Domo arigato, Mr. Watson: IBM computer takes on 'Jeopardy'

Years of research and development went into IBM Research's Watson, a computer with the ability to process human language in such a way that it's capable of challenging the most formidable game-show knowledge champions.

Caroline McCarthy Former Staff writer, CNET News
Caroline McCarthy, a CNET News staff writer, is a downtown Manhattanite happily addicted to social-media tools and restaurant blogs. Her pre-CNET resume includes interning at an IT security firm and brewing cappuccinos.
Caroline McCarthy
5 min read

Former 'Jeopardy' all-time champions Ken Jennings (left) and Brad Rutter flank a TV screen connected to Watson, a computer capable of challenging and potentially beating them in the famed game show. Caroline McCarthy/CNET

YORKTOWN HEIGHTS, N.Y.--The first words publicly spoken by a talking computer named Watson were, "WHAT IS JERICHO?"

Watson was following the rules. Like any contestant on game show "Jeopardy," the IBM Research-built machine was required to phrase his answer in the form of a question. And Watson was playing "Jeopardy." More specifically, it was a test run this morning at IBM Research's headquarters in preparation for a televised weekend challenge against famed "Jeopardy" champions Ken Jennings and Brad Rutter, and Watson nearly shut out those champions in a category about female archaeologists called "Chicks Dig It."

But then in the category's final answer, Ken Jennings--who holds the record for winning the most consecutive "Jeopardy" games in a row--bested Watson. Jennings' question-phrased response: "What is a Neanderthal?"

It was a fitting start to the test match, because what Watson is really all about is what separated us from the Neanderthals in the first place: the evolution of high-level intelligence, the complexities of the Homo sapiens brain, and the depths of human cognition--and whether we have finally begun to crack the code in creating a human-built machine that can start to approach this kind of grasp on language.

From folklore hero John Henry's Pyrrhic victory against a steam-powered steel hammer to Arthur C. Clarke's HAL 9000 and the Superman foe Brainiac, the narrative of human versus machine has fueled sentiments of both excitement and fear. Now there's Watson, named for IBM founder Thomas J. Watson, whose creators at IBM say is a hallmark of far more significant accomplishments than being able to challenge both Jennings and fellow "Jeopardy" champ Brad Rutter in a question-and-answer match. They also want the world to know he isn't evil (yet).

"We are at a very special moment in time here," IBM Research director John E. Kelly III told the audience. "We're at a moment in time where computers and computer capability has approached in this dimension the ability of humans, and the fact that the demonstration was so close shows that these two beings, these two lines are crossing. What will happen in the final tournament, we don't know."

Previewing IBM's Jeopardy Challenge (photos)

See all photos

Watson cannot see or hear, which means it cannot handle "Jeopardy" audio or visual clues, but it can wager on Daily Doubles. It can read and speak, and is packed full of dictionaries, reference books, thesauri, and an impressive literary canon, but it is not connected to the Internet and therefore has no Google-enabled advantage over a human competitor.

The game's prompts are fed to Watson, whose hardware fills an entire room underneath the auditorium in which the televised "Jeopardy" challenges will take place, in plain text at the same time that they appear onscreen for the human contestants. Over the past four years, IBM researchers improved the computer's capability first to answer enough questions correctly so that it would keep a positive score in the game, then to respond correctly at the rate of the average "Jeopardy" player (about 60 percent accuracy), and then enough to challenge "Jeopardy" legends like Jennings and Rutter.

In Thursday's test match, Watson narrowly beat Jennings, who made up for his lack of knowledge of female archaeologists like Mary Leakey and Dorothy Garrod by dominating a "Children's Books" category, and more soundly defeated Rutter ("Watson and I don't have kids," Rutter surmised afterward with regard to Jennings' dominance in the kiddie-lit category).

IBM's legacy of machine-versus-man gameplay goes back a decade and a half: In 1996, an IBM-built computer called "Deep Blue" became the first machine to beat a reigning world chess champion when it defeated the Russian chess master Garry Kasparov; Kasparov won the overall match. An upgraded version of Deep Blue then soundly defeated Kasparov the following year, and the defeated chess champion accused IBM of breaking the rules by giving the computer some kind of human assistance.

"Off of that we got tremendous computer science learning that is applied to our systems today," Kelly explained, saying that he believes the research behind Watson will "have impacts on society far beyond the latest widgets that people are worrying about today."

That's because IBM, which is currently commemorating its 100th anniversary, believes that Watson represents a groundbreaking innovation in artificial intelligence because of its ability to process the complexities, nuances, and subtleties of human language. This is something that could have far-reaching implications in fields as varied as academia, government, and particularly health care.

"We've created a system that can interact in a very, very special way," Kelly said. In artificial intelligence, "People spend their lifetimes trying to advance that science inches. What Watson does, and has demonstrated, is the ability to advance the field of art intelligence by miles. People who are experts in this area who have seen Watson privately said, 'I never thought I would have seen this in my lifetime.'"

It was a significantly bigger challenge to build than Deep Blue, according to David Ferrucci, the IBM Research "investigator" in charge of building DeepQA, the "massively parallel probabilistic evidence-based architecture technology" that powers Watson and has been in development since 2007.

"'Jeopardy' is a very, very different challenge than chess," Ferrucci said. "In chess we have this finite, mathematically-defined search space, very precise rules, and we need a very different sort of algorithm...It's explicit, it's unambiguous, it's exacting. When we deal with language things are very, very different."

Watson's big challenge will come later this week when he makes his television debut, which Kelly called "an excellent challenge against which we can measure the progress of this system." It's also a charitable endeavor: Rutter and Jennings will donate half their winnings to charity, whereas every dollar won by Watson will be donated to either humanitarian organization World Vision or IBM social research foundation World Community Grid. And when the game is over, IBM Research will turn to investigating Watson's more serious purposes.

As it turns out, Watson also has some comedic skills, eliciting some chuckles from the audience when, after successfully responding to four prompts about female archaeologists, declared in his computerized monotone, "Let's finish 'Chicks Dig It.'"

But then, as it turns out, Jennings snagged that $1,000 answer. Neanderthals, indeed.