What if there were a computer that could answer virtually any question? IBM engineers are developing such a machine, teaching it to compete on the quiz show Jeopardy. In February 2011, it will face off in a nationally televised game against two of the game’s greatest all-time winners, Ken Jennings and Brad Rutter. Final Jeopardy tells the riveting story behind the match.
Final Jeopardy carries readers on a captivating journey from the IBM lab to the podium. The story features brilliant Ph.D.s, Hollywood moguls, knowledge-obsessed Jeopardy masters — and a very special collection of silicon and circuitry named Watson. It is a classic match of Man vs. Machine, not seen since Deep Blue bested chess grandmaster Garry Kasparov. But Watson will need to do more than churn through chess moves or find a relevant web page. It will have to understand language, including puns and irony, and master everything from history and literature to science, arts, and entertainment.
At its heart, Final Jeopardy is about the future of knowledge. What can we teach machines? What will Watson’s heirs be capable of in ten or twenty years? And where does that leave humans? As fast and fun as the game itself, Final Jeopardy shows how smart machines will fit into our world — and how they’ll disrupt it.
Q: What did you come to most admire about the researchers working to develop Watson?
A: I found myself admiring their meticulous engineering. I’ve always enjoyed stories of great engineering, from the building of the Panama Canal to the rescue of Apollo 13. The work on Watson fits into that genre. It involves continual problem-solving, innovations, incremental improvements, and above all, endless patience. To do this work, the Jeopardy team had to break down the way we think, the way we understand sentences and concepts and facts, into tiny components, and then teach them to Watson.
I have to say, I came out of this process with an ever greater appreciation for the magic that takes place between our ears. We don’t give ourselves enough credit. Seriously. They had to build a roaring complex of computers and supply it with enough electricity to light up an entire town, all this just to approximate the question-answering part of the organ we carry in our heads. Unlike Watson, our thinking machines can run for hours on just a cup of coffee and a donut.
Q: What was most surprising to you about Watson’s behavior?
A: Two things: First, its speed. When researchers describe all of the work it takes for a machine to make sense of a question and hunt down possible answers, it makes perfect sense that the process would take a computer two hours. And in the beginning, it did. The fact that they engineered that two-hour process into a mere three seconds is astounding.
The second surprise was that Watson could be so amazingly smart on one question, then laughably clueless on the next. How could it ever conclude that the Russian word for good-bye would be "cholesterol"? How could it confuse Oliver Twist with the Pet Shop Boys? But you know what? I found that when I watched Watson screw up, I had even greater appreciation for the work involved when it got things right. If it got everything right, Watson wouldn’t be the fallible (and entertaining) machine that it is. It would just be magic--which really is not nearly as impressive.
Q: So who is in charge of picking the clues for the final match? Do you think the arrangements for the match are fair?
A: In the end, Jeopardy chose 30 games that the writers had prepared for the Jeopardy season that began in the summer of 2010—before they knew for sure that a man-machine match would take place. Each of the games was given a number. Then they had an outside compliance company select two of the games by number. I think they’ve made the game as fair as possible.
I should add that Watson’s scientific test comes from a bigger set of matches. The machine took on human champions in 56 matches in the fall of 2010. It won a majority of those matches. And for the field of question-answering, those games mean more than the televised showdown, simply because there are more of them.
Q: If you had to place bets, who do you think will win the match?
A: I would bet on the computer, because there’s only one computer facing two humans. Ken Jennings and Brad Rutter will compete against each other in areas where humans are presumably strongest—clues associated with puns, humor, and difficult context. The computer, I would think, might dominate in the factual realms it "understands." That said, I’d never wager much on one game. One lucky Daily Double or Final Jeopardy can decide a single match.
Q: Should people who don’t watch Jeopardy care about this story? If so, why?
