From: Michael DeHaan Date: 2004-12-11T04:57:21+09:00 Subject: Re: [QUIZ] Learning Tic-Tac-Toe (#11) On Sat, 11 Dec 2004 04:42:31 +0900, Brian Schr�der wrote: > On Sat, 11 Dec 2004 01:22:30 +0900 > > > James Edward Gray II wrote: > > > On Dec 10, 2004, at 10:11 AM, Brian Schr�der wrote: > > > > > I expect the world to tell me what the legal moves are. Otherwise my > > > agent > > > would have knowledge of workings of the world that she should not have. > > > > I'm using a Board object, in my playing around. My goal in writing the > > quiz was to outlaw you teaching your program any strategy. It must > > learn that. > > > > I personally am okay with seeing the board, but you make your personal > > challenge as hard as you like it! > > > > James Edward Gray II > > > > > In this case, would a minimax-a/b algorithm be teaching the ai a strategie? > Tic-Tac-Toe is so shallow, it can be searched completely by minimax-a/b, so > this would be a one-page solution. > > Regards, > > Brian > > -- > Brian Schr�der > http://www.brian-schroeder.de/ > > Man, I really wish I had the energy after work to work on these instead of playing Gran Turismo and the like :) Minimax isn't a learning algorithm, and that's a bit too easy because you'd be able to completely solve the game (full-depth). Rather than ban such things, how about making the problem harder? Arbitrary MxN grids and a set time limit of X seconds for each move. (Say X = 30 seconds on a 1GHz box or something). At some level, Minimax would break down without a good fitness algorithm attached.