So the game turns from "the absurdly smart computer stomps everyone" to "the absurdly smart (but deliberately crippled) computer may or may not stomp everyone, depending mostly on random chance."
That doesn't sound like a great improvement, imho.
That would be more "fair," by which I mean that it would remove the obviously computer-favored dynamic of buzzer reflexes. Of course, it wouldn't make good TV, it wouldn't really be Jeopardy!, and I suspect Watson would lose. Against normal season contestants Watson would probably fair well (in a buzzerless competition), but against human champions Watson wouldn't stand a chance.
There's no way to know what portion of the clues the human contestants knew the correct responses to. I've watched Ken and Brad play extensively though, so I very strongly suspect that they easily knew at least as many as Watson.
I would like some stats about that. As I said, Watson could have correctly answered 84% of the questions, nobody could answer 6% of the questions. That leaves a very thin margin for the humans, if past games are any indication one that is probably not large enough.
That would give a more precise idea of just how many of the questions Watson was able to answer correctly, but at that point the discussion starts to feel like people proposing changes to the rules of chess in order to screw Deep Blue.
