Sunday, May 27, 2007

An Interesting Game?

In a 1972 review in Computing Reviews, Z. G. Vranesic said flatly that no computer would give a Chess Master an interesting game in the 20th Century. I was surprised that he would bet against technology over such a long time scale--28 years.

Zvonko was a friend and colleague, and I had a great deal of respect for him. He was an excellent chess player (an International Master and player for Canada in the Chess Olympiad), and I had followed with interest his work with Michael Valenti developing the CHUTE chess-playing program.

So I bearded him in his office, saying "28 years is a long time." He had no idea what I was talking about, so I waved the review in front of him, and asked him why he'd made such a bold prediction. He maintained that it was reasonable, and offered to wager on the prediction. We settled on the stakes being a bottle of spirits of the winner's choice.

We also needed to agree on the definition of "interesting game." We agreed that, in the spirit of the review, the definition should involve capability over a series of games, not whether some single game was, by some criterion, "interesting." We finally decided that a chess player was capable of giving another player "an interesting game" if it was able to win or draw at least half the time. This did not require it to be of equal strength, but at least it could not be a consistent loser.

Sometime later Zvonko came to my office, and said he'd been thinking, and 28 years was a long time. He'd like to win the wager sooner, by redefining its term and terms. We agreed on 10 years and Expert, rather than 28 years and Master.

So what happened?

In 1977 CHESS 4.5 won the Minnesota Open winning 5 games and losing one. It had a performance rating of 2271, i.e., was already above the Expert level.

On May 11, 1997, World Champion Garry Kasparov lost a six-game match against IBM's Deep Blue computer and program, with a score of 3.5-2.5. Two wins for Deep Blue, one for Kasparov and three ties.

So, for either form of the wager, I won handily. (But I never did get that bottle of spirits.)

Moral: Don't bet against technology in the long term.
Moral 2: Collect at the time you win.

In 1957, Herbert Simon forecast that "within 10 years, a computer will routinely beat the world's best player." Wrong.

In 1965 Herbert Simon predicted that "machines will be capable, within twenty years, of doing any work that a man can do" (Simon 1965, p. 96). Wrong again.

Moral: Even Nobel Prize winning geniuses's predictions aren't always right.

Other tidbits on computers and chess:
First known computer chess program.
ACM Computer Chess.
Louis Kessler's Chess and Computer Chess Links.
Computer chess and Waterloo University.