etc. The problem is that they are virtual life. They do nothing except exist in a computer world. To win the war, we would have to show something a bit more concrete, ie things that are useful in the real-world, where we can actually touch them. That's why solutions such as the Challenge are used to illustrate the point.

"The question is incoherent; the fitness/selection criterion is reproductive success"

That's part of it. There is also environmental pressures.

And of course, reproductive succuss certainly did not generate the initial reproductive machine in the first place.

"which is not the sort of thing that can be designed or not designed."

How so?

Anyway, my challenge stands. Mine is a lot most difficult than Dave Thomas's. Let's see if anyone can pull it off. It would be wonderful if someone could.

Nobody has ever demonstrated that it's even possible for a flagellum to have been assembled the way MET asserts. For that to be true, at least one precise (down the molecule) developmental pathway would have to be demonstrated.

For all we know, the laws of nature might actually *forbid* any putative pathway that could be devised due to various chemical interactions. It's an open question.

Mike1962: "When somebody can find some randomly produced fitness algorithms that use random inputs to generate some complex virtual machines like the flagellum, please let me know. The war will be over."

Demallien: "Mike, the problem is that the type of system you are talking about...is a system that reproduces successfully depending on it's ability to reproduce (as opposed to systems like the challenge, which reproduce successfully based on their ability to solve a problem). Such systems can (and have) been created."

I explicitly specified "virtual machines like the flagellum." Are you saying that something like that has been generated by randomly produced fitness algorithms acting on random inputs? If so, I'd like to see them.

mike1962 wrote Which is kind of the point, for guys like me: what is the origin of fitness/selection criteria that let to the bio-entities we actually see on this planet? Designed or not? Particularly with regards to the origin of life.

There is also environmental pressures.

And of course, reproductive succuss certainly did not generate the initial reproductive machine in the first place.

"which is not the sort of thing that can be designed or not designed." How so?

Anyway, my challenge stands.

It's self-evident that "reproductive success" as a fitness/selection criterion is not the sort of thing that can be designed. If you think not, please explain what it could mean to design it.

Stevaroni: "Frankly, I'm baffled. You're telling me that you don't see selection all around us?"

Of course I do. Did I seem to imply that I don't? Variation and selection can certainly account for much, if not all, of what we see on the scale of wombats and gazelles. And we can artificially select for our own ends. However, when it gets down to the small scale specific protein machines such as the flagellum, nobody has proffered any precise, definite specified chemical pathways for it's development, that is demonstrably possible simply given *chemical* considerations, forgetting the irreducibility business. Just assuming that it's possible certainly doesn't convince me. It's an open question.

Mike1962: "There is also environmental pressures."

Popper: "Uh, no, those are not "fitness/selection criteria". And do you really want to debate whether environmental pressures are designed? Of course some are --- by humans. Anything beyond that is an irrelevant metaphysical question."

Hmm. I'll have to digest this puzzling statement and get back to you.

"It works for wombats, it works for bacteria. It even works for single molecules (your much-debated origin of life - No, I can't prove how it happened, but that's besides the point, I can show that if it happened, yes, we could get here from there)."

OK, then, are you going to be the one to provide a precise chemical pathway to the flagellum? Not necessarily the real one, but one that would work, given what we know about chemical processes? That would be great! :)

Just assuming that it's possible certainly doesn't convince me.

However, when it gets down to the small scale specific protein machines such as the flagellum, nobody has proffered any precise, definite specified chemical pathways for it's development, that is demonstrably possible simply given *chemical* considerations, forgetting the irreducibility business. Just assuming that it's possible certainly doesn't convince me. It's an open question.

Popper: "It's not an open question that there's some causal explanation for the existence of flagella --- not in science."

Where can I find a blow by blow evolutionary sequence of the flagellum, with no gaps whatsoever?

I explicitly specified "virtual machines like the flagellum." Are you saying that something like that has been generated by randomly produced fitness algorithms acting on random inputs? If so, I'd like to see them.

