“I never get your limits, Watson. There are unexplored possibilities about you.”
– Sherlock Holmes
Sherlock Holmes, I suspect, would have despised the computerized Watson on display this week on “Jeopardy.” The machine proved to be everything that the fictional Holmes prided himself on being — coldly calculating, decisive and unemotional — 0nly much more so.
As Ken Jennings and Brad Rutter, the two most decorated human champions in “Jeopardy” history, looked on in frustration, the IBM computer buzzed in first on question after question, answering most of them correctly and leaving millions in the television audience aghast that their fellow carbon-based life forms had been bested so easily.
It’s a funny thing: We don’t feel challenged by machines that are stronger than we are, or faster than we are. But we can get a little nervous about machines that we suspect might outsmart us. Intelligence is supposed to define us as a species — that’s why we’ve branded ourselves “Homo sapiens,” the knowing man — so machines that call our intellectual dominance into question can be a bit unsettling.
Or, as the defeated Jennings playfully scrawled on his Final Jeopardy answer: “I for one welcome our new computer overlords.”
Scientists, of course, tell us that it’s irrational to feel threatened by our own machines. They’re no doubt right — we’re just being silly. Silly and foolish. Silly little irrational foolish people, that’s us. If only we had a purely logical computer to do our thinking for us, right?
This has been a long time coming. Fifteen years ago, I had the chance to cover the historic six-game chess match between Garry Kasparov, the best chess player in history, against an IBM computer called Deep Blue. The machine was capable of analyzing 100 million chess positions a second, while Kasparov at his peak could analyze just three. Yet Kasparov somehow won that particular challenge, and at a press conference afterward he boasted about being the defender of mankind, beating back the onslaught of the machine.
A year later, he lost pretty badly in a rematch. The onslaught of the machine resumed.
In reality, there’s no danger of our machines outsmarting us, however entertaining we might find the idea. (As irrational beings, we do like to scare ourselves a bit). Think of it this way: According to IBM, Watson runs on 90 of the company’s Power 750 servers, and draws on a database roughly equivalent to a million books. In other words, it takes more than five tons of raw computing power to compete against a human brain of roughly three pounds. And while Watson is a single-purpose tool much like a potato peeler — its sole function is playing Jeopardy — the human brain is capable of doing so much more.
For example, building a machine capable of playing Jeopardy.
Bill Murdock, a 2001 Georgia Tech grad and an IBM researcher on the Watson project, concedes that it’s really no contest. “Absolutely,” he says, “human beings are really, really impressive.” In fact, he says, a different set of questions or categories in this week’s contest might have produced a human winner. In pre-match sparring against lesser Jeopardy champions, he says, Watson won just 71 percent of the time.
And as viewers saw this week, it doesn’t always produce the right answer. In a Final Jeopardy question about U.S. cities, Watson for some reason answered “Toronto.” Nobody is sure why, but using the deductive power available only to we humans, Murdock surmises that somewhere deep in its database, Watson has a list titled “American and Canadian cities,” in which Toronto is included. If the list had been properly titled “American OR Canadian cities,” Watson might not have made that mistake.
In other words, human error.
In our awe of our own technology, we sometimes forget that weakness. Watson and its offspring are not decision-makers; they are tools to assist us in decision-making. Like the Watson who assisted Sherlock Holmes, and the Watson who lent a hand to Alexander Graham Bell, computerized Watsons will always play second banana.