Writing for IEEE Spectrum, Joanna Goodrich says that Deep Blue beat Kasparov because was just so fast.
The supercomputer could explore up to 200 million possible chess positions per second with its AI program.
But it wasn’t. Fast enough. Not really. IBM didn’t expect to win, just to lose less badly. Kasparov won the first game. Lost the second. Drew three.
In an account I read years ago (Pandolfini?), it came down to psychology. Deep Blue was doing better than expected, and K started to doubt his preparation or understanding.
Then Deep Blue played a move that spooked Kasparov into thinking it was far faster than it really was, and (uncharacteristically) he panicked and resigned.
But the machine itself had panicked. A bug made that move random.
This recent article roughly agrees with my decades-old memory, supplying the bit I forgot about it being a bug.