— mike1962

Popper: "It's not an open question that there's some causal explanation for the existence of flagella --- not in science." Where can I find a blow by blow evolutionary sequence of the flagellum, with no gaps whatsoever?

Popper: Look up "Occam's Razor" and "inference to the best explanation". "For all we know", there are "laws of nature" that forbid the migration of continents, but there's no evidence of of such laws. Really, learn something about science."

Occam's Razor is a good guiding principle, indeed. And I accept your explanation as the best possible explanations thus far if we disallow a priori any intelligent cause. There are still gaps in our knowledge with respect to the precise chemical pathways of things like the flagellum. I'm interested in someone filling these gaps. I would be willing to consider an intelligent source as the *only* way, if it could be determined that there is no chemical pathway *possible* without intelligent tweaking. Nevertheless, MET has not "explained" the flagellum to MY satisfcation. Forgive me if my standard of "explanation" is tougher than yours.

Nevertheless, MET has not "explained" the flagellum to MY satisfcation. Forgive me if my standard of "explanation" is tougher than yours.

Your standard is intellectually dishonest. I'll view it otherwise when you either provide a molecule-by-molecule explanation of the movement of the continents or assert that there could be laws of nature that prevent such movement, that there may be no causal explanation of how the continents got where they are, and that without without such a "blow by blow" explanation, tectonics does not explain how they got where they are to your satisfaction, etc. Of course, if you say that, I will consider you to be a loon.

popper: "Where can I find an exact chemical breakdown of the migration of continents? Demands for such sequences are intellectually dishonest, and your persistence in asking this question is trollery."

If you can't do it, fine.

As far continents go, we can detect processes right now that are *demonstrably* capable of producing what exists today. You, me, and the milkman can reproduce these sorts of productions on a small scale which can be extrapolated, without hopping any gaps in chemistry or physics. Can the same be done for the flagellum? I haven't seen it. If it exists I would seriously love to read about it. Selection pressures working on variations within a population cannot be scaled down to chemical and atomic levels without the induction of developmental gaps which are not analogous to the macro scale selection.

Insisting on a precise, detailed, and (most importantly, gap free) developmental pathway, is scientific question. I suppose it would be trollish for me to continue asking for something that obviously isn't going to be forthcoming. I got my answer, which is you can't provide an answer. And that's fine with me. If and when somebody can, please let me know. I would much appreciate it.

And I accept your explanation as the best possible explanations thus far if we disallow a priori any intelligent cause.

As far continents go, we can detect processes right now that are *demonstrably* capable of producing what exists today.

When somebody can find some randomly produced fitness algorithms that use random inputs to generate some complex virtual machines like the flagellum, please let me know. The war will be over.

— mike1962

You, me, and the milkman can reproduce these sorts of productions on a small scale which can be extrapolated, without hopping any gaps in chemistry or physics.

Can the same be done for the flagellum?

I haven't seen it.

the milkman can reproduce these sorts of productions on a small scale which can be extrapolated

I don't see the connection here.

Just a note from a lowly English teacher here--I've been reading the dialog between you all and Cordova, specifically the part where he disavows the Irreducible Complexity argument that begins with a completely constructed, functional entity and instead maintains that IDers "don't think that way." This is after declaring that the end result of your AI experiment is specified beforehand, therefore invalidating the experiment.

You'll notice that he also never specifies the way in which he wants us to think that IDers *do* think.

Incidentally, in Kitzmiller v Dover, you'll find that Michael Behe admits that the IC argument "doesn't address the central task of natural selection," so it's kind of amazing they're still using this argument.

It occurs to me that if you are reduced to saying the tautological "look, there is no empirical data that shows evolution to work any other way than has been empirically observed," then, although you have not lost the argument, the usefulness of continuing the debate is somewhat questionable.

