by Andrew Boyd
Today, Jeopardy!. The University of Houston's College of Engineering presents this series about the machines that make our civilization run, and the people whose ingenuity created them.
What comes to mind when you hear the word, "Watson?" For many, it's Sherlock Holmes's friend and narrator Dr. John Watson, made famous in the phrase, "Elementary my dear Watson." But there's a new Watson on the block, as reported in the New York Times.
Watson — it goes by only one name — is a computer that plays the popular trivia game Jeopardy!. Watson was developed at IBM's T. J. Watson Research Center, the same center that gave us the chess playing computer Deep Blue — the computer that beat world champion Gary Kasparov in 1997.
Deep Blue's victory caused quite a stir, but it didn't have much impact on the world of artificial intelligence. Deep Blue was a number cruncher's dream — a machine so powerful it could simply evaluate more moves than a human could ever hope to. What's really amazing is that human's can play chess so well without being giant number crunchers.
Jeopardy! is very different than chess. Consider the clue "A word meaning huge, but not necessarily woolly." The word 'huge' has many synonyms, so simply checking a thesaurus won't do the job. We need help from the word "woolly." It's not a very common word, and most of us have probably heard of it in reference to woolly mammoths. And there's what we're looking for: the word "mammoth."
The process by which we arrive at "mammoth" isn't easily expressed as a distinct set of rules, making it very difficult to program. The question, 'What is the capital of Turkmenistan?' is relatively easy for computers since it's an unambiguous request for a fact. But the game Jeopardy! isn't just about simple facts. The clues often involve complex word play.
So Watson's engineers haven't sought a single set of rules. Instead, their algorithms use many different sets of rules to arrive at many possible solutions. If Watson's sufficiently confident about a particular solution, it rings in with a response. In effect, Watson takes its best guess.
The approach isn't earth shattering — the real work is in the details. And, like Deep Blue, Watson takes advantage of tremendous computing power to draw on vast stores of data. But the design concept's quite different from that behind Deep Blue. And it seems to be working. Watson plays a pretty good game, and Jeopardy!'s producers have agreed to let Watson compete with former champions.
But whatever the outcome, Watson's still a long way from the Holy Grail of artificial intelligence, which is intelligence. Even the most upbeat researchers wouldn't claim Watson thinks. For now, and for as far as we can see, we'll just have to rely on that most remarkable question-and-answer machine — the human brain.
I'm Andy Boyd at the University of Houston, where we're interested in the way inventive minds work.
Notes and references:
For a related episode, see KASPAROV AND DEEP BLUE.
C. Thompson. 'What is IBM's Watson?' The New York Times Magazine, June 20, 2010.
The pictures of Blue Gene, the computer hardware that supports both Deep Blue and Watson, are from the Web site of Argonne National Laboratory. The Jeopardy! picture is from the game show's Web site.