help with neural nets?

i have just found breve and it is a dream come true. really. :p

as a modest starter-project i would love to try giving agents three  
lights, red green and blue, and three corresponding photocells, to  
see if i can get a colour-language to evolve. if we add a green food  
and red threat for example, would warning and/or lying behaviours  
emerge? what use might be found for the unused colourspaces like purple?

i am assuming that this is beyond the sort of twitch-based  
instinctual "intelligence" of the Walkers, and i would need a three  
or more layer neural net [input, middle[s], out]. [discuss.] if i  
follow correctly, the idea is that you feed a list of inputs/sensors  
into the net class, and it will then follow dynamic "habits" to  
provide outputs/actuator responses?

do you still need the busful of monkeys generational approach or can  
one candidate on its own evolve at a reasonable rate?

i guess my question really is, where can i find tutorials or more  
sample steve code than comes bundled, and particularly a tutorial  
neural net project to show how to hook one up? i am not having much  
luck on google, partly because some thoughtless people have used  
"breve" as a common word in their language.

i can think of a number of famous dead guys who would just wet  
themselves over tools like these, and it's great to see these techs  
evolving into mass games like spore and gta. meanwhile the pentagon  
say robot armies within ten years, but then we get virtual girl/ 
boyfriends to make up for it, and all the smart dust you can inhale.  
exciting times to have the experience of "being alive".