It's pretty clear you are dealing with ideologs here. I can't say it better than seventypercent did yesterday in a Fark.com thread:

"The lessons of the Cold War taught us that a real commitment to science and technology can produce a generation that would end up winning that war for us. And now that we face a threat that many would consider more grave than the Soviet Union, that lesson has been forgotten by many Americans, who are now descending into a frenzied pit of religious fanaticism, not unlike their fundamentalist brethren on the other side of the globe -- the same people they claim we're at war with."

Yours,
DistendedPendulusFrenulum

We're back to the flagellum again.

Somehow I seem to recall that back around tehre were some references to a very recent paper tracking down the little spinner protein and showing it was a single mutation version of a cell wall protein (or something along those lines). Which is the kind of information that should have laid this moronic argument to rest.

I though it was posted here, but I can't find the key search term.

Anybody remember the link?

"I don't see the connection here. It isn't even true; the movement of the continents results from processes that are currently far more opaque to us than the processes of the cell."

Point taken. I'm not qualified to comment on geological topics such as that. My apologies.

Which is the kind of information that should have laid this moronic argument to rest.

I babbled... Somehow I seem to recall that back around tehre were some references

I'm not qualified to comment on geological topics such as that.

When somebody can find some randomly produced fitness algorithms that use random inputs to generate some complex virtual machines like the flagellum, please let me know. The war will be over.

— mike1962

I don't know about steveroni's cell wall mutation, but here's Nick Matzke's paper on "a model for the origin of the bacterial flagellum":

http://www.talkdesign.org/faqs/flagellum.html

Popper: "Then renounce plate tectonics as a satisfactory explanation of the locations of the continents and demonstrate that you are a loon, or admit that you're intellectually dishonest in not applying the same standards in both cases."

I have no reason to renounce plate tectonics. I don't know enough about the subject to make a judgement. Nor do I renounce evolution, per se. What I stand unconvinced of is MET to explain many things, like the flagellum. You've provided me with no information I can use, so it seems to be a waste of time chatting with you (singular.)

but here's Nick Matzke's paper on "a model for the origin of the bacterial flagellum

I have no reason to renounce plate tectonics.

You've provided me with no information I can use, so it seems to be a waste of time chatting with you (singular.)

Well, Cordova's response is up over at UD, and predictably its simply "this isn't evidence of evolution, because the computer program was created by human intelligence." Since I've already shown the error in this thinking, I won't do so again, but I will repeat that Sal's response is a joke.

I would like to add that NDE explains everything in general, because it explains nothing in particular. Impossible to falsify such a concept because it does not deal with particulars. Rather it is an ideological axiom. Axioms are not subject to falsification by definition. And this particular axiom is simply the mirror corollary of an a priori bias against design, supernatural OR natural. It's amazing to me how bad philosophy can pass as a "scientific" paradigm.

the ID argument is dependent upon setting a scientifically unreasonable burden of proof for the theory of evolution.

Popper: "Here's an indication of where mike1962 is coming from: ... It's amazing to me how bad philosophy can pass as a "scientific" paradigm."

Indeed. See my most recent post at UD.

The CSI camp and the MET both operate under a particular philosophy. Nothing wrong with that, and it's impossible to the avoid. The only thing is, one should not pretent to be merely methologically naturalistic when one is actually philsophical materialistic.

At any rate, I see your view regarding the flagellum comments above, as one that's ideologically based. Imagine chastisting someone for damanding evidence to fill in the gaps, and telling him he's being an unscientific, ignorant troll. That seems to me like ideology talking. Whatever it is, it doesn't help anyone learn or understand. Nothing has been gain in the trasactions, except that I'm even more confident that my challenges will go unanswered. Enough for this "troll" for now.

Scordova: "I would like to address your concerns regarding CSI."

Mike1962: No, I have not read No Free Lunch, or any of his books, except for excerpts, and also discussions related to it. But I do believe I understand the gist of it, however, from an information theory angle.

