Re: New AI Prize

> An AI can only be a complex as its environment

You can imagine all manner of complicated intelligence going into playing
the game of tictactoe or chess, even though the "environment" of the game
isn't all that complex (compared to the real world).  If a bunch of
game-playing entities are put into a world where their opponents form the
"environment," and they get into a sort of arms race around playing better
and better at the game, then you can imagine sophisticated behaviors will
arise.

Think about Karl Sims' cube game.  Simulated bots were fighting to control
a cube.  The game itself was impoverished.  Yet, because of the
competitive aspect and the arms race which arose, the robots and their
strategies became remarkably complicated.

As a follow up to that work, consider some of these projects: 
http://demo.cs.brandeis.edu/pr/robotics.html

All of those involved evolution in an impoverished world; e.g. the bots 
lived on flat plane with no obstacles, and were incented to move forward.  
No complicated tasks.  Nevertheless they developed interesting and novel 
ways of moving.  Many of those walking robots have been transferred to the 
real world as you'll see on that page.

Designating certain parts as "environment" is an artificial way of
dividing things up, I think.  What seems to be more important are the
types of interactions which can occur and how entities can respond to
those interactions.  It appears that artificial worlds which resemble the
real world "naturally" provide the right kinds of interactions, but they
do not exhaust all the possibilities.

Anthony