Two 'Jeopardy!' aces to square off against computer

The future bragging rights of the human brain depend, in part, on a Lancaster college dropout-turned-actor and comedian.

This week, millions will be watching Jeopardy! to see whether an IBM computer system named Watson can defeat the quiz show's two best players ever.

Since the showdown involves interpreting difficult language, experts say it is far more significant than the much-ballyhooed triumph of IBM's Deep Blue over world chess champ Garry Kasparov in 1997.

"It is a big deal," said David Blei, a machine-learning specialist at Princeton University. ". . . I'll be glued to the TV."

Two games will be played during three shows, airing Monday through Wednesday, seen locally at 7 p.m. on 6ABC.

The famous contestant is Ken Jennings, who, as a software engineer from Salt Lake City, won a record 74 straight games in 2004.

The other guy is Brad Rutter, 33, a Manheim Township High School grad who was working at a Lancaster record store in 2000 when he began a Jeopardy! streak of his own.

He had left Johns Hopkins University without finishing his history degree - "School and I have never really gotten along," he said - but still won five games, the limit then, collecting $55,102, along with two Chevrolet Camaros.

Then he won three straight showdowns against former champs, outdueling Jennings to win the 2005 "Ultimate Tournament of Champions" and $2 million, making Rutter the show's top money winner ever.

He won, in part, because he really got into a "zone," anticipating when to buzz in, he said.

A few years ago, he headed for Hollywood, seeking acting roles and doing stand-up at comedy clubs. He does a "pretty passable" impression of Sean Connery as a bumbling Celebrity Jeopardy! contestant, à la Saturday Night Live.

Now he's going up against what Nova, the PBS science show, called the "Smartest Machine on Earth."

Top prize is $1 million, with $300,000 for the runner-up, $200,000 for third. The humans have pledged half of their haul to charities, IBM all of its.

Rutter, a voracious reader as a child, has earmarked his dough to go to Lancaster County libraries for technological improvements. "Hopefully, somebody else growing up in Lancaster can have the same kind of passion sparked," he said.

On a normal night, nine million people watch the nation's most popular syndicated quiz show, a spokesman said.

Viewers will see Watson's "avatar," a glowing globe with radiating streaks, on a screen behind the center podium. All the hardware, the size of 10 refrigerators, would never fit. With humanlike speech, the computer, named for legendary IBM president Thomas J. Watson, will politely pick categories and respond - in the form of questions, of course.

The avatar's lights will vary blinking rates and colors to suggest how hard Watson is thinking and how confident it is, as its top responses are shared on TV screens.

 Comprehending a clue and fetching the correct response is no easy task, especially for a computer. Unlike chess, language has been an unexpectedly tough nut for computers to crack - even if children can. Add in the tortured syntax and wordplay common on Jeopardy! and traps and pitfalls arise.

Consider this actual clue: "This designer's Opium intoxicates me with its blend of tangerine, plum & cloves." It sounds as if it's about fruit or drugs, and the category title - "You Smell Yummy!" - is little help. (The correct response: "Who is Yves St. Laurent?")

In early auditions for Jeopardy! producers, Watson repeated opponents' wrong answers, got stumped by Roman numerals, and even had trouble distinguishing he and she, suggesting Richard Nixon was once first lady.

 But IBM's brainstormers plowed on, and Watson, designed to learn from its mistakes, kept improving. It helps to have memory banks containing 200 million pages of references, including entire encyclopedias and newspaper archives, and 15 terabytes of RAM or "thinking" memory - a couple thousand times what's in a good PC - operating at 80 teraflops, or 80 trillion calculations per second. Like the players, it has no access to the Internet.

"This is a bigger deal than Deep Blue," said Mitch Marcus, artificial intelligence professor at the University of Pennsylvania, calling chess "a simple game with a complicated strategy with very little extension to anything else in the world."

Systems like Watson could dramatically help in fields such as medical diagnostics, said Katharine Frase, an IBM vice president who graduated from Bryn Mawr College and got her Ph.D. from the University of Pennsylvania.

"The medical literature is changing so fast, it's almost inhuman to think that anyone could keep it all at the front of their brains all the time," she said. So a question-and-answer system that quickly tells the smartest possibilities should be invaluable.

"This is truly, we think, a leap forward in having a computer understand how we as humans use language, how we actually write or speak," Frase said.

Not that Watson is any kind of jack-of-all-trades know-it-all. It doesn't hear the clues - they get delivered electronically - and no audio and video clues will be used, since Watson is geared for text, not speech or visual analysis.

But that will come, with strides being made all the time in speech and image recognition.

Adding humanlike qualities - a name, a voice, a "face" of sorts - doesn't mean Watson thinks like us, says Frank Lee, a cognitive science specialist at Drexel University.

"It is like P.T. Barnum. It's for show," he said.

Also, he prefers to see machines as tools, not rivals.

"To show that it could beat a person is playing on this deep innate fear that people have of technology . . . taking over. But if you take away the layers of the show, there really isn't anything to worry about at all."

Watson, designed with Jeopardy! in mind, can't handle problems beyond its programming or its store of knowledge - but humans can and do every day, Lee said.

Marcus agreed, contending Watson couldn't pass a college test cold. "It would fail almost anything, because most tests require some reasoning," he said.

"This is a kind of idiot savant that is better than any human being alive at retrieving information," he said. ". . . It's a trained monkey to do a task that has a lot of real-world applications."

For an ultimate TV test, wait till a machine can win Survivor, he said.

It requires "a combination of physical things, mental things, and reasoning about what other people are thinking," Marcus said. ". . . Understanding people's intentions would be remarkably hard for a computer to do."

 As to who won, Rutter's lips were sealed, of course.

"What I can say is that I'm happy with how I did," he said.

Not that strong compu-tition wasn't deflating.

"It's just human to feel that way," Rutter said.

But then he found a more philosophical perspective: "It was people who built Watson and programmed it, built the algorithm, and so on, so I guess humanity wins no matter what happens."