The problem with the CSI angle as I see it is that you still cannot detect design vs accident by mathematics alone if one's philosophy is wrongheaded. The concept of information originator and information receiver is irrelevant when one accepts the idea that an *infinite* variety of combinations are possible within the universe, or multiverse. The problem with the MET-only crowd is not their science, even though it is limited, and doesn't explain what they think it does. The problem with them is their philosophy, which is ridiculous. They embrace a methodological meterialism a priori (which usually turns out to be a philosophical materialism in disguise), and then skip along with the notion that matter *can* combine in just the right way to get the self-replicating process going, and then *can* reproduce and develop along the lines MET suggests. Talk about calling the blue sky a different color. They deny the sky is even possible, let alone blue. There's no use trying to convince someone that the sky is blue if they don't believe a sky is possible.

I am agnostic on design. I'm likewise agnostic about MET in many respects. MET fails to explain many aspects of the cellular world. And by explain, I mean explain in detail, of how something like the flagellum *could* (not *did) come together by unguided processes. I'm not saying it can't. But they can't show that it can. I'm the agnostic here. MET adherents are the ones making the positive claim about their theory. If it's true, fine. I'd love to see real proof.

Anyway, CSI doesn't seem to help a guy like me, since it's *conceivable* (if particulars are of no importance) that ANYTIHNG can happen given enough universes and enough time. And the anti-ID MET adherents are simply not on the same playing field philosophically with the pro-ID guys.

I see your view regarding the flagellum comments above, as one that's ideologically based.

Imagine chastisting someone for damanding evidence to fill in the gaps, and telling him he's being an unscientific, ignorant troll

Re "I'll bet your pocket protectors are overstuffed."

I resemble that remark!

I'm not saying it can't. But they can't show that it can.

I see your view regarding the flagellum comments above, as one that's ideologically based.

Now, look at the Mutate() function, and you can see that he jury-rigged the mutation rates on his organism's DNA to minimize mutations in # of steiner points. Further, it is restricted to a preset m_varbnodes variable. How would a GA know a priori to limit m_varbnodes to m_fixdnodes - 2?

Computers can compute certain things faster than us, that is why they exist to help us. For Thomas to argue that evolution is smarter than humans because computers can compute faster than humans is a non-sequitur. ... No backstage explaining to do. The little puzzle Thomas offered was amusing. It only shows that computers can compute faster than humans and that some at Pandas thumb can connect dots with lines better than others. What he didn't prove was mindless undesigned agents can create Genetic Algorithms from scratch.

To show that the processes of selection and mutation, in a reproducing population of organisms with variations in their genotypes (DNA) and corresponding phenotypes (e.g. networks), can lead to striking results featuring both Irreducible Complexity and Complex Specified Information (CSI), and all without front-loading the algorithm with any specifics regarding the actual problem solutions.

I have no reason to renounce plate tectonics. I don't know enough about the subject to make a judgement.

Nor do I renounce evolution, per se. What I stand unconvinced of is MET to explain many things, like the flagellum. You've provided me with no information I can use, so it seems to be a waste of time chatting with you (singular.)

who or what is "MET"?

Popper,

I've seen the Matzke treatment before, and read much of the stuff on his website. Sorry. Not impressive. May too many gaps left unexplained. I'll will be glad to hash through it, if you like.

"I believe this is ID/DI/UD speak for "materialistic evolutionary theory"."

No, rather Modern Evolutionary Theory

Dave:
Could you address all of Lurker's post #121509, please? I think he's making some rather relevant claims. Such as:

Why is the probability of a mutation different for each type of locus? Indeed, why not just use a "packed" sequence of bits (instead of a char sequence) and just invert one random bit? (You could use a 1024x1024 grid to avoid redundancy in coordinates.) Even if relatively frequent mutations in the "number of nodes" gene would cause a lot of excess garbage (hopeless candidates), you seem to be inviting criticism by biasing mutation frequencies.

Most importantly, what is your response to Lurker's claim that your code makes use of a global optimizer which kicks lineages off local optima?

Not impressive. May too many gaps left unexplained.

