I really should be writing a paper about genetic algorithms and evolution right now. To explain why we are maiming existing programs, to make something that is not evolving a solution but a bunch of wacky things.
It's an old strategy, to simulate evolution in finding a solution. You just use a fitness function to evaluate your agents, and then breed the best ones from a generation.
Our idea was at first to do it the other way around, breed the worst few and see what happens. But before even testing this we realized it wouldn't be different from regular evolution, just with the fitnessfunction reversed. The aim is to give the individuals in our simulated world the chance to evolve the way they want.
Humans, for example, don't have the problem with survival as much as they used to anymore. At least not in this part of the world. But there are still rules they have to live by, telling them was NOT to do. For the rest they are pretty much free to choose who and what they want to be, usually only limited my their imagination (in my honest opinion).
We are going to try and implement this kind of freedom in an evolutionary algorithm. The concept is still distilling in our brains, but right now we think we might be able to make a program and breed not the top rated individuals, but random ones from the entire population.
If these virtual entities have no raison d'être and are just allowed and even encouraged to do random, innovative things: what will happen?! They will probably be flopping about the place, doing nothing... mmm doing nothing...

So now I have to look up some articles, and whip up a good story of 3 pages. Wish me luck!

posted by Marian @ 5:35 PM 3 comments

At 11:21 AM,  Anonymous said...

Hmm, like you said: evolution is no longer an issue for humans and they have moved on. This allows them to be free. Trying to capture that in a (random) evolutionary algorithm is just... paradoxal.