Over here I talked about a design challenge posed by Dave Thomas of Panda’s Thumb regarding Steiner problems and solutions generated by genetic algorithms (GA). The main purpose of the challenge was to demonstrate the criticisms offered by intelligent design (ID) advocates were baseless. Well the results are in, folks! How did the ID folks do? How did yours truly do? Venture forth past the fold to find out!

First off, a couple early results, one designed, one “grown” with a GA, were so similar it led Dave to issue a new challenge. Since it is aimed specifically at the ID crowd, I won’t go into details. He asks them to demonstrate how these two nearly identical solution can be analyzed with ID tools to determine which is designed and which merely appears designed. The curious can review the article themselves for more information.

Now, the solution. Well the ID crowd did, in fact generate correct solutions, of which there are two. But only the GA generated both solutions. And independant home-brew GA got one of them. Five people solved it qualitatively, which I tried but failed at. So the conclusion? GAs perform just as well as people here. The ID nay-sayer, however, failed to demonstrate that the program had been “front-loaded” with the solution. This is no surprise.

But the neat thing is the GA generated many solutions that were nearly as good as the optimal solution, and no human designers came up with all of them. These he calls “MacGyver” solutions because they use unexpected elements in a unique and innovative way. So the first one is only 0.6% longer than the optimum and it was found by six people. The second MacGyver was 2.1% longer and was found by nine people… including me, baby, yeaaaahh!!! Check it out! My name’s right there (whoops, Deely, now my name’s leaked, too). Actually the 10th MacGyver was the very first thing I put down on paper. I generated six solutions in my qualitative search, but #10 was the first thing that popped into my head since one half of it is the Steiner network for four points. Since I beat that total length quickly, I didn’t bother to submit it. I guess I should have because my name would have been published twice!

So the conclusion reached is as follows. If you don’t understand the ID lingo, let me know and I’ll try to find links. But your best bet is to use the Intarweb – duh.

“…anyone who comes across one of the Steiner or MacGyver designs shown in the Design Room would probably infer design, especially if they knew of the effort required to derive such designs. But, given that these “designs” can also be found simply by breeding herds of alphanumeric strings, it becomes clear that evolutionary processes can also produce the appearance of design.

To those Designers who found “MacGyvers” instead of the official Solution – pat yourself on the back. You have helped to show that the McGyver solutions are Complex Specified Information in their own right.”

And on a side note, I submitted another design. While he did not list it, with a total distance of 1648.52, it has only three MacGyver’s that perform better than it. I think Dave didn’t include it because it was the same basic geometry as the other solution (MacGyver #2). But, out of the three characteristic paths present in that shape two in the second design were shorter than the corresponding paths in the “real” MacGyver. See for yourself:

steiner1.gif steiner2.gif

Dave e-mailed me back and said he had not seen the second solution before.