Dave: Could you address all of Lurker's post #121509, please? I think he's making some rather relevant claims. Such as: Why is the probability of a mutation different for each type of locus? Indeed, why not just use a "packed" sequence of bits (instead of a char sequence) and just invert one random bit? (You could use a 1024x1024 grid to avoid redundancy in coordinates.) Even if relatively frequent mutations in the "number of nodes" gene would cause a lot of excess garbage (hopeless candidates), you seem to be inviting criticism by biasing mutation frequencies. Most importantly, what is your response to Lurker's claim that your code makes use of a global optimizer which kicks lineages off local optima?

Yeah, Salvador is busy denying that it means anything, and he is relying on a dull tautology to do it. Basically, if you simulate evolution, then you cannot conclude that X evolved because you designed the experiment, thus, you can only conclude that it was designed. This is as dumb as saying that if you simulate lightning on a lab, then all lightning was caused by humans.

Like I said, these people need to be treated as psychics, deviners, dowsers, etc, and made to admit the fairness of a particular test BEFORE it was tested. Then afterward, we can call them on their dodging when they change their tune. Like when Behe said that a two-part IC system would be unlikely to evolve, and then dover hits, he admits that a 2-part system can be IC, under OATH. Then afterward, the evolution of a two-part system is elucidated, and Behe claims that THREE (!) parts are required for an IC system. Slam-dunk, clear move of the goalpost that everyone should be made aware of.

So that Salvador Cordova is dodging all of this at UD is no surprise, and he has taken the lame excuse that it is due to a computer's superior calculating skills. The computer is the environment, and he has a computer too, but the tools being tested were a genetic evolution algorhythm and his brain. His brain lost, and he'll be spinning for a while.

The point is, he will never change his position based upon evidence, just as a psychic wouldn't, but there are many people out there who would change their positions based upon a clear demonstration that he and other IDers admit to beforehand, and then you show them fleeing after the test is completed.

I would like to see a test where the fitness algorithms themselves are randomly generated, and capable of generating interesting virtual "entities".

In other words, your challenge doesn't help to answer whether a particular entity IS designed or not.

How does your challenge relate to the bacteria flagellum, for example, with regards to the question of design?

Anyway, my challenge stands. Mine is a lot most difficult than Dave Thomas's. Let's see if anyone can pull it off. It would be wonderful if someone could.

Woot, /me feels 1337 now.

For the record: the easiest way to figure out what the correct topology should be is to think soap bubbles. Note that if two soap films meet at an angle of less than 120 degrees, the system is not stable. The result is that the two films form a node that zips them together until the required angle is achieved.

In the above case, for example, the tenth MacGuyver is the obvious starting point. The two films will then start to zip in towards the middle, producing a rather odd MacGuyver - a version of the second MacGuyver with a diamond instead of a point in the middle. This diamond will naturally contract to a point, at which time the only way to proceed is by breaking symmetry. Which generates the correct solution.

The question I find really interesting is: can you ever have more than one stable solution for a given network? I'd imagine you can, but have yet to find an example.

earlier, Stevaroni was wondering about the thread discussing flagella that appeared on PT recently.

I think this is what you were thinking of Steve:

http://www.pandasthumb.org/archives/2006/08/friday_flagellu.html

[Dave Thomas:] I used computers simply to show that simulating evolutionary processes (reproduction, mutation, selection) over time can indeed produce precise solutions to difficult math problems, with no a priori knowledge of the correct answer(s).

The question I find really interesting is: can you ever have more than one stable solution for a given network? I'd imagine you can, but have yet to find an example.

What GA's do is real evolution. The information in them actually evolves, and it does so via the interplay of the three key processes that, when they occur together, inevitably produce evolutionary change: Replication, variation, and selection.

The notion of biological organisms solving "problems" is on the other hand quite problematic, for the simple reason that there is no single specified problem an organism has to solve. Problems change all the time in biological evolution.

