I think that this is what impresses folks most about programs like this: Humans playing 20 questions will usually go for questions to which we think the answer is “yes”, and so we think of someone getting a bunch of "yes"es in a row is a sign that he’s doing well. But when you’re confident that the answer to a question is “yes”, it’s a waste of a question to ask it: Why ask when you already know the answer? With a well-chosen question, a “no” answer will give exactly as much information as a “yes”, and so a computer that appears to be getting nowhere (i.e., it’s getting a lot of "no"s) is still making good progress.
Couldn’t get Roland Burris. (Kept trying to guess Deval Patrick or Jesse Jackson). Screwing around with it I could get it to guess Roland Burris if I answered the question “Is he a member of the current government” incorrectly with “Yes”, instead of the correct “No”. So it doesn’t appear to be entirely up to date with that.
It got Hazumu from Kashimashi Girl Meets Girl after 20 questions. 2 of which I answered ‘don’t know’ to and 2 it didn’t have an expected answer.
Not a single one asking about her gender swap. Or her alien encounter. Or the fact that it’s an anime/manga series. And only tangentially the fact that she’s in a love triangle with two other girls (it asks if she’s in a love triangle, and if she’s in love with a man, but that’s it, and I wouldn’t call that enough to call it).
I’m not surprised it got her. I am SHOCKED it got her with the information it had.
It got Warden Norton from Shawshank Redemption amazingly quickly with very little pointed questions. I was pretty surprised when he got it right. It didn’t really ask anything specific about prisons or even being a bad guy. The closest it asked was “did he lead a group of people” and “did he use guns” and I’m pretty sure that’s awfully broad. I even responded “I don’t know” when he asked if the character wore glasses.