In the second day of his appearance on TV's Jeopardy, IBM's Watson supercomputer laid waste to his human opponents. But even Watson, with 90 IBM Power 750 servers, has weaknesses.
On Tuesday, the Smartest Computer in the World blew past his carbon-based opponents, champions Ken Jennings and Brad Rutter. At the evening's conclusion, the computer had racked up $35,734, while Rutter had $10,400 and Jennings $4,800. At the end of Monday's game, Watson was tied with Rutter at $,5000 each, with Jennings at $2,000. The match concludes Wednesday.
'Fundamentally Different' from Humans
The winner gets $1 million. Since Watson is not designed for shopping, IBM will donate the prize to charity if its creation wins. The two humans have said they'll donate half if they manage to pull it out on Wednesday.
Eric Nyberg, a Carnegie Mellon University professor of computer science who headed a CMU team that helped IBM, has noted that Watson's responses are "fundamentally different" from humans. For instance, a human contestant can buzz and take a few seconds to answer, while Watson will only buzz when it has a high degree of confidence that the answer is correct. Watson can process and evaluate two million pages of information in three seconds. But if the question is short, Watson is at a disadvantage because it takes the full three seconds on every question.
More difficult questions can throw Watson off. In a brief moment of cheer for human partisans, Watson whiffed badly on the Final Jeopardy question Tuesday. The category was U.S. Cities, and the answer clue was: "This city's largest airport is named for a World War II hero; its second largest, for a World War II battle." The correct response was "What is Chicago?" Both humans answered correctly. Watson, who has never spent much time in airports, much less U.S. ones, answered "What is Toronto?"
On its Smarter Planet blog, IBM's David Ferruci, Watson's project manager, was quoted as saying that the computer had learned that category names "only weakly suggest the kind of answer that is expected, and, therefore, the machine downgrades their significance." Additionally, he noted, there are cities in the U.S. named Toronto, and the Toronto in Canada has an American League baseball team.
Watson's Post-Jeopardy Career
But the good news for Terminator-wannabes is that Watson is continually learning how categories work in Jeopardy. It has also learned to be careful about bets. As its level of confidence for Tuesday's Final Jeopardy question was low, it only bet $947 of the amassed earnings.
On the IBM blog, IBM's Steven Hamm noted some of the uses for Watson's intelligence after it calms down from Jeopardy fame. In medicine, he predicted, "a patient may describe to a doctor a certain symptom or a high level of pain, which, on the surface, may seem to be an important clue to the cause of the ailment." But Watson could discover, from sifting through and weighing its huge data store, that another symptom could be the real key to the diagnosis.
Other applications could include sifting through case law to find relevant precedents, helping governments process requests by social-services clients or businesses, assisting intelligence agencies in the analysis of streams of information, or lending a hand at call centers.