Alumnus Invents Computer that Wins at Jeopardy

David Ferrucci ’83 and his team from IBM created a computer called Watson.

It took a team of 25 engineers four years to make answering Jeopardy questions elementary for a computer named Watson. But the team managed just that as IBM’s Watson started slowly in a three-game competition that aired in February 2011 against two of the leading Jeopardy winners of all time, and won handily by day three. Watson earned a total of $77,147, compared to Ken Jennings with $24,000 and Brad Rutter with $21,600.  

Alumnus David Ferrucci ’83 and his team put Watson to the test on Jeopardy by programming the supercomputer made up of 2,880 IBM Power750 cores and 15 terabytes of memory.

Based on decades of work, Ferrucci and his team helped Watson learn how to analyze and interpret natural language content rather than just search facts to answer the game show questions correctly.

This wasn’t the first time IBM created a computer to take down a master. In 1997, Deep Blue defeated chess grandmaster Garry Kasparov, although Deep Blue was able to use a much more typical set of computer skills to play chess. The feat of natural language has set Watson apart as a standard in computer and artificial intelligence.

“Natural language is a classically difficult problem for computers,” says Ferrucci, IBM Fellow and vice president for the Semantic Analysis Integration Department at the Thomas J. Watson Research Center at IBM where he leads the Watson/DeepQA research project.

He described how questions on Jeopardy are particularly difficult for a computer because of the nuances of the English language and the amount of interpretation needed.

“We didn’t know what type of questions we were going to get,” Ferrucci says.

Computers that Care

Although decades of research influenced their work, the Watson project began at the end