vetta project

I’ve lost count of the number of times I’ve seen the 1997 match between Deep Blue and Kasparov listed as a major success of AI. It was a moment in AI history, for sure, but I’ve always felt that the victory was a bit hollow. For example, not allowing Kasparov enough rest and then not giving Kasparov the chance for a rematch. As such, over the last week I’ve been following the Deep Fritz vs. Kramnik games with great interest — hoping for the decisive victory of machine over man that 1997 somehow didn’t quite deliver. In case you missed the results, it was 6 games in total that produced 4 draws and 2 wins to Deep Fritz.

Although the results give a clear victory to machine over man, it’s only when you take a closer look at what has happened that the extent of things become clear. In none of the games did Kramnik appear to be significantly ahead, indeed it was the machine that was almost always on the attack looking for a win, while Kramnik was defending and playing for a draw. Even more telling, Deep Fritz would sometimes make strange moves that the Grand Master commentators could not explain… at least not until a few moves later when the subtle reasoning behind the move became apparent. Deep Fritz was playing at a level which the best chess minds in the world were struggling just to understand, let alone beat.

Perhaps the most significant thing of all, at least from my perspective, was that Deep Fritz was not a multi-million dollar super computer. It was just a few high end PCs. This was possible partly due to Deep Fritz having better algorithms than Deep Blue, and partly due to 9 years of Moore’s law. Should the Fritz code be run on a modern day super computer, it is clear that it would totally crush any human.