Kasparov vs. X3D Fritz. Machines apparently are not good at the positional stuff, i.e., strategy! They key insights come from the analysis of move 18.

This useless-looking move confused most of the commentators, but to anyone with extensive anti-computer chess experience it makes perfect sense. The rook protects the f2 pawn, a potential weak spot, but why would you protect something that isn't being attacked?

The reason goes into how computers think. Its brute force calculation can only go so deep, even with four super-fast processors. Black's only possible source of counterplay in this position is to push its f-pawn and open up an attack against the area around the white king, f2 in particular.

If X3D Fritz's search, usually running 12-20 half-moves deep, ever reaches a position in which it sees success in such an attack it will put such a plan in motion. On the other hand, if it cannot reach a favorable position in its searches it will never play the initial moves required. With the rook on b2 protecting f2 already, the potential weakness of that critical square is somewhat hidden from the computer's search.

X3D Fritz can't just play it anyway like a human would, knowing that everything else is useless. A machine has to receive a positive evaluation from its search to play a move and always plays the move that gives it the best evaluation. Since X3D Fritz sees no danger here for itself it is content to play moves that do nothing, but don't cause any negative effect either. It twiddles its virtual reality thumbs. Any human would say, "I have to do SOMETHING."

The computer, not seeing any immediate tactical chances, fails to develop any kind of strategy. The same can be said of the RON AI. Or the AI for any strategy game. For computers, it's all tactics.
posted by Tim at 10:42 AM