Posted 19-Mar-11 23:39 by White_Mage
Yes it could work
Well, newb1 I hear you. I have very similar ideas to yours and I know specific ways to make AI learn in a game, like C-Evo, so I disagree that it is impossible. It takes tremendous time of coding, debugging and AI learning for a game like this to see desirable outcome. There are several ways to address this learning problem and in principle high quality AI could be created. Game rule changes would be, however, disastrous for AI’s which already specialized for a specific version of the game (although this is not out of question either). Chess AI currently is far beyond any human’s capability, but as AndersI pointed out, leaning is not the reason why they are so strong. The game is too simple, so AI capitalizes on computing speed. Chess AI simply does not need to learn to beat best humans and chess is the most well funded game AI development project, by far. C-Evo probably does need learning. Since all of us do this C-evo project as hobby and not for profit, development of strong learning AI may take a long time, but I firmly believe that we are on the path towards your vision and I hope that it will be delivered at some point, perhaps for C-Evo, perhaps for some other similar game. In my estimation, a well funded, small group of AI experts could deliver miracles in as short as 3 years. Until then we rely on individual enthusiasts spending sporadic time on development, very few having knowledge+skills+time+determination to make it happen.