IBM supercomputer to play Jeopardy against humans
You know how computers play chess? Heck, Barnes & Noble’s EBR (E-Book Reader) the NOOK (sic) can do it.
That’s never impressed me as “artificial intelligence”.
Humans who are really good at playing chess aren’t thinking like your typical human. Part (and only part) of chess is memorizing games and recognizing that the game you are playing is the same as a game previously played.
What humans have been able to do better in chess than computers can (besides psychological tricks) is knowing immediately which pieces don’t matter.
A good human chess player can glance at the board and know that several of the pawns, for example, don’t matter this time.
You could hypothetically have a computer look at every piece and every possible move each time and each consequence of every move…but that’s a huge, complicated process, even for the best computers.
Computers can do it with something relatively simple like Tic Tac Toe, of course, but it took an incredible amount of work for IBM’s Deep Blue to beat Gary Kasparov on May 11, 1997…and that win was disputed by Kasparov.
I was impressed by that, but didn’t see it as human-like thinking.
I’ve been saying for years that what would impress me was a computer that could play Password.
You may not know Password, but it was a popular game show for decades. To hit a current cultural touchstone, Betty White was on it a lot (and married its host, Allen Ludden). It was part of a famous Odd Couple episode.
The premise was simple. There were two pairs of players (a celebrity and a contestant on each side is what most people remember). One player on each side is given the same “secret word”. The ones who know what the word is take turns trying to get their partner to guess the word. They do it by saying one word. No gestures, and you can’t use part of the word. There were some inflection tricks: if you using a rising inflection, it meant you wanted the opposite. If you sustained the word, it meant you wanted a following word. If you sort of used a dismissive delivery, it meant your word was an example of a class.
For example, let’s say the word was “tiger”. You might say, “CAT!” really loudly and growly. Your partner might guess, “lion”. The player on the other side who knew might sort of shrug and say, “Tony.” That person might get “tiger” from the big cat you suggested and Tony the Tiger as an example of the class.
Betty White was incredibly good at this, by the way.
Felix, on the Odd Couple, wasn’t, but I don’t want to spoil that episode for you if you ever see it. He gave very intellectual clues instead of going for the simple, obvious thing.
When I watched the game, there were many times someone would get an answer…and it wouldn’t be at all obvious how it happened.
That’s why I would be very impressed with a computer that could play Password. We don’t really understand how it works. What is there in the way the person says, “bird” that can get “robin” one time and “eagle” the next and both be right? Some of it may be coincidence, some of it is clearly cultural, and some of it, I think, is very subtle physiology…narrowing of the eyes, raising the eyebrow on one side, leaning in…that kind of thing.
To my knowledge, nobody’s done that with a computer yet.
However, IBM has taken on a new challenge.
It’s a new software program that runs on a supercomputer, named Watson…and it plays Jeopardy.
You might think, right off, that computers playing Jeopardy would be easy. You hear a question (well, an answer) on the show, and you may rush to look it up on the internet, using a computer.
I usually know almost all the answers…there are some categories where I’m not as good, though. I don’t use a computer during Jeopardy: I could during final, and I certainly could if I was a phone-a-friend on Who Wants to be a Millionaire? I tested it with Millionaire for about a week: I never failed to find the answer online in the thirty seconds.
But it’s trickier than you might think. Looking up answers for Jeopardy would be all about understanding the question and phrasing the search.
Oh, and I’ll use question from here on out. Technically, on Jeopardy, they give you an answer and you give them a question. But the question really is what is the answer shown to you.
The questions often involve puns and tangential information.
You have to really understand the English language (and American English at that) to play Jeopardy, in addition to knowing the answers. Oh, and there is that pesky ringing in thing, as well…the timing of that gets a lot of people.
IBM has been working on Watson, a software program that understands human language in quite subtle ways and can respond in kind.
That’s very impressive!
Here, take a look at this video:
So, it appears Watson is going to play against human Jeopardy champions later this year. I assume Ken Jennings might be one of them.
It’s not going to be the knowledge that matters…it’s going to be understanding the questions.
It will be fascinating to watch. It will be a huge step forward in computers understanding people. We all know that’s coming, eventually. Will you be able to call a computer “advice nurse”? Computerized 911? Clearly you can’t if the computer doesn’t understand what you want. That means it has to understand language…imperfectly used language, at that, in a stressful situation.
That would be a ways away. Watson is a prototype right now. You aren’t going to get KITT-like computers in your car at a marketable price even if Watson wins.
Oh, and computers might beat Ken Jennings, but I think Betty White’s Password record is safe for years to come. 😉
UPDATE: We now have the dates! February 14-16 2011 and Watson will play Ken Jennings and Brad Rutter. http://engt.co/gfJY1r
This post by Bufo Calvin originally appeared in the The Measured Circle blog.