nullWatson faces its human rivals in a practice round. Image: Jeopardy / IBM

Ken Jennings and Brad Rutter are accustomed to making others feel the heat as they blaze through Jeopardy clue after Jeopardy clue. But tonight, the quiz show's two greatest champions will oppose a player who can't be psyched out.

It's time for the world to meet Watson.

IBM's Jeopardy-playing computer system appears to viewers at home as an avatar of the Earth on a black screen. In fact, it is a system years in the making, and perhaps the most impressive attempt ever to create a question-answering computer that understands the nuances of human language.

Watson is not connected to the Internet, but its databases overflow with books, scripts, dictionaries, and whatever other material lead researcher David Ferrucci could pack in. Storing information is the computer's strong suit; the grand artificial intelligence challenge of Jeopardy is the subtlety of words.


When the bright lights of Jeopardy go up tonight, there will be no human handler to tell Watson where inside its mighty databases to seek the answers. It must parse each clue and category title to figure out what it's being asked. It must race through its databases, find relevant search terms, and pick out the right response with a high level of confidence. It must understand the puns and geeky quirks of America's Favorite Quiz Show. It must beat two Jeopardy champions to the buzzer. And it too must voice its responses in the form of a question.

Jeopardy is among the most human of games—much more so than the mathematical and closed system of chess, which IBM's Deep Blue mastered back in the 1990s. But is Jeopardy therefore the greatest gaming challenge for a machine? Not quite, AI researcher and computer science professor Bart Massey of Portland State University says. Checkers, chess, Scrabble, bridge, backgammon, poker, Stratego, and more—they've all seen software designers hoping to create the system that cracks the game. But the history of building computers to play games is the history of overestimating the rise of the machines and underestimating the strength of the brain. Watson could crush its all-star competition over the next three days, but computers even today lag behind the best human players in many of our favorite games—including some surprising ones.

The Grid: Where Computers Reign

Picture the playing areas of tic-tac-toe or Connect Four. They are small and two-dimensional; the two players take turns making their marks on the playing surface in an attempt to place three or four marks in a row, while preventing the opponent from doing the same. This elegant simplicity makes these diversions kids’ games, which players give up for more interesting pursuits (gaming or otherwise) as they grow up. The setup also makes these games perfect for computers.

There's no hidden info in tic-tac-toe, Massey writes in his notes (pdf) on AI and games. There's no luck, either. A tic-tac-toe-playing computer learns the same skill of forcing a draw if necessary that many a child has learned after plenty of losses. Most importantly, a computer can do what a child cannot: simulate every possible outcome and choose the perfect move. As a result, tic-tac-toe and Connect Four fall into the category of "solved" games. Computers play them perfectly. You stand no chance.

But what happens when the board gets bigger, and the rules a little more complex? Things get messier.

Next page: Computer programs face off against the masters of checkers, chess, and the ancient board game go.