I lost this blitz game on FICS, but sense that I missed a better opportunity, so here is my analysis game for getting an advantage.
It takes Crafty over a minute to spot 18.Qd3 for White, and before that it thinks Black is winning. A couple minutes after finding Qd3, it realizes the position is even, but at least one minute after before it thinks Black’s advantage is only tiny.
A human would probably choose White right away.
I had to have the guts to tell my chess engine that it was wrong, that White has nice compensation for the pawn, even though this would have been impracticable for me to calculate over the board in a G/90 setting. I didn’t find the continuation right away, but liked the c4 (mine) and Qe2 idea. For years, I would lose such a position to a computer because I cannot calculate like it, lots of ways for White to go wrong, but this time, with Crafty’s help also, I was able to find it. Can you imagine if one us actually beat Crafty by deeper tactics(?). Believe it or not, this is possible. I think it’s the consistency and aggressive play, particularly when challenging human foibles, rather than other computers, that gives an engine such a high possible rating.
Like I say, this is a quad-core, 2.4 Ghz phantom AMD machine (ASUS MB), Dual RAM 2GB in parallel, but acting as a fast 1GB RAM (633 Mhz FSB, plus 2x throughput).
This is why I always look at computer continuations skeptically at first when I read them on a blog. If anyone puts this position up against Fritz or Rybka, let me know how well they performed on this position. 🙂
I would imagine that there are other programs with a more human-like heuristic in their algorithms that don’t go by calculation alone but more by “the look” on the board, as a human would do. They may get it right, right away.
I noticed when looking at GM games on CA that even that engine would not spot the tactics that the GMs did, for a minute or so, mating attack type of attacks. It’s more surprising though, when you or I find these. Crafty does go in over it’s head frequently, though.
I’ll go so far as to say that someone like Capablanca could probably beat a computer that plays this way easily enough. For Capa, he may have looked 12 moves deep, but may have only looked at “the best one” on each move, as he liked to say, so that is only 12 moves that he has to hold in his head, not a million or more, as with a computer.