TOPICS > Science

A: This Computer Could Defeat You at ‘Jeopardy!’ Q: What is Watson?

February 14, 2011 at 5:34 PM EDT

GWEN IFILL: Man versus machine, scientists just love putting that notion to the test. Now a competition is under way to judge whether the latest and greatest supercomputer can actually think. The face-off occurs this week on the popular game show “Jeopardy.”

And NewsHour science correspondent Miles O’Brien jumps right in.

MILES O’BRIEN: Now, I’m not a guy to make a lot of excuses, but I didn’t get much sleep the night before I found myself here, getting ready to play Watson, arguably the smartest computer in the world, in a game of “Jeopardy.”

MAN: Same category, $1,200.


MILES O’BRIEN: That’s David Ferrucci, Watson’s proud papa.

DAVID FERRUCCI, Watson Project, IBM: So, you’re looking at 10 racks of power 750. So, there’s 10 racks. There’s 90 what they call power 750 servers.

MILES O’BRIEN: He introduced me to his silicon progeny.

DAVID FERRUCCI: So, overall, there’s about 2,880 cores in that system, about 15 terabytes of RAM.

MILES O’BRIEN: For those of us who don’t have a doctorate in computer science, Watson is equivalent to about 6,000 high-end home computers. But the secret sauce is the software that gives Watson the ability to understand language like no computer ever has.

MAN: Kathleen Kenyon’s excavation of this city mentioned in Joshua showed the walls had been repaired 17 times.


WATSON, supercomputer: What is Jericho.

MAN: Correct.

MILES O’BRIEN: Well enough to play “Jeopardy” at the highest level with top money winner Brad Rutter and Ken Jennings, who won 74 games in a row in 2004. Jennings’ amazing run caught the nation’s attention, including some IBM executives, who were looking on for a follow-up to their man-versus-machine chess triumph. The computer they called Deep Blue beat grandmaster Garry Kasparov in a celebrated tournament in 1997.

They wondered if a machine could beat the best humans at “Jeopardy.”

DAVID FERRUCCI: We knew it would be hard. We knew it wasn’t be easy. But to have an opportunity to get the resources to sort of push the limits in advance of technology was just irresistible.

MILES O’BRIEN: Getting machines to truly understand language is the Holy Grail of a field called artificial intelligence — you know, making computers more like us, able to comprehend, learn and solve problems. In the early days of computing, it seemed so easy.

MAN: I confidently expect that, within 10 or 15 years, we will find emerging from the laboratories something not too far from the robot of science fiction fame.

MILES O’BRIEN: Artificial intelligence is already here. It is used to make more accurate weather forecasts. It decides what movies and books you might like. But a computer that can match human intellect remains an elusive goal.

MARVIN MINSKY, Massachusetts Institute of Technology: There’s still no machine that can solve everyday commonsensical problems.

MILES O’BRIEN: Marvin Minsky of MIT says that with a healthy dose of chagrin. He’s one of the fathers of artificial intelligence. And he told me his colleagues got on the wrong track a few decades ago, trying to create a single mathematical model of the human brain.

MARVIN MINSKY: I’m a little disappointed that most people look for the magic bullet: What’s the trick that will make machines more intelligent?

And it seems to me that we know from brain science, if you look at the brain, it’s like 40 different computers. In fact, if you look in a big book on neuroscience, you will find maybe 300 or 400 descriptions of different parts of the brain that do different things.

MILES O’BRIEN: That’s the way David Ferrucci sees it. He and his team of two dozen wrote many formulas or algorithms to teach Watson language skills.

DAVID FERRUCCI: Language is not going to emerge from a silver bullet. There’s not going to be one algorithm that just understands language. It’s going to be a lot of different algorithms. They’re going to look at and interpret the language from different perspectives. And somehow, we’re going to be able to combine them.

MILES O’BRIEN: It took four intense years for them to write all the algorithms that make the machine Ken Jennings-ready. The behind-the-scenes drama played out in the PBS “NOVA” special “The Smartest Machine on Earth.”

MAN: Administrative Professionals Day and National CPAs Goof-Off Day.


WATSON: What is holiday?

MAN: No, that’s not even close, really.

MILES O’BRIEN: For a computer, “Jeopardy” is much, much harder than chess. Deep Blue beat Kasparov by playing out every possible outcome of every possible move every time.

