“I was in attendance at the US Go Congress match yesterday where history was made: the go program MoGo, running on an 800-core supercomputer, beat 8-dan professional go player Myungwan Kim in a 9-stone handicap game. Most in the audience were shocked at the computer’s performance; it was naturally assumed that the computer would be slaughtered, as usual. Go is often seen as the last bastion of human superiority over computers in the domain of board games. But if Moore’s law continues to hold up, today’s result suggests that the days of human superiority may be numbered.”

I am a Go player, and started my ph.d. research on computer approaches to go in the early 90′s. This is an amazing achievement. Some commentators have downplayed the significance because the Go program received a 9-stone handicap. But what they don’t realize is that a serious amateur Go player (like myself) would not likely be able to beat a professional 8-dan player with that same handicap.
The approach used by these Go programs, which involves simulating millions of random games to the very end and backing up the outcomes to select the best current move, is similar to what my friend Bruce Abramson developed in his PhD work on “Expected Outcome” model of learning and search.
It’s interesting to see these ideas take 20 years to yield fruit.
I think we’re going to see a lot more progress based on AI ideas developed in the 80′s and early 90′s over the next 5 years.
My prediction for when a Go program beats a human professional with no handicap: 2015.

• A variant of chinese checkers in which up to 6 players could play at a time.
• A game called “blocks and corridors” in which the two players make simultaneous moves on identical 3×5 grids. Each player has a single piece that start at the top right of their grid and win by being the first player to get his piece to the bottom of the grid. On each turn, a player either moves his piece or sets up a block between two rows on the opponent’s board. The block covers the left or right two squares (meaning the blocked player has to move to the square on the unblocked column before moving down to the next row).
• A standard version of Othello.

The teams had worked really hard on their players and had implemented a variety of interesting visualizations and strategies, including:

• Multi-player minimax searches.
• Searching breadth and depth first in parallel.
• Making a series of (mostly) random probes to the end of the game and then taking the expected outcome for each player to decide on the statisticaly best move.
• Compiling the game descriptions into C and and then assembly language for extremely fast calculation.
• Analyzing the rules for invariants and strongly connected components (including a graphic display).
• A graphic display that shows search tree considered by the program, with gradient colored nodes indicating the program’s evaluation of each position. (This was beautiful!).

Unfortunately, demo effects set in to the detriment of each of the teams. Hence, most of the programs, including ones that had performed flawlessly in practice rounds just the day before, made illegal moves, or failed to move in time. The students were understandably disappointed. Nevertheless, Mike Genesereth and I were both impressed with what they accomplished. I hope the teams continue to debug their programs so we can see what happens when the programs play as they were intended to do.
There will be a general game playing competition at the upcoming AAAI national conference in July. Since Stanford is the organizer, these students won’t be able to compete for the prize, but I expect their programs will make strong showings in the open competition.

My love of games started when my grandfather taught me to play chess at the age of 5 and continues through this day. While I have studied chess and go as a tournament player, I also like the challenge of learning new games. Part of the fun for me is playing with the rules to discover the strategies that follow from them. This is part of the fascination that led to my research on computer game-playing for specific games (e.g. chess, go, bridge) and ultimately to my thesis work on general game playing.

Publications

• Gamb”ack, B.; Rayner, M.; and Pell, B. 1991. An Architecture for a Sophisticated Mechanical Bridge Player.In Levy, D., and Beal, D., eds., Heuristic Programming in Artificial Intelligence 2 – The Second Computer Olympiad. Ellis Horwood.
• Pell, B. 1991b. Exploratory Learning in the Game of GO. In Levy, D., and Beal, D., eds., Heuristic programming in Artificial Intelligence 2 – The Second Computer Olympiad. Ellis Horwood.
• Pell, B. 1992a. Metagame: A New Challenge for Games and Learning.In van den Herik, H., and Allis, L., eds., Heuristic Programming in Artificial Intelligence 3 – The Third Computer Olympiad. Ellis Horwood.
• Pell, B. 1992b. Metagame in Symmetric, Chess-Like Games.In van den Herik, H., and Allis, L., eds., Heuristic Programming in Artificial Intelligence 3 – The Third Computer Olympiad. Ellis Horwood.
• Pell, B. 1992c. Logic Programming for General Game Playing.University of Cambridge Computer Laboratory Technical Report No. 302.
• Pell, B. 1993. Strategy Generation and Evaluation for Meta-Game Playing.Ph.D. Dissertation, Computer Laboratory, University of Cambridge. Also appears as University of Cambridge Computer Laboratory Technical Report No. 315.
• Pell, B. 1994. A Strategic Metagame Player for General Chess-Like Games. In Procs. of AAAI-94 (1994), 1378-1385.
• Pell, B. 1993. A Strategic Metagame Player for General Chess-Like Games. In Procs. of AAAI Fall Symposium on Games: Learning and Planning.
• Matsubara, H., and Pell, B. 1994. Applying metagamer to shogi. In Game Programming Workshop in Japan’94. Kanagawa, Japan: Japanese Computer Shogi Association.
• Pell, B. 1996. A Strategic Metagame Player for General Chess-Like Games.Computational Intelligence 12(1).
• Pell, B.; Epstein, S. L.; and Levinson, R. 1996. Introduction to the special issue on games: Structure and learning.Computational Intelligence 12(1).

Other Links

Related work on Metagame and General Game Playing Programs:

• Danny Cron’s
  Metagame Page

• Jay
  Scott’s Metagame Page

• Stanford General Game Player Project.