A soapy sort of solution
August 21, 2006 6:39 PM   Subscribe

The design challenge. After some work on genetic algorithms was accused of having 'frontloaded' solutions, Dave Thomas issued the challenge - human design vs his mutating code to find Steiner Trees^. If the answer is frontloaded, it should be derivable. And now the results are in.
posted by Sparx (7 comments total) 1 user marked this as a favorite
 
Neat, but not very convincing (and I want to puke every time I hear the word ID). I don't think any creationist would deny that a computer algorithm could optimize math problems numerically faster than someone could do it empirically. I think they're main hangup is evolution coming up with new problems I.e, a bacterial flagellum may be optimized, but how could it have appeared and be set the challenge of moving the organism in the first place? (I know the rebuttals to this, I'm just pointing out the ideological flaws in the linked demonstration).
posted by Popular Ethics at 7:03 PM on August 21, 2006


While I find genetic programming extremely interesting, I find it a bit of a stretch to use it to justify biological evolution. Salesmen keep on travelling, despite the problem being NP-complete.
posted by GuyZero at 7:09 PM on August 21, 2006


Yup - it certainly won't change anyone's mind - (and, of course, humans did find the answer). I just liked it because it was a funky way of demonstrating visually a whole raft of different nerdcepts, from problem spaces to how GAs work to evolutionary niches and how the same problem can be solved in a variety of ways.

Of course, the backdrop of Culture War backbiting is just gravy.
posted by Sparx at 7:10 PM on August 21, 2006


Mefi.summon(bevets);
posted by bashos_frog at 7:17 PM on August 21, 2006 [1 favorite]


Sparx: Interesting stuff.

I guess I am having a difficult time with differentiating the predecided "weasel" target from Dave Thomas' GA program, which uses a "shortest length" criteria to decide fitness.

With sufficient generations, a randomized string can be "bred" to the "weasel" target. Likewise, with sufficient generations, Thomas' random strings will approach the optimum arrangement of shortest lengths — which by definition will very nearly approximate a Steiner tree.

Deciding ahead of time that "shortest length" decides reproductive fitness will, by virtue of the process, force selection of offspring that most closely approximate a Steiner tree.

Not sure that ID has anything to do with this thought. I find GA algorithms fascinating. Clearly it has use for quickly throwing away solutions that we can measure ahead of time as inefficient or of little utility.
posted by Blazecock Pileon at 7:29 PM on August 21, 2006


BP: The difference is with the "weasel" the target is known in advance - and so it is open to allegations that the answer is snuck in (which it has been, nobody is denying it, but ID folks tend to miss the point of the exercise) and therefore has an intelligence behind its result guiding it toward that specific target and that level of information complexity.

With the tree problem there are multiple solutions, including two optimal ones (owing to rotational symmetry) - so which one has been snuck in and whereabouts is it expressed? And those optimal solutions are not arrived at every time - 'near enough' solutions also arise a majority of the time.

The experiment thus shows that, in this model, the GA (modelling evolutionary processes of mutation and combination), with the fitness function of shortness (modelling the environment) is sufficient to produce a wide range of novel, complex (and indeed, some of them are irreduceably complex by the standard ID definition) solutions to the Steiner problem - even optimal solutions occasionally - without any reference to a specified target like "weasel" or a set of co-ordinates for the solution.

The environment part is the tricky one to get your head around - it's a form of evolutionary pressure - not a determination of shape - which is why you get the variety and novelty. By extension, if shortness was the single environmental selection factor acting on Earth (um - because we lived in shrinking caves or something), you'd get a wide variety of almost optimally short creatures which also exhibit irreduceable complexity.

The word 'solution' is also something of a misnomer in this case - it suggests that that's all we are after in the model. What's interesting really is the variety, complexity and novelty created by the process

Or that's my understanding of it.
posted by Sparx at 8:16 PM on August 21, 2006


Thanks for that great explanation, Sparx. I was myself thinking, "ok, I got it, great!" and then, "uh... huh?" The article was presented in an "it speaks for itself" way, but for those not in-the-know it was a little confusing. Your explanation made things abundantly clear.
posted by PigAlien at 10:11 AM on August 23, 2006


« Older sacreligious smackdown?   |   Look upon my frozen blue face, ye Mighty, and... Newer »


This thread has been archived and is closed to new comments