Regarding the flagellum: It would certainly be nice to have a detailed history of the evolution of the flagellum. On the other hand, given the number of degrees of freedom of protein folding and interaction, resulting in an extremely high dimensionality of the "search space" traversed by evolution, I believe that we can with confidence make the following prediction from evolutionary theory:

The detailed, mutation-by-mutation pathway of evolution of the flagellum (and indeed, or most biological structures) will never be known

(Of course, that doesn't mean that we won't be able to come up with models, but they will always contain a substantial degree of speculation)

So does this meant that we must reject evolutionary theory because it refuses to provide us with an answer that we would like to have? If so, it will be in good company, because we will also have to discard such things as quantum physics, which perversely refuses to provide us with the exact trajectory of a photon through a slit. And Newtonian physics, which perversely refuses to provide us with an exact solution of the n-body problem, enabling us to accurately extrapolate planetary positions for any time.

But the scientific criterion for rejection of a theory is not whether it fails to answer a question that we would like it to answer---it is whether its predictions are contradicted by the results of experiment and observation. Of course, for a theory to even be in the running, it has to make such testable predictions. Here, of course, evolutionary theory has no problem, because it serves up a wealth of predictions, testable by methods ranging from computer simulations of genetic algorithms to genomic sequence comparisons (and so far has accumulated a remarkable track record of success).

ID on the other hand has a big problem. Predictions, after all, arise from the limitations of a model: the things that it cannot do. And since the ID crowd, apparently for political reasons, is unable to get specific about the nature---and particularly the limitations---of their hypothetical designer, they have been unable to make any testable predictions. The closest they can get to a prediction is "No pathway for evolution of the flagellum exists." Unfortunately, this kind of prediction is next to worthless, because to test it, they would have to show that they have examined every possible sequence of mutations from every possible set of protein precursors, and determined whether all of them are blocked by low-fitness "chasms" in the fitness landscape. Of course, they have no desire to take on this impossible task themselves. So instead, they turn to biologists, and say---"You must carry out the impossible studies to test our theory; if you can't disprove it, then our theory must be right."

But of course, biologists have no interest in trying to carry out this impossible study, because they have a rich theory that makes lots of testable predictions, and are busy testing those predictions and making discoveries. And all the ID guys can do is yell, "Hey, what about the flagellum?"

I wrote a program where digital creatures shaped like circles had evolving nueral network brains. The brains output was a force vector. The circles had a energy level that increased whe they touched food and decreased when they touched each other. A circle cloned its self when it reached a high energy level and died when it reached 0. The clone had a sightly mutated brain. There is no fitness function, only a digital world. No goal. Eventually, in about a half hour, there were circles that can track down food and avoid other circles. They actually could avoid other circles at high speed nd find food.It was very clear that their movemets were not at all random. The input to the neural net was x,y position of its' self, nearest food, and nearest circle. What, if any are the problems with this aproach (mike?)? What was amazng is how fast smary circles evolved. I typed this on a nintendo ds lite...sorry for bad and short setences

I encourage making comparisons between GA and RA(Random Algorithm) by the number of generations multiplied by the number of individuals versus the number of random attempts for a solution. Not only does this seem 'fair', but it is enough to demonstration how quickly GA develops a 'good enough' solution in comparison to RA's stumbling in the dark...

— David B. Benson

I typed this on a nintendo ds lite

— mempko

DaveThomas wrote:
A neat sci-fi story exists where a 3D person is plucked into the 4th dimension, turned around, and placed back into the 3D world to find himself to be his mirror image.
---
I remember reading this story :) The only problem with it is that such a person wouldn't be unable to survive for long, for a very trivial reason (can you find it)?

I remember reading this story :) The only problem with it is that such a person wouldn't be unable to survive for long, for a very trivial reason (can you find it)?

Marek 14,
Presumably because his amino acids would all be the wrong handedness. He would starve, being unable to digest and process food.

Paul

Damn server time outs.

If I may, a few people have read Mike####'s demands for *a* point by point evolution of the flagella for *the* point by point evolution that actually produced it. This isn't actually fair. Mike's demand is possible, while the other is something we will probably never achieve without time travel.

