The opponents have felt each other out and learned they are all fallible. Tonight (Feb. 16) the two humans and the computer who have competed in "Jeopardy!" meet one more time, with their combined cash total from their two games determining the overall winner.
The first of the two "Jeopardy!" matches in the battle between brains and computer chips concluded last night, and Watson summarily smoked his human competitors, finishing with $35,734 to Brad Rutter's $10,400 and Ken Jennings' $4,800.
Although that puts the IBM-produced computer in a very strong position tonight, the computer's road to victory was not without a couple of bumps, and an expert in artificial intelligence says the human contestants have a chance if they can "outstrategize" Watson.
For an analysis of how Watson performed – and to assess the humans' chances – TechNewsDaily contacted Ted Senator, vice president and technical fellow at the Science Applications International Corp. Senator, a former "Jeopardy!" champ himself, is also secretary-treasurer of the Association for the Advancement of Artificial Intelligence.
Tied, but not for long
When yesterday's Double Jeopardy round began, Watson and Rutter — the biggest money-winner in "Jeopardy!" history — were tied at $5,000, and Jennings — who has recorded the game show's most victories — had $2,000. [Read: Post-Game Analysis: How Watson Did in Round 1 of 'Jeopardy!' ]
Jennings, with the lowest score, went first, but Watson got right response to the clue, then proceeded to rip the board to shreds.
Interestingly, as in round one the day before, once Watson had control of the board, he selected the fourth clue down in the category farthest to the left. In Single Jeopardy, the machine scored a Daily Double here.
"While it’s known that this square is the most likely location of a Daily Double," said Senator, "I wonder if the conditional probability of its being located there in Double Jeopardy, given that it was there in Single Jeopardy, has been calculated."
Watson had been trained to hunt for Daily Doubles, and after getting the first six clues of the round right, the computer hit paydirt. Watson wagered a hefty, but bizarre, $6,435 – human contestants never bother to place a Daily Double wager that isn't rounded off into the hundreds – and host Alex Trebek joked, "I won't ask." Watson nailed the question.
A small slip-up
The next clue, however, broke Watson's streak. It read: "In May 2010, five paintings worth $125 million by Braque, Matisse & 3 others left Paris' Museum of this art period." Watson answered "What is Picasso?" apparently mistaking the artist for an art period, given the "what" phrasing of the answer.
Rutter and Jennings both buzzed in and answered incorrectly as well, offering "Cubism" and "Impressionism." (The right response: "What is Modern Art?")
"Missing the type of answer seems to be a mistake that Watson makes and should be fixed in a future version," commented Senator.
Still in control, Watson answered the ninth clue of the round correctly before hitting upon the second and final Daily Double of the night – a crushing blow to his competitors, who at this point were well behind.
Watson busted out another head-scratching wager – $1,246 – and got this clue: "The ancient 'Lion of Nimrud' went missing from this city's National Museum in 2003 (along with a lot of other stuff)."
Watson, whose top three answer choices and their corresponding confidence scores are displayed on television and for audience members, was only 32 percent sure of his top answer. Cutely, the computer was programmed to say "I'll take a guess" before offering "What is Baghdad?" which turned out to be dead-on.
Leaving humans in the dust
Of the remaining 20 clues, Watson answered 14 correctly and none wrong; Jennings got three right and Rutter two.
One clue, however, revealed a particular linguistic vulnerability for Watson. The clue: "A Goya stolen (but recovered) in 2006 belonged to a museum in this city (Ohio, not Spain)."
The correct response, which Jennings got, was "What is Toledo?"
"Watson misses the 'not Ohio' part of the clue and has as its highest-ranked answer 'Madrid' – showing again its difficulty with negation and with complex patterns in multi-part clues," Senator noted.
Going into the Final Jeopardy clue, Watson could not lose as long as it wagered appropriately. The machine had $36,681 to Rutter's $5,400 and Jennings' $2,400.
A final-round botch
But chinks in Watson's armor were on full display in this final round. The category was "U.S. Cities." The clue: "Its largest airport is named for a World War II hero; its second largest, for a World War II battle."
Both humans wrote out, "What is Chicago?" the right response. Watson, however, wrote on his lectern, "What is Toronto?????" with the five question marks indicating the computer's lack of confidence in the answer, as Trebek explained to the audience. (To get an inkling of what Watson was "thinking," Toronto's biggest airport is named after Lester Pearson, who served as a Canadian diplomat in World War II.)
"It's the same mistakes as we've seen earlier," Senator said. "Watson doesn't use much information from the category name, and has trouble when it has to match two abstract patterns and find the intersection between them."
Yet Watson had bet only a paltry, if puzzling, $947. Its winnings dipped to $35,734, allowing it to close out its remarkable day still comfortably ahead.
Hope for humans?
Is it lights out for humankind in this battle of man versus machine? Senator weighed in.
"Other than hoping for the types of clues that Watson finds difficult, the humans have to outstrategize Watson – Watson doesn’t adapt to its opponents, it just plays the game," Senator told TechNewsDaily. "Perhaps if they make risky bets, or buzz in even when they're not sure" – a tactic Jennings tried twice yesterday, with mixed success – "they may have a chance."
"Or perhaps Watson will experience a catastrophic failure," Senator said. "Otherwise it wins."