A: Jeopardy is just a showcase for a new type of machine. Look, we’re going to be living with these things, working with them, and using them as external lobes of our brains. Final Jeopardy follows the education of one such machine. Readers, I’m hoping, will get a feel for its potential as well as its limitations. And that will help them understand what skills and knowledge they’ll need to carry around in their own heads. Of course, I’m also hoping they’ll enjoy the story.
Q: Doesn’t Google already answer all our questions? What makes Watson so special?
A: Google is so useful that we sometimes forget how much more it could tell us. First, it doesn’t answer questions. It usually just points in the direction of the information we’re looking for and leaves the rest of the brain work to us. Watson, by contrast, puts together pieces of information from different sources and comes up with answers or hypotheses. A simple example. Let’s say you want to buy a dog for your aging parents. It has to be small, quiet, well-behaved, able to tolerate long periods indoors, and friendly enough to sit on a lap. A hunt for such a dog on a search engine would require lots of different searches (unless someone had happened to write an article about precisely that challenge). But a machine like Watson would understand the sentence in English, read through several thousand documents, and put together a list of candidate dogs. It acts much more like a knowledgeable person.
Q: Did Watson become more or less "human-like" to you over the course of the project?
A: It’s funny. When it’s playing Jeopardy, I find myself referring to it as “him .” But when you hang around with the researchers, the human evaporates and Watson returns to its truer form, that of an enormously sophisticated computer program. Sometimes it’s tempting to think of Watson as a “brain.” But as you deal with it, you see that it’s not even close to a full brain. It just handles information retrieval and question-answering. In short, it’s a tool.
Q: How does the rest of the computing world see Watson? Is this the true path to Artificial Intelligence?
A: Many folks in AI resent Watson, some of them to the point of loathing. You see, there’s this dream of building machines that blend the intelligence of humans with the data-processing wizardry of computers. For people who hold that vision—and many still do—Watson is almost a false prophet. In the realm of Jeopardy, it acts like a human. But instead of processing information the way we do, it just crunches numbers. It doesn’t really know or understand anything, or come up with ideas. Yet it works—and it gets to parade its smarts on national TV! If we as a society settle for machines like Watson, will we continue to fund the ambitious research that seeks to replicate the magic of the human brain? For many, that’s the true path to AI. But as far as I’m concerned, there doesn’t have to be just one path.
Q:How did IBM decide what to name Watson and who created its public image?
A:There was lots of debate within IBM about Watson’s name and image. How human should it be? Many worried that the public would view Watson as scary: a machine that learns our secrets and steals our jobs. So they decided to limit Watson’s human qualities. They would give its friendly, masculine voice a machine-like overtone. And its face, if you could call it that, would simply be a circular avatar—no eyes, nose or mouth, just streaming patterns representing flowing data. Despite these choices, I’ve noticed that fellow Jeopardy players immediately start to respond to Watson as another human—and not necessarily a friendly one. It’s playing the game, after all. And it usually beats them.
As far as the name, IBM entertained loads of possibilities. They considered THINQ, Ace, even EureQA, a blend of Eureka with QA, for question answering. In the end, they picked Watson, for IBM’s founder, Thomas J. Watson. In the literary world, it also fit into the stories of Sherlock Holmes, a master question-answerer. Of course, in those stories, Watson was only the assistant to the true genius. But considering the widespread fears surrounding smart computers, maybe it made sense to name the question-answering machine after Holmes’ plodding number two.
Q:If Watson-like systems do become ubiquitous, what will that mean for humans?
A:There’s no question about it. Machines like Watson are going to become part of our lives. They’ll be manning call centers, answering questions in offices, factories and emergency rooms. And they’ll be available to all of us through our smart phones, often answering spoken questions. This intelligence will become, de facto, part of the human brain. Each one of us will have to figure out how to leverage these smart systems for our own good—and not be replaced by them. Our brains are still the most intricate, complex and brilliant thinking machines on earth. But we have to figure out how to use them in concert with the machinery we’re building.