Deep Blue defeats Kasparov

Back to Computer Chess article

On May 11, 1997, Kasparov lost a six game match against deep blue with a score of 3.5-2.5. Two wins for Deep Blue, one for Kasparov and three ties. (Kramnik update at the end)

I think most people (including myself) will remember this date as the day that a computer won the world chess championship.

Those who have read my article will ask if I have any comments regarding the paragraph:

Given the previous history of computer chess, by 1998 Deep Thought should be playing at the level of the top three players of the world. If by that time the original developers are not tired of computer chess, then sometime in the next decade DT should beat a Kasparov-strength player, which most likely will not be the chess champion then (hey! with time humans develop and play better too!). If no major developments occur in the human side of chess, a computer should beat the world champion (whoever she may be) by the year 2022, 70 years after ``sometime in the next decade'' was predicted for the first time.

And indeed, I do. Lets go over it phrase by phrase:

".... by 1998 Deep Thought should be playing at the level of the top three players of the world."
Given that I wrote this paragraph in 1993, I would say it was accurate enough.
"If by that time the original developers are not tired of computer chess, then sometime in the next decade DT should beat a Kasparov-strength player"
Here's where I have a quibble. I had assumed that whoever accepted a match against the best computer at that time would demand to see the computer in action before playing against it. This means that the computer would have been thoroughly examined by humans before it ever went against the champion.
Why did I assume this? Because it had more or less been that way in the past. Computers played in computer tournaments, which were open to the public, and thus players could learn from them. In fact a year before the match, I wrote in rec.games.chess.computer
My prediction is that Kasparov won't be beaten this century, provided he has access to the scores of games between the computer and other GM's. If the Deep Blue team stays at it, they might be able to beat the world champion in those conditions sometime in 2000-2010.
Do I then consider Deep Blue's victory "not valid"? Not at all.
There are other board games, such as backgammon, for which I understand a computer won the world championship without having played many public matches. So it seems to be established practice for computers to win world championships "by surprise".
"If no major developments occur in the human side of chess, a computer should beat the world champion (whoever she may be) by the year 2022".
Here I was plain wrong. Even allowing for public playing on Deep Thought's part, I think a successor forty times faster would be invincible. As Deep Blue uses parallel hardware, in principle such a machine could be built today, if the funds were made available.
Even with limited funding, I think one could certainly build such a machine before the end of 2010, which is well before 2022.
Why did I miss by so much? There are two reasons.
First, I thought that human players played at a higher level. I've read many books where Grand Master level matches are analyzed. From the analysis in those books it would seem that blunders are rare. But lately, when computers do post-mortem analysis of matches they tend to find errors at a more superficial level that I would have expected them to. I mean, we know Grand Masters do not play perfect chess, but I thought almost all their errors would be deep and far reaching. It turns out the percentage of "trivial" errors is higher than I originally thought.
Secondly, I didn't take into account that, as computer chess design teams got closer to the holy grial of beating the world champ, they would try that much harder.

What did we learn from Deep Blue's victory? A lot. First and foremost is that a sensible evaluation function, combined with a deep enough search, results in moves that can be construed as intelligent and creative. In other words: You can make up with speed what you lack in intelligence, seems to be the lesson here, at least as far as chess goes.

Quite a few people have been saying lately that chess is particularly suited for computer attack. This is not so. While the computer certainly benefits from a well defined rule set, the evaluation of a position is quite subjective. This type of knowledge has been traditionally one of the hardest things to encode in computer terms.

In fact, Deep Blue made up for this deficiency by sheer speed. However, given the significant improvement in the level of play between last year's match and this years, one can only admire the work done by IM Joel Benjamin, GM Miguel Illescas and player/programmer Murray Campbell.

I take the functional view of artificial intelligence. "If it seems intelligent it is intelligent", or paraphrasing Forest Gump Intelligent is as intelligent does.

While there might be deep phylosophical reasons why a machine that emulates intelligence is not necessarily intelligent, I don't care much. The reason why we want to emulate intelligence with computers is not to prove a point (at least it isn't the driving force), but to have computers aid humans in intelligent decision making.

Today, if I were to play a game of chess against Kasparov, I would definitely like to have Deep Blue next to me, giving me advice.

more to come....

In October 2003, Deep Fritz, a commercial chess program played a match against the world champion Kramnik. The match ended up in a draw. In this case Kramnik had access to Deep Fritz before the match. This results seems to support the contention that if the human player had access to the computer program, a computer chess champion would not have been possible until sometime in 2000-2010.