RAY KURZWEIL, author and futurist: It’s just doing a logic puzzle very quickly. And computers are good at that.

MILES O’BRIEN: Futurist Ray Kurzweil is author of “The Age of Intelligent Machines” and several other books on the rise of artificial intelligence. He’s impressed with Watson’s ability to understand something as nuanced and complex as human language.

RAY KURZWEIL: It gets a query involving metaphors and puns and similes and jokes and other cultural references. It has a wide knowledge base. It can parse these complex statements that have different attributes organized in a hierarchical fashion.

MILES O’BRIEN: Watson cannot be connected to the Internet when it plays “Jeopardy.” It wouldn’t be fair. So, the team filled its memory banks with the entire “World Book Encyclopedia,” Wikipedia, the Internet Movie Database, much of The New York Times archive and the Bible.

But synthesizing all the data is the key. To do that, the team turns to a technique called machine learning, which teaches computers by example. Rather than trying to define the letter A, programmers instead give the machine millions of examples, and it figures out a unifying pattern, so it can recognize an A that it has never seen before.

Watson also ingested thousands of correctly answered “Jeopardy” questions, so it could learn the patterns of success in the game.

Do you find yourself wanting to call Watson he?

DAVID FERRUCCI: I make an effort to call it “it.” But yes,occasionally,I slip into the “he.”

MAN: To waste little by little or applying (OFF-MIKE) similar to the (OFF-MIKE)?



MAN: (OFF-MIKE) looking for.

MILES O’BRIEN: Damn. Damn, he’s fast. OK.

He — I mean it — is a formidable foe, as I was learning. Did I mention I didn’t have a very good breakfast that morning?

WATSON: Let’s finish the northern most capital city.

MAN: Let’s do it.

Manila, Kathmandu, and Jakarta.


WATSON: What is Kathmandu?

MILES O’BRIEN: He was fast on that one. He knew that category.

Let’s do Presidential Rhyme Time for $200, please.

MAN: Here we go.

Barack’s Andean pack animals.


WATSON: What is Obama’s llamas?

MAN: Obama’s llamas.


MAN: Yes, got it. That’s what this category is all about.

MILES O’BRIEN: It amazed me how Watson gets all the jokes, the wordplay and the puns that are hallmarks of “Jeopardy.” And Watson gets smarter with each answer.

DAVID FERRUCCI: And it learns based on the right answers how to adjust its interpretation. And now, from not being confident, it starts to get more confident in the right answers. And then it can sort of jump in.

MILES O’BRIEN: So, Watson surprises you?

DAVID FERRUCCI: Oh, yes. Oh, absolutely. In fact, you know, people say, oh, why did it get that wrong? I don’t know. Why did it get that right? I don’t know.

MILES O’BRIEN: Computers that learn, understand and even surprise us? Hmm. What could go wrong with that?

KEIR DULLEA, actor: Oh, HAL, do you read me? Do you read me, HAL?

DOUGLAS RAIN, actor: Affirmative, Dave. I read you.

MILES O’BRIEN: Oh, yes, there is that: a machine that becomes a psychopathic murderer.

KEIR DULLEA: Open the pod bay doors, please, HAL.

DOUGLAS RAIN: I’m sorry, Dave. I’m afraid I can’t do that.

RAY KURZWEIL: Certainly, artificial intelligence can be destructive. It’s already used in our weapons. We have smart weapons that have their own intelligence. We can send a cruise missile across the world, and it intelligently navigates and makes its own decisions.

Technology can be destructive, particularly in the wrong hands. The positive side is that these tools can help overcome human suffering, help cure disease, alleviate poverty, solve the energy problem, clean up the environment. I mean, there’s a lot of good things we can do with more intelligent technology.

MILES O’BRIEN: If you get a machine that fully understands language and can learn, you just sit back and watch, right?

DAVID FERRUCCI: Right. But we’re not — we’re not — we’re not quite there yet. We’re not there where we can — we can get a computer that completely understands language.

And there are all different ways of learning. And this is — you know, this way of learning is very powerful, and we have made great advance with it. But if you compare that to the enormity of the ways that humans learn, again, just scratching the surface.

MILES O’BRIEN: So, if Watson is only scratching the surface, where does that leave me? Pretty grim, eh? Did I mention I’ve had some terrible stiffness in my thumb lately?

GWEN IFILL: You can watch all of Miles’ apparently futile battle with Watson on our website.