The contest between man and machine on Jeopardy! was decided when IBM’s Watson computer landed on the second Daily Double on day three. The clue was: “This two-word phrase means the power to take private property for public use as long as there is just compensation.” Watson’s response: “What is eminent domain?”
The audience (mostly IBMers, since the show was taped in an IBM Research auditorium) went nuts. Former Jeopardy! champion Ken Jennings covered his mouth with his hands. He knew the contest was likely over. “Ken realizes he can’t catch up,” David Ferrucci, head of the IBM Research team that created Watson, commented at a viewing of the Jeopardy! segments on Monday morning.
A few minutes later it was over. But our wondering about its implications has only just begun.
Watson had shown some of his limitations during this segment. The computer totally misunderstood the category that required the answers to also be a type of key on a computer keyboard, and, in another category, Actors who Direct, he knew the answers but the humans beat him to the buzzer.
Yet the overall impression created by Watson during three days of competition was how “smart” he was–and not just how smart for a computer. He had beaten humans at a game that is considered to be a fair measure of human intelligence. Watson finished with $77,147, compared to $24,000 for Jennings and $21,600 for Brad Rutter, another top Jeopardy! champion.
But does Watson really think? This is a question that Ferrucci frequently hears from journalists. He’s wary of making bold pronouncements. When I asked him the question on Monday, he answered: “How do you define think? Does a submarine swim?” His point is that Watson doesn’t think the same way that we humans do, but he’s thinking none the less.
Thousands of little algorithms allow Watson to break down human language into pieces, search for answers in his vast database of information, and assess his confidence in the accuracy of the potential answers he comes up with. He answers questions without truly understanding them, but he answers them none the less.
A debate is raging about the future relationship of computers and humans. Will the machines become smarter than we are? In addition to the formidable coverage of Watson’s Jeopardy! match, The Atlantic and Time magazines took on the issue in recent cover stories. Ever since the first electronic computers were created in the 1940s, they have been compared to human brains. Now they’re actually capable of doing some very sophisticated thinking–as the Jeopardy! contest illustrates. And, in the not-to0-distant future, it seems likely, they will be smarter than us in a lot of ways.
The implications are huge. This crossing over, which computer scientist Ray Kurzweil calls “the singularity,” may not happen in 2045, as Kurzweil predicts, but it will happen in some form some day. We humans will have to prepare for it–agreeing to and enforcing a set of rules governing the use of computers similarly to the way we deal with genetics.
In the meantime, the Deep QA technology that’s the “brains” behind Watson is nothing to fear. It’s a tool for extending the capabilities of humans, not for replacing them or controlling them. Humans have faced these kinds of challenges with each major advance in technology over the short history of our species. We create machines to do some of the things we do quicker or better or cheaper– and we figure out something else to do with our strong arms and supple brains. We reinvent ourselves.
Ferrucci, for one, isn’t worried that computers will become our masters someday. “I think human intelligence will continue to subsume computer intelligence–not the other way around,” he said. “The Internet came along. It didn’t consume us. We consumed it.”
Some explanations of how Watson plays Jeopardy! from IBM researchers.
Here’s how Watson knows what it knows from Jon Lenchner:
Here’s a post on Watson’s wagering strategies from Gerald Tesauro:
Here’s some info on how Watson sees, hears and speaks from Dave Gondek: