by May 10, 2002 0 comments

May 1997 was a critical juncture in the history of the evolution of computers. IBM’s Deep Blue, a chess-playing computer, beat Gary Kasparov, possibly the world’s greatest chess player. Deep Blue won the six-match series 3.5 – 2.5–three draws, two victories for Deep Blue and one for Kasparov. (A draw is half a point for both players, winners get one point, losers get zero.) Finally, man was beaten by machine in the most cerebral of activities!

If you go through IBM’s site for the match (or rather rematch; the first match between the two was played a year earlier and ended in a 4-2 victory for Kasparov–three wins, one loss, two draws for Kasporov), you will not find the word supercomputer anywhere. What you will find is a rather muted reference to an HPC machine. By now, you and I know that it is one and the same thing. 

Before being defeated by Deep Blue in 1997,
Kasparov had, in earlier matches, defeated both Deep Blue and its predecessor, Deep Thought 

For the rematch, unlike Kasparov, Deep Blue had undergone a complete refitting of hardware and software!

The original Deep Blue of 1997 was a 6 feet 5 inch tall, 1.4 ton heavy supercomputer built out of a 32 Processor IBM SP2, with 256 specially designed chess accelerator chips. Only half of these were used in play, the other half being backups in case of failure.

In contrast, the Deep Blue of 1998 used a 32 node RS/6000 SP server (these machines have since been renamed the PSeries e-servers) with 30 P2SC processors that delivered twice the processing power of the earlier version, and used all 256 chess processors. With this configuration, Deep Blue is capable of making 100 to 200 billion moves within the allocated three minutes for a move.

Chess and computers
Chess-playing computers go back to 1958. 

But it was only in the early eighties that a chess computer capable of master level play (Kasparov is a grandmaster) could be created. This was the Belle from Bell Labs. 

Deep Blue has its origins in Chip Test, a chess-playing computer designed by Feng-Hsiung Hsu, Thomas Anantharaman, Murray Campbell and Andreas Nowatzyk, graduate students at the Carnegie-Mellon University in 1985. 

Chip Test led to Deep Thought, using about 200 off the shelf processors along with two specialized chess processors. Deep Thought was the first chess computer to achieve grandmaster-level play.

Subsequently, Hsu, Anantharaman and Campbell joined IBM’s Thomas J Watson Research Centre and continued work on the chess computer. Deep Thought led to Deep Though II, and then came Deep Blue. 

For the rematch, Joel Benjamin, an American grandmaster, joined the Deep Blue team.

Deep Blue has no IQ and uses a brute force method of deciding which move to make. That is, it evaluated all possible moves up to six sets of moves or 12 plies (two moves, one by each player is a ply) forward to determine which is the best move to make.

Even Deep Blue cannot analyze all possible such moves in the allotted three minutes. Also, what if the seventh move ends in a checkmate for its own king? So it uses alpha-beta searches and chess rules and position values to reduce the number of calculations it has to make to a manageable few billion. Also, if the number of choices for a move is limited, then Deep Blue can search many more plies ahead.

What next?
Officially, the Deep Blue effort seems to be over. It does not look like there will be another chess match between a world champion and a machine. But then, you never know.

Krishna Kumar

No Comments so far

Jump into a conversation

No Comments Yet!

You can be the one to start a conversation.