The game of go and AI

Discussion in 'Intelligence & Machines' started by Fen, Jul 24, 2002.

Thread Status:
Not open for further replies.
  1. Fen Registered Senior Member

    Messages:
    292
    If you study up on go, you will realize it presents one the most interesting problems in AI. AI go programs, despite millions in rewards, currently can only beat only the most amateur players.
     
  2. Google AdSense Guest Advertisement



    to hide all adverts.
  3. Dinosaur Rational Skeptic Valued Senior Member

    Messages:
    4,618
    There are many who consider Go to be more difficult than chess. It is at least on the same level as chess for a human being.

    For a computer it is far more difficult than chess because there is no known approach which allows number crunching power to be usefully applied to Go.

    All current high level chess programs use an algorithm which does an excellent job of assigning a valid numerical score to positions on the chess board. The chess playing programs look for moves and series of moves which lead to positions with a high score.

    Von Neuman mini-max strategy is used, which is effective due to the validity of the scoring algorithm. While Deep Blue seems to display intelligent behavior, I would not call it an AI program because it really is just an example of number crunching being applied to a difficult task.

    When I last paid attention to Bridge playing programs, they were also poor compared to a good (but not expert) human player. There is money to be made wtih a good bridge playing program, so the motivation is there.
     
  4. Google AdSense Guest Advertisement



    to hide all adverts.
  5. c1earwater Registered Member

    Messages:
    25
    I did my thesis about Go. I tried to use genetic programming to find a min-max function that could analyse a 7x7 Go board. The functions were of the type like the emboss functions in paint programs and would fill up the board with black and white to predict the final position.

    It did not work. Still, I got my thesis.

    Current "good" go programs work by pattern recognision and localised number crunching. This works pretty well, but it's nowhere near the "Grand Master" level. I believe the best programs so far are at the level of an amateur 5th Kyu.
     
  6. Google AdSense Guest Advertisement



    to hide all adverts.
  7. Zero Banned Banned

    Messages:
    2,355
    It's gonna be a loooooong time before AI can do Go decently. There are simply too many patterns and possible outcomes to analyze.
     
Thread Status:
Not open for further replies.

Share This Page