Enviroment/Organisms

[yrwyddfa]

Now, if you started everyone off with the basics, and allowed us to optimize, we would both learn, and be able to compete. Kind of start everyone off at basically the same level. It may be biased, but not more than other contests.

Some people will be able to code simple neural nets, some bayesian intelligence (remember this is on a time basis: if you can't complete the task quick enough then you lose your move) The simplest, but most effective, should stand a better chance of winning.

[yrwyddfa]

Simple rules produce very very complex results.

[yrwyddfa]

So that's enough from me. Who's in, then?

[Shaggy Hiker]

Actually, I'm already working on something that is very similar to this. My problem is the GUI (I SUCK as an artist).

The basic program is a lake with 8 organisms. The lowest level organisms feed on three base resource pools that are determined by a chaotic equation. The reason for this is that a paer showed that three populations of plankton in a water body are guaranteed to be in a chaotic equilibrium. However, this may not be important. The original idea behind this program was a population simulation, but the behavioral aspects have come to dominate, and the population sim is now just a basis for encounters.

Originally, I had each organism determining one of 16 different canned behaviors based on 7 yes/no questions. Seven may not seem like many, but this was back in the bad old days of memory models (under DOS), and my design had a limit right around there. Eventually, I scrapped that in favor of the current idea. I had over 1,000 genes/species determining behavior, and behavior was driven at a VERY low level. Unfortunately, somebody broke into my house and stole the computer that had the newest version of that design on it. That left me in a funk for almost a year. I have now redesigned the genes (though not yet coded the new design) in an even more elaborate scheme using as many as 2,000 genes/species.

I am currently hung up on my inability to do graphical programming. I am such a horrid artist that drawing on the screen is mostly trial and error. I knew this shortcoming in advance, so why I thought I would be able to do this is beyond me, but I'm slowly working on it.

I firmly believe that GA's, if given sufficient complexity, can solve most any problem. Having worked with a few of them, I would even go so far as to speculate that the technique I am using would allow for sufficient intelligence to evolve such that a robot could be dropped into a random kitchen, and given the existence of the proper items somewhere in the kitchen, it would be able to create a cheese omelette. The major stumbling block to this would be visual recognition, which is itself a MAJOR issue.

However, I have been sidetracked again. One of the other programs lost in the theft was the newest version of another odd GA program I wrote that evolves predictive equations. This was written in VB6, and it belongs in .NET, so, now that the new version was stolen, rather than fix the existing version, I am re-writing it the way it should be done.

Having said all that, I would sum it up like so: I think that idea is good enough that I have been pursuing it as a hobby for a few years now.

[yrwyddfa]

Actually, I'm already working on something that is very similar to this. My problem is the GUI (I SUCK as an artist).

I know the feeling.

The basic program is a lake with 8 organisms. The lowest level organisms feed on three base resource pools that are determined by a chaotic equation.

Do you use the May's logistic equation? Have you read the softening of the conjecture?

The reason for this is that a paer showed that three populations of plankton in a water body are guaranteed to be in a chaotic equilibrium. However, this may not be important. The original idea behind this program was a population simulation, but the behavioral aspects have come to dominate, and the population sim is now just a basis for encounters.

Are you saying that emergent behaviour rather than environmental factors determine the system, now?

I firmly believe that GA's, if given sufficient complexity, can solve most any problem. Having worked with a few of them, I would even go so far as to speculate that the technique I am using would allow for sufficient intelligence to evolve such that a robot could be dropped into a random kitchen, and given the existence of the proper items somewhere in the kitchen, it would be able to create a cheese omelette. The major stumbling block to this would be visual recognition, which is itself a MAJOR issue.

I think that some recent work in evolving neural nets, and genetic programming (as oppose to genetic algorithms) will make this a reality once the computing power is there. I think that Sony's(?) new 'Cell' processor is a massive step in the right direction.

However, I have been sidetracked again. One of the other programs lost in the theft was the newest version of another odd GA program I wrote that evolves predictive equations.

Cool: genetic programming, eh!