Computer Chess II

by Dino Surendran

But now the sun of media attention on Deep Blue has set and we are left gazing at the night sky of computer chess. At smaller specks of light, but myriads of them.

Our particular star is Dr. Ingo Althofer, Mathematics Professor at the Friedrich-Schiller University in Jena, Germany.

His is a story about one man who, without all the financial support that the scientists at IBM had, developed a chess-playing mechanism that took on some of the best players in the world. And sometimes defeated them too.

The main problem that computers face when playing chess, or any other game for that matter, is strategy. Apart from simple rules, such as `a queen is more powerful than a bishop or `when you have this configuration pieces, do that', it is difficult to program machines to have the kind of bird's eye view of the game that humans have. On the other hand, their ability to consider millions of moves in seconds means that they can play a very solid game with very few silly mistakes.

Somehow, this works! The combined system is much better than any of its individual parts. For example, in early 1992 3-Hirn consisted of the `Mephisto Lyon 68030' (rating 2260), the `Chess Machine' (rating 2230) and Althofer (rating 1900 - a good club level). This system defeated International Master Dr Helmut Reefschlager (rating 2405) 4.5:3.5 in an 8-game series.

A quick explanation of the ratings referred to above. They are the Elo ratings (after their inventor, the Hungarian statistician Arpad Elo) used by chess federations to rank their players. A rule of thumb is that you will defeat someone 200 points below you 75% of the time. Beginners start in the hundreds while Grandmasters are 2500 plus, with Kasparov at an all-time high of 2820.

3-Hirn played over a hundred games between 1985 and 1996, typically using the most advanced commercially available computer chess programs available at the time. In other words, 3-Hirn was a changing system, which became increasingly powerful as better chess software was put on the market. In 1996 it defeated Grandmaster Gennady Timoschenko with rating 2530. Then the computer part of 3-Hirn consisted of Fritz 4 (rating 2440 on the hardware used) running in 2-best mode, meaning that this single program would suggest the two best moves possible at each stage.

The most impressive perfomance of 3-Hirn was in 1997. This was over Grandmaster Arthur Yusupov (see photo -- Althofer's facing the computers, while Yusupov is facing defeat), at the time No. 32 in the world with a rating of 2640. Chess fans may remember him as being Vishwanathan Anand's second in the FIDE World Championship matches played by the Indian against Karpov in January 1998. However this was not standard chess but a variant - Shuffle Chess, where the game begins with a random arrangement of the back row of pieces.

Furthermore, the two programs (ratings 2500 each) used were each run in 3-best mode, i.e. they would suggest the best three moves at each stage. In other words the human controller had up to six choices of moves instead of two. Yusupov was defeated 5-3 in a series of eight games.

The next challenge was obviously a Top Ten player, followed by the World Champion. But finding a willing player was hard. After all, all the pressure would be on the human player in such a match and it would not be worth his while to enter one - unless enough money was up for grabs. But what company would sponsor such a match when the question of Man Vs. Machine in chess had already been resolved (as far as the popular press was concerned) by the Deep Blue Victory?

But 3-Hirn is not a machine. It is a Man+Machine, i.e. a man enhancing his capabilities using machines. Humans have always used tools to better themselves and there is no fundamental difference here. The question should not be 'is this right?' but rather 'how can we use machines better?' In this case, competitions involving systems like 3-Hirn would encourage people to investigate more effective ways of welding the talents of humans and computers.

While Althofer writes a book on his experiments, it is possible for us to also take a step back and look at his ideas. Why should they apply to chess alone? A medical doctor faced with a situation beyond his ken could ask two computer medical programs for their `opinion' and then pick one using his own knowledge. That would be a better idea than just inquiring from one program. It's a simple idea and no doubt it is used by some physicians. But not by many, and therein lies scope for new ideas and improvements.

And why stop with medicine? Economic planning, industrial optimization, business management, bio-technology - all could benefit. In fact, any area where the sound judgements of computer programs are well married with the far-sighted and intuitive strategic thrust that is the strength of the human mind.