Sam Trenholme wrote on Thu, Jun 5, 2008 06:34 PM UTC:
Yes, I'm sure it is possible to make a computer program that can play Arimaa well; however to do so will require breakthroughs in AI that we've tried to do for decades without success.
Basically, the computer in front of you is a complex adding machine. It doesn't think nor recognize patterns the way a human does. Yes, we've made the adding machines complex enough that they can do things like play music and movies, and even play Chess well. But we haven't been able to have it so computers can, for example, translate from one language to another without the translation being so bad it's just about not readable.
Nor have we been able to get a computer to play a game with a high branching factor, like Go or Arimaa well. Computers play Chess very differently from humans; they just look at all of the possible moves, using 'alpha-beta' pruning to determine which moves are and are not looking at. They don't recognize patterns; they just see possible future moves and how much material they have.
A computer needs to evaluate millions of possible positions to play as well as a human who only looks at dozens of possible positions. Computers aren't able to really see a given position to evaluate how good it is; they only play as well as we do because they basically brute force through just about every possible chess move so many moves down.
Games like Go and Arimaa are good because brute force just doesn't work with these games. In order to have a computer play these games well, we will have to make a true AI breakthrough. Which will probably have consequences far beyond just having a computer playing some abstract game really well.
Basically, the computer in front of you is a complex adding machine. It doesn't think nor recognize patterns the way a human does. Yes, we've made the adding machines complex enough that they can do things like play music and movies, and even play Chess well. But we haven't been able to have it so computers can, for example, translate from one language to another without the translation being so bad it's just about not readable.
Nor have we been able to get a computer to play a game with a high branching factor, like Go or Arimaa well. Computers play Chess very differently from humans; they just look at all of the possible moves, using 'alpha-beta' pruning to determine which moves are and are not looking at. They don't recognize patterns; they just see possible future moves and how much material they have.
A computer needs to evaluate millions of possible positions to play as well as a human who only looks at dozens of possible positions. Computers aren't able to really see a given position to evaluate how good it is; they only play as well as we do because they basically brute force through just about every possible chess move so many moves down.
Games like Go and Arimaa are good because brute force just doesn't work with these games. In order to have a computer play these games well, we will have to make a true AI breakthrough. Which will probably have consequences far beyond just having a computer playing some abstract game really well.