What Computers Can't Do, Part 2
Nevertheless, at the risk of falling into my own trap, I think there are at least two types of positions where chess software is relatively likely to go astray.
(1) Frog-in-the-pot positions.
It's allegedly the case that if one throws a frog into a pot of boiling water (why would anyone do this??), it will immediately strive to escape, whereas if a frog is put in tepid water which is then heated, it won't attempt to escape even when the water attains a full boil.
The chess software version of this occurs when its opponent is building an attack, but the threats are still too far away to be seen. In such positions, the computer might continue on its merry way, seeking objectives on the opposite side of the board from where its king is, only to find that when the threats do come, they're unstoppable.
There are some defensive positions that a human can instantly see are unbreachable, but unless it's in a tablebase or the 50-move rule is looming, it won't recognize that the position is in fact equal.
A third sort of example traditionally given occurs in positions with locked pawns. Standard lore has it that the computer turns into a blithering idiot once a pawn chain has arisen, but I think things aren't as clear here as they used to be. The one game Kasparov won in his match with X3D Fritz was a triumph of that strategy, but I have also seen chess software play very well in locked-up French and King's Indian Defenses.
Of course, programmers are aware of all of these issues, and improvements in both software and hardware the past few years have reduced chess programs' vulnerability to the first and third weaknesses in particular. So for those who want to play anti-computer chess, your task is growing more and more difficult, and my advice is just to play good moves: you'll learn more, you won't hurt your style and you'll probably even do better!