Doesn't change the fact that Mike's demand is unreasonable. The weight of evidence that life did evolve is astounding; demanding that one particularly bit of it be demonstrated to that degree is just be persnickety.

For several years, I had a screensaver that evolved organisms out of colored lines I found online. It was fun to watch. Green lines produced energy. Red and blue lines consumed energy to maintain. Red lines stole energy if they touched a green line in another organism, but not if they touched a blue. (I don't remember how red/red interacted.) You could vary the amounts of energy each line type used or produced to create different selection environments.

It was fun to watch organisms evolve. You could see ecosystems try to develop with primarily green producers, trying to protect themselves with a minimal shell of blue, etc. Good fun. And no selection function to sneak in design! (Not that they really ended up "designed".)

I imported the opera browser. Its good enough to surf and post comments, access gmail, etc. I enjoy it very much

If I may, a few people have read Mike####'s demands for *a* point by point evolution of the flagella for *the* point by point evolution that actually produced it. This isn't actually fair. Mike's demand is possible, while the other is something we will probably never achieve without time travel.

"DaveThomas wrote: A neat sci-fi story exists where a 3D person is plucked into the 4th dimension, turned around, and placed back into the 3D world to find himself to be his mirror image. ---- I remember reading this story :) The only problem with it is that such a person wouldn't be unable to survive for long, for a very trivial reason (can you find it)?

— Marek 14

"If I may exercise a pet peeve of mine, genetic algorithms do not "simulate" evolutionary processes, they *are* evolutionary processes.

Similarly, a computer does not "simulate" addition and other numeric operations, it actually performs actual addition, etc.

Information processes, like evolution and addition, are transformations that information undergoes. When information in a computer (or anywhere else) is subjected to some form of selection, variation, and replication, the transformations that occur in the information is real honest-to-goodness evolution, not a simulation thereof."

I think such a view, especially abstracted with the concept of information, invites all sorts of problems. Information isn't sufficient to describe addition or evolution. And it is problematic to discuss even the numeric operations in a computer.

Granted, you can set up natural addition and subtraction with the help of bitoperations on the positions on vectors or similar algorithms, simulating moving beans between containers. But when you look at computer numerics you find that they are essentially binary 1's (unsigned) or 2's (signed) complement addition. It barely captures integer addition/subtraction on low numbers.

When you get into numeric on reals (or decimal approximations thereof) it doesn't capture for example superposition of field strengths when adding charges, due to roundoff errors. So you have to argue about a limited "ghost in the machine here", soemwhat removed from idealised mathematical addition operations or most real additive processes.

But considering this I think that if anything could be said to really happen inside a computer instead of being modelled by its algorithms, evolution of the GA type is probably the *best* example. The objects mapped inside are actually operated on by a selective process.

If I may, a few people have read Mike####'s demands for *a* point by point evolution of the flagella for *the* point by point evolution that actually produced it.

A question here - would this mirror image thing also reverse the electric and magnetic properties of the subatomic particles? ;)

P.S. The "spin" of subatomic particles is a fundamental physical attribute, distinct from the fundamental physical attribute of charge. Electrons and positrons both have spin of 1/2.

For the 6-node problem discussed here, there are indeed two unique Steiner solutions. Each is rotationally symmetric (like a playing card), but each has different handedness.

Re "Electrons don't left and right sides, so mirror-imaging them wouldn't change anything ... certainly not their charge."

Not right/left, true. But I don't know if they might have geometric distinctions (from positrons) in some other dimension (not one of the four we can detect directly). I don't know enough physics to judge that offhand.

Henry

Henry:

"But I don't know if they might have geometric distinctions (from positrons) in some other dimension (not one of the four we can detect directly)."

According to string theory they have, and some of them may be large (brane worlds). But I doubt a rotation affects this.

