Emergent behaviour in simple robots

[deMangler]

Some people might find this interesting.

http://www.plosbiology.org/article/info%3Adoi%2F10.1371%2Fjournal.pbio.1000292

The interesting thing I find here is not the way the robots evolve, which is the main point of the experiment, but that apparently motivated and complex, even co-operative behaviour emerges from groups of units with simple instruction sets.
The fact that this behaviour has emerged through random selection seems to hint at how difficult it is to code for programmers who have a 'top down' mindset.
I am wondering if part of this approach could be used to develop behaviours for units in AI War, not to interfere with the stateless nature of the in-game AI, but to run out-of-game 'evolution sessions' from which code might be incorporated into the main game.
Anyway, got me thinking.....

[x4000]

Very cool.  Evolutionary AI is not really an area of AI that holds a ton of interest for me directly at the moment, as the pitfalls of units learning the wrong behaviors can be rather severe, and the whole thing takes a ton of tuning, evaluation, and CPU time to create a database for, but it's always interesting to hear what others are doing with their systems.

[Echo35]

Evolutionary AI is really creepy. I read an article not too long ago about robots that were designed to look for "food" and signal other robots when they found it, and after a few trials, their AI code was "Bred" together and a new generation started. After a few hundred generations, the robots started lying to each other and hoarding the "food". It was fascinating.

[Spikey00]

Quite a fascinating subject, isn't it?

[Mánagarmr]

Evolutionary AI is really creepy. I read an article not too long ago about robots that were designed to look for "food" and signal other robots when they found it, and after a few trials, their AI code was "Bred" together and a new generation started. After a few hundred generations, the robots started lying to each other and hoarding the "food". It was fascinating.

If you have a link to this study, please share!

[deMangler]

Evolutionary AI is really creepy. I read an article not too long ago about robots that were designed to look for "food" and signal other robots when they found it, and after a few trials, their AI code was "Bred" together and a new generation started. After a few hundred generations, the robots started lying to each other and hoarding the "food". It was fascinating.

If you have a link to this study, please share!

http://www.pnas.org/content/106/37/15786.abstract

There ya go!

dM

<edit>Dunno if that is the study Echo35 was referring to, but it is the one I remember.... <edit>

[Mánagarmr]

Thanks for the link!

[deMangler]

Evolutionary AI is really creepy. I read an article not too long ago about robots that were designed to look for "food" and signal other robots when they found it, and after a few trials, their AI code was "Bred" together and a new generation started. After a few hundred generations, the robots started lying to each other and hoarding the "food". It was fascinating.

If you have a link to this study, please share!

I just noticed that the other link I posted does not point to the full text.
Sorry about that.

Here is a link to the full text.

http://www.pnas.org/content/106/37/15786.full

also some links to other interesting articles about this in popular science.

http://www.wired.com/wiredscience/2009/08/real-life-decepticons-robots-learn-to-cheat/
http://www.physorg.com/news172304708.html
http://www.hplusmagazine.com/articles/ai/darwins-robots