Side/Comments on What if Design is Compatible with Evolution

[Rope (Rope Kojonen)]

@Roy

Btw, Rope’s statement is incorrect, but not for the reason you give. The expertise needed to get GAs to produce useful results (in little time) isn’t expertise in design, but expertise in the relevant field that allows one to narrow down the GA’s solution scope without excluding successful outcomes.

For example, when I implemented a GA for finding multi-dimensional Steiner trees, I was able to improve the GA’s output not by changing the design of the GA, or the fitness function, but by changing the nature of the ‘organisms’ to not have direct connections between the fixed nodes. This reduced the GA’s tendency to get stuck on solutions that were suboptimal in some problems, without eliminating them as solutions in the cases where they weren’t suboptimal.

Maybe there is indeed some misunderstanding or a language / terminology barrier here, because I would understand what you describe yourself as doing as “design” in the sense of purposefully arranging stuff. I would see the nature of the organisms as part of the GA’s design. Perhaps we are talking past each other here.

@Dan_Eastwood

@Roy, playing by my own rules, here are my thoughts, please quote or steal as you see fit.

@Rope

Jeavons probably does not mention those examples, because his article is a response to / evaluation of my book in which I do also point out examples of genetic algorithms producing designs that were unanticipated. So there was no need to talk about examples that we both agree about, given that this does not change the conclusions in any way - you still need plenty of design expertise to get such algorithms to produce interesting results. I hope you can give Jeavons the benefit of the doubt, since he is an expert in this field. See his details here .

That statement is incorrect. GAs always produce results, but often not the result desired by the programmer. Expertise is needed to corral them into a particular task, but this presumes a programmer.

Why does this show the statement to be incorrect in your opinion? Not that I did not write about results generally, but about “interesting results”. I guess it depends on what you mean by “interesting”. I will grant that if you set the bar for “interesting” sufficiently low then you can disagree.