At first glance this might not seem like such a big deal, but on reflection I think a human-versus-computer match on Jeopardy! would be an amazing thing to watch.

As I understand the challenge, it’s going to be less about being a vast trivia repository (where the computer could obviously dominate) and more reliant on breakthroughs in human-language AI. People are usually going to muff a Daily Double because they don’t know the answer; the computer, on the other hand, will blow it far more often because it doesn’t precisely understand the question.
(Okay, I understand that on Jeopardy! the question is the answer and vice versa, but you know what I mean.)

Consider this Jeopardy “answer,” for example, under “Misnomers” for $400:

Despite its name, some paleontologists don’t think this ancient animal was even part of the cat family.”

This is no simple Google search. In this game, the name of the category usually provides an oblique clue, the phrasing of the answer is subtly and cleverly helpful in a variable way depending on the money at stake, etc. You probably came up with “saber-tooth tiger” immediately when you looked at the line above, but imagine what a computer would have to overcome in order to achieve the same result in a reasonable amount of time. And that’s a comparatively easy one. Like I said, I think this contest would really be something to see.

More on the story here from the New York Times. (The site might require a free registration.)

If you’ve got a few hours to kill, check out the Jeopardy Archive here.