Spin is basically a topological effect with no left/right problem, if I understand correctly. (See Popper's comment.) But there are P (and CP) violations, so parity has a small effect from geometry. A parity transformation is flipping all coordinates and that is AFAIK possible by rotations in a higher dimension. But it wasn't done here.

Actually, I wonder if he would starve. Are proteins from right-handed aminoacids just undigestable or are they downright toxic? If so, even touching a biological material (like an apple) could have some unpleasant effects...

Of course, the story was written by H.G. Wells, who could know nothing about the assymetry of living world - I think.

They are certainly indigestible. They may also be anything else. A stereoisomer can have any chemical properties any other molecule can, relatively unrelated (biologically, their general properties will be similar).

Thalidomide, for instance, is a powerful, safe anti-nausea drug, but it's stereoisomer causes nasty birth defects.

Maybe slightly off-topic here, but since we're talking about genetic algorithms (apologies if posted before):

http://www.demo.cs.brandeis.edu/golem/press/UsNews.pdf

...[Brandeis University computer scientist Jordan Pollack] and mechanical engineer Hod Lipson run the Golem Project, a colony of machines that evolve and give birth to other machines without human guidance."

Wonder what a cool name for their next generation project would be...hmm...oh, how about "SkyNet"? Hey, that's a cool name!

I'm not real familiar with Steiner trees, so my question is how was the exact solution determined? Is there some type of general way to do it? How can you be sure that your best solution actually was the best solution, and that with a little more playing you couldn't get something with an even shorter length than your 1586.53?

How can you be sure that your best solution actually was the best solution, and that with a little more playing you couldn't get something with an even shorter length than your 1586.53?

Hi Dave,

Do you know of any place to find sample mutation algorithms for GAs?

It looks like your code hardcodes probabilities for certain mutations, then mutates one(?) locus at a time.

I'm trying to put together a simple GA-based program that takes a set of input bits and manipulates them into a set of (user-defined target) output bits using only NAND operators.

Not sure exactly how the "DNA" is going to look yet, but if a mutation can only replace one locus at a time, it's going to be impossible for the GA to co-opt "groups" of operations that might serve a generally useful purpose [e.g., (a NAND a) NAND (b NAND b), equivalent to (a OR b)]. Unless I allow the reproduction mechanism to mutate, which is a bit more work than I want to do right now...

Anyway...any resources you can think of that would help me with this?

I'm trying to put together a simple GA-based program that takes a set of input bits and manipulates them into a set of (user-defined target) output bits using only NAND operators.

Dave Thomas wrote: In other words, consilience works for me.

Dave,

You inspired me to develop a genetic algorithm engine to solve the Steiner Problem you described. I went through your three articles, making copious notes, then spent an afternoon building the core functionality. I tried it out on the six node challenge problem and consistently got solutions with a segment length more than 400 longer than the Steiner solution.

Over the course of the past week I've spent an hour or two every day adding to my engine, implementing more sophisticated crossover, trying alternative selection techniques, changing the encoding of variable node positions from standard binary to Gray code, and running test after test. Still, my evolved solutions remained stubbornly 30% longer than yours.

I was getting concerned. Had you, no doubt unknowingly, introduced information about the solution in your encoding of the problem? Was this a potential source of embarrassment for The Panda's Thumb? Did I have an ethical obligation to publish my results?

Before contacting you with these questions, I went back once more to the original articles. It turns out that I'd written down the wrong value in one section of my notes. I have been hacking furiously in an attempt to get the fitness of my solution to the six node challenge problem down to that of the five node example you'd provided. In fact, my engine had been generating Steiner solutions and MacGyvers all along, even in its most naive implementation.

The code is available if anyone is interested in having a look. I used a simple binary string rather than your more complex encoding. My first version allowed the number of variable nodes to mutate, but as you also reported, this leads to local optima with no variable nodes.

Thanks for the interesting puzzle.

Regards,

Patrick

As regards "global" versus "local," I don't have the foggiest notion of what you're referring to. In the GA, an organism is represented by a DNA string, period.