TITLE: Intelligent Game Playing in the News AUTHOR: Eugene Wallingford DATE: July 23, 2007 1:59 PM DESC: ----- BODY: Two current events have me thinking about AI, one good and one sad. First, after reporting last week that checkers has been solved by Jonathan Schaeffer's team at the University of Alberta, this week I can look forward to the Man vs. Machine Poker Challenge at AAAI'07 The computer protagonist in this event, Polaris, also hails from Alberta and Schaeffer's poker group. In this event, which gets under way shortly in Vancouver, Polaris will play a duplicate match against two elite human pros, Phil Laak and Ali Eslami. Laak and Eslami will play opposite sides of the same deal against Polaris, in an attempt to eliminate the luck of the draw from the result. I don't know much about computer card-playing. Back when I was teaching AI in the mid-1990s, I used Matthew Ginsberg's text, and from his research learned a bit about programs that play bridge. Of course, bridge players tend to view their game as a more intellectual task than poker (and as more complex than, say, chess), whereas poker introduces the human element of bluffing. It will be fun seeing how a "purely rational" being like Polaris bluffs and responds to bluffs in this match. If poker is anything at all like chess, I figure that the program's dispassionate stance will help it respond to bluffs in a powerful way. Making bluffs seems a different animal altogether. I wish I could be in Vancouver to see the matches. Back in 1996 I was fortunate to be at AAAI'96 in Philadelphia for the first Kasparov-Deep Blue match. The human champ won a close match that year before losing to Deep Blue the next. We could tell from Kasparov's demeanor and behavior during this match, as well as from his public statements, that he was concerned that humans retain their superiority over machines. Emotion and mental intimidation were always a part of his chess. On the contrary, former World Series of Poker champion Laak seems unconcerned at the prospect that Polaris might beat him in this match, or soon; indeed, he seems to enjoy the challenge and understand the computational disadvantage that we humans face in these endeavors. That's a healthier attitude, both long term and for playing his match this week. But I appreciated Kasparov's energy during that 1996 match, as it gave us demonstrative cues about his state of mind. I'll never forget the time he made a winning move and set back smugly to put his wristwatch back on. Whenever Garry put his watch back on, we knew that he thought he was done with the hard working of winning the game The second story is sadder. Donald Michie, a pioneer in machine learning, has died. Unlike many of the other founders of my first love in computing, I never had any particular connection to Michie or his work, though I knew his name well from the series of volumes on machine learning that he compiled and edited, as they are staples of most university libraries. But then I read in his linked Times On-Line article:
In 1960 he built Menace, the Matchbox Educable Noughts and Crosses Engine, a game-playing machine consisting of 300 matchboxes and a collection of glass beads of different colours.
We Americans know Noughts and Crosses as tic-tac-toe. It turns out that Michie's game-playing machine -- one that needed a human CPU and peripherals in order to run -- was the inspiration for an article by Martin Gardner, which I read as a sophomore or junior in high school. This article was one of my first introductions to machine learning and fueled the initial flame of my love for AI. I even built Gardner's variant on Michie's machine, a set of matchboxes to play Hexapawn and watched it learn to play a perfect game. It was no Chinook or Deep Blue, but it made this teenager's mind marvel at the possibilities of machine intelligence. So, I did have a more direct connection to Michie, and had simply forgotten! RIP, Dr. Michie. -----