UConn Witnesses Man vs. Machine in Jeopardy! Challenge

<p>Watson, powered by IBM POWER7, is a work-load optimized system that can answer questions posed in natural language over a nearly unlimited range of knowledge. Photo provided by IBM Watson</p>

The IBM supercomputer Watson can answer questions posed in natural language over a nearly unlimited range of knowledge. Photo provided by IBM

As more than 300 students, faculty, and staff poured into a lecture hall Wednesday night to watch an epic battle of man vs. machine, the air was thick with hushed, excited conversation.

On a massive dropdown screen, the audience watched as two of Jeopardy!’s best-ever human contestants – Ken Jennings and Brad Rutter – faced off against an expressionless, motionless (some might even say creepy) black avatar named Watson, an amazing new super-computer built by IBM with the assistance of professors and scientists from eight universities around the world.

Atif Rakin, a fourth-year UConn engineering physics major, was one of those who came to the UConn viewing party to witness history being made.

“This is like a lot of the science fiction stuff you read about as a kid,” Rakin said during a brief break in the show. “The idea of humans versus technology is amazing.”

UConn alumna Sharon Nunes, IBM vice president for smarter cities strategy & solutions, attended Wednesday’s viewing event and answered questions about Watson’s development and cutting-edge technology.

<p>Sharon Nunes, IBM VP and UConn alum, speaks with UConn students and staff to answer questions about "Watson," a supercomputer with the ability to play Jeopardy. Photo by Lauren Cunningham</p>

UConn alumna Sharon Nunes, a vice president at IBM, answers questions from UConn students and faculty about Watson. Photo by Lauren Cunningham

Watson, named after IBM founder Thomas J. Watson, has the ability to understand the subtle nuances of English language and process widely diverse bits of information in milliseconds in order to find the best answer and compete on the popular trivia game show.

“It stretches human imagination,” said UConn engineering dean Mun Y. Choi. “An event like this demonstrates there are limitless possibilities.”

Watson battled Jennings and Rutter in two Jeopardy! matches that spanned three consecutive days beginning on Feb. 14. On the first day, Watson broke out to an early lead, only to find itself tied with Rutter, with each earning $5,000 at the round’s end. Jennings trailed with $2,000. On the second night, Watson expanded its lead to $35,734, in an impressive surge that outpaced Rutter with $10,400 and Jennings at $4,800.

After Wednesday’s final round, Watson won the contest with $77,147 in total earnings. Jennings ended in second place with $24,000, and Rutter’s final tally amounted to $21,600. All of Watson’s winnings, which included a grand prize of $1 million, will be donated to charity. Jennings and Rutter are donating half of their earnings to charity.

“It’s a huge milestone,” said Carmen Gambardella, a junior engineering major, after watching Watson’s impressive performance.

IBM and Jeopardy! representatives were quick to point out that the computer was subjected to the same game play constraints as its human counterparts. Watson had each Jeopardy! question fed to it as text electronically, and had to process the information quickly before weighing in by pressing a button when it believed it had a correct answer. The computer – actually a mass of 90 IBM Power 750 high-intensity servers running the Linux operating system – was represented on the Jeopardy! set by an avatar patterned after IBM’s Smarter Planet icon and provided guesses in a digitized voice.

Watson took four years to build, at an estimated cost of up to $2 billion (IBM isn’t saying exactly how much). The computer was not connected to the Internet: it had to rely on 200 million pages of natural language content (the equivalent of reading 1 million books) fed into it by its creators. These included encyclopedias, the Bible, the Internet Movie Database, archives of The New York Times, and thousands of previous Jeopardy! clues.

All that information was tucked inside 16 Terabytes of memory that was repeatedly scoured by IBM’s commercially-available POWER7 processing cores running at 3.55GHz and armed with IBM’s advanced Deep QA (Question Answering) technology to enable Watson to offer a response in less than three seconds. The system has the ability to operate at more than 80 Teraflops (millions of operations per second).

WatsonAvatar_lg

The avatar for Watson, IBM's latest supercomputer.

Jeopardy! presented a unique challenge for Watson’s computing capabilities because of its variety of question types, broad range of topics, and reliance on the nuances of natural language. In order to compete, the supercomputer used advanced algorithms to analyze irony, subtle meaning, and riddles in a way that computers traditionally cannot. In doing so, it was able to process an enormous number of simultaneous tasks and data while analyzing information in real time.

“I found it really impressive how quickly Watson could decode each question,” said Christian Wilkie, a senior majoring in electrical engineering.

IBM has suggested that a computer’s ability to process this kind of data represents remarkable progress in natural language processing. The technology behind Watson could be adapted to solve problems in various fields such as health care, where such systems could help medical professionals accurately diagnose and treat patients by quickly analyzing large quantities of the latest available trade and research information. Other possible applications include improving online self-service help desks, providing tourists with information, and prompting customer support via the telephone.

Robert McCartney, an associate professor of computer science and engineering and an expert on artificial intelligence, said Watson’s technology “would be useful where you have little bits of information you can’t remember. Systems like this will find answers that are close to the answer you’re looking for.”

But Watson was not infallible. In its initial Jeopardy! go-around, noted Christine Wakefield, a fourth semester chemical, materials, and biomolecular engineering major: “Sometimes Watson would have the answer right away, but sometimes he wasn’t even close.”

Watson’s lack of familiarity with the human experience also showed during the second game, when it was presented with the clue “A loose-fitting dress hanging straight from the shoulders to below the waist.” Watson responded, “What is a chemise?” The correct answer was “What is a shift?”

The audience was able to watch Watson “think” during the contest, through the use of a graphic at the bottom of the television screen. This “confidence bar” displayed the computer’s confidence level in preparing three possible best answers. If it felt none of its immediately available options was clearly the answer, the system took longer to press the buzzer and respond. This disadvantage allowed Jennings to build a short lead in the second game, until Watson bounced back with a string of quick, correct answers to riddles.

Morad Behandish, a graduate student in mechanical engineering, saw Watson’s arrival as a glimpse of the future.

“What I am wondering is if humans can one day create a computer that’s almost as capable as the human brain,” Behandish said. “This has been a debate for a while, and I wonder if I can see it in my lifetime.”

The viewing party was sponsored by the schools of business, education, engineering, fine arts, liberal arts and sciences, nursing, and pharmacy, the College of Agriculture and Natural Resources, and UITS.

***********************************************************************

Would you like to test your trivia knowledge against a computer? Check out The New York Times Magazine‘s interactive website, where players can test their skills against Watson.

Additional information about how Watson was created and what it means for computer technology can be found on the IBM website.