I knew before I started this post, that I was going to be diving into the territory of chess anathema. However, one cannot look at the great chess players of the world and pass over Deep Blue. The two matches between Deep Blue and Garry Kasparov will remain some of the most important in chess history.
The tail of Deep Blue is rooted in two independent story lines; machine learning and chess itself. One can even say, that the two story lines have always been headed for a collision course. Very few activities of the human mind have so much rigidity, while at the same time forcing the player to use creativity. The line between creativity and rigidity has been the ultimate goal of machine learning from its infancy.
Nuts and Bolts
Deep Blue as a computer was nothing fancy. At the time there where larger, more powerful super computers. However, the software was where the computer stood out. Although many of the details still remain a secret, we know a bit about what it took to beat Kasparov. The computer was said to have studied over 4,000 board positions and 700,000 grandmaster matches. It was likely that many of Kasparov’s own matches had been studied.
While library moves where not a new idea in creating a chess playing computer, it was the implementation of that library that made Deep Blue great. Not only would it look at potential moves, but it would look at the likely moves beyond that as deep as 20 moves or more in some cases. It was estimated that for every half move the computer looked at, it added 50 to 70 Elo points to its strength.
Match One
While most of us see match two as the real pinnacle of achievement, I am going to actually say, match one was more important. Yes, it was important that Deep Blue eventually won out over a standard match. But the fact that it won at all against a reigning world champion under tournament conditions was such a huge achievement, that no one would have guessed that it was possible.
When the first match was aired, it was largely thought that Kasparov came out keep because he wasn’t expecting the strength of Deep Blue’s play. This was evidenced by the many materialistic plays typical of computer. It is more likely that Kasparov was sticking to his traditional style of play. This put the game well within Deep Blue’s wheel house and in the end, the decisive capture of Kasparov’s knight proved too much for him.
However, we all know how the match ended, 4-2.
Match Two
After his triumphant defeat, the IBM team wanted to have another try. Feeling that he so soundly beat Deep Blue, Kasparov agreed to a re-match. We all know how it ended, with Kasparov’s defeat, however, the game was not without his controversy and intrigue.
Kasparov had claimed that he saw imagination and insight beyond how the computer had previously performed. He made the claim that there must had been some form of human intervention. The IBM team responded that there was no human intervention during match play.
It has been actually shown to be a bug in the software. During the first game of the match, Deep Blue simply made a play by moving its rook. This seemed to show creativity, but was actually a safety and default move made by Deep Blue when it could decide what move to make. Nate Silver later concluded that because Kasparov felt there was a superior intelligence at play, it caused him to change his game play in the second game. Eventually, Kasparov was so upset because he felt cheating was involved that he resigned the second game, even though it has been shown a draw was an option.
Another notable event of the match was the first game. Kasparov opened with an irregular opening. His thought process was to throw the computer off of his book of plays. This was likely the first documented anti-computer tactic. While common now, was a novel move during that time.
The Aftermath
One of the creators of IBM’s Deep Blue wanted another rematch, to see if he could beat Kasparov more soundly. However, Kasparov has never agreed. This did spawn a whole new group of chess playing computers. Recently, several have been successful against world chess champions. Even more notable, they no longer require a super computer, but can run on high end PC. Also, a new area of chess was developed around using anti-computer tactics, which have been used with varying success in many high profile matches.
In the end, I don’t think the question is if computers can beat human anymore. They certainly can. I think it goes back to the dichotomy I presented at the start. Perhaps computers are forcing us to use more creativity rather than a rigid set of defined actions.