Wolff Morrow Print E-mail
By Joel Benjamin   
June 21, 2007
Dear Joel,

On the tenth anniversary of the rematch between Garry Kasparov and Deep Blue, I decided to plug in the moves from the "controversial" game 2 into today's modern PC chess programs.


Interestingly Fritz 10 plays virtually ALL moves identical to Deep Blue's choices, including the supposedly human-only move of 37.Be4. In fact, Be4 is found instantly by all the programs I tried!

In the end, there were only 3 moves in the game that Fritz 10 did not agree with and those were 26.f4!, 36.axb5, and the final move 45.Ra6? allowing black a perpetual check draw. Of these the only one that really seems out of the blue is 26.f4, which I believe Seirawan dubbed as "Simply superb". My take is that Deep Blue must have had a value parameter set where it gave considerable merit to such a pawn push, which is simply not equally valued by today's programs.

At any rate, my question is do you feel that since modern chess programs choose nearly identical moves to Deep Blue in this game, it gives vindication to the IBM team? I'd love to hear your thoughts on this. Thanks!


Wolff Morrow

I don't think the IBM team needed vindication. Kasparov's argument was based on the premise that Deep Blue's moves could be discounted if other computers did not replicate them. This is like Kasparov telling Karpov his move is invalid because weaker players didn't find the move even when given lots of time. For what it's worth, the notorious log printouts are available at http://researchweb.watson.ibm.com/deepblue/watch/html/c.shtml, though they are very hard for a layman to understand. I've had a copy of the logs for ten years; they show the technical process by which Deep Blue chose its moves and the "principal variation" (PV), the sequence of moves Deep Blue considered to be best play at each level of its search.

The move 26.f4 was a result of my laboratory work. I spent a great deal of time forcing Deep Blue to play closed positions and trying to get it to utilize "levers" along the lines of the move in game two. We may have set the value for levers unusually high because we wanted to make sure Deep Blue would play them.

I think a number of moves from that game (especially those associated with building up with the rooks on the a-file before opening it) would not have been played by the commercial programs of that day. It is a shame that Kasparov met Deep Blue's groundbreaking play with suspicion rather than admiration. Programmers were able to make many of their subsequent advances because they learned from Deep Blue's accomplishments.

Joel Benjamin