When I was in High School my teacher had put on an educational movie about computer AI. It was from the eighties so the picture quality was horrible, but I learned something very interesting. Life forms on the computer. The movie had shown thinking, learning life forms. Here's what I remember on the PBS video I seen. They had shown the computer scientist playing with the organisms. Giving them appendages and they were learning to move. I distinctly remember one of the scientist giving the life form a sort of fish body and they were practically swimming. The life forms were very basic. Just various rectangular cubes connecting to do what it was learning. When I was young I was nieve but now I'm older and thinking back upon what I have learned it high school and what I know now, I find that quite interesting. Please someone care to comment? I'm sure there are other past students whom remember that presentation. Imagine if we could make a game with these sorts of organisms as the AI in the game. Constantly learning how you play, and getting smarter on how to beat you.

sounds like its mostly hype and for show

the description doesnt indicate any special AI method used
and it doesnt sound like anything requiring any large stretch over current methods, so i doubt its much of a breakthrough

but hey, if that gets your rocks off
check this: http://demo.cs.brandeis.edu/golem/
they have their computer 'evolve' robot designs, then they go ahead and build them in the real world
Oh Noes! its Terminator Judgement Day again!



sorry dude, but after youve taken classes in AI and Computability Theory, youll be pretty jaded too when you hear people get excited over AI that 'thinks' or 'lives'

I remember that show as well. They were using genetic algorithms and neural networks, IIRC.

I don't think the AI should be trying to beat the player, it should be creating a fun challenge for him. It's like wrestling with a younger sibling. You aren't trying to beat him, you are just trying to keep it interesting.

I've seen what you're talking about on the net, but it wasn't quite as you describe: they didn't design the organisms at all. They set up a simple physics system, and created a genetic system so that the body shape and neural net that ran it could be put into a genetic programming system, then they did standard evolution on everything using various statistics as the fitness function: for example, for swimming they might let each bot swim for 5 seconds and use distance from original position as the fitness function, or for another they put two bots facing eachother with a 'puck' in the middle and used the position of hte puck as the fitness function (bot had to get it to its own side or something like that).

I think it was done by John Koza, but I'm not certain and I can't find it right now.

This sounds like two things: Conway's famous "Game of Life," which shows you can get evolving, complex "creatures" from very simple mathematical rules on a 2D black-and-white grid, and game guru Will Wright's upcoming project "Spore."

There's a more complex version of the Life simulation called "Avida," too.

Quote:

Original post by Kris Schnee This sounds like two things: Conway's famous "Game of Life,"[...]

Not at all what I was talking about. The project I mentioned was a fully 3d environment (though everything was made of rectangular prisms of various dimensions). Also, it was an old project because the few renderings available were rendered on a supercomputer and consisted of untextured rectangular prisms.

Check these out.

-Kirk



http://www.frams.alife.pl/

It's the "Blockies" by Carl Sims. The videos are still very impressive to me :) The creatures evolved many methods of locomotion that looked very natural, especially the various swimming patterns.

Quote:

Original post by Extrarius I've seen what you're talking about on the net, but it wasn't quite as you describe: they didn't design the organisms at all. They set up a simple physics system, and created a genetic system so that the body shape and neural net that ran it could be put into a genetic programming system, then they did standard evolution on everything using various statistics as the fitness function: for example, for swimming they might let each bot swim for 5 seconds and use distance from original position as the fitness function, or for another they put two bots facing eachother with a 'puck' in the middle and used the position of hte puck as the fitness function (bot had to get it to its own side or something like that). I think it was done by John Koza, but I'm not certain and I can't find it right now.

I think this what I saw...

It's pretty interesting how they have made learning computers.... Maybe one day they will have the first robot to learn as a human.

20 years later...

We'll prabably be in a cage at the zoo under the endangered species section.

I think learning is the key to A.I. A.I interaction in video games is boring. Npcs lack personality, learning,tatics, pathing and alot more. They are completely static and never evolving which is stupid and boring. I believe this can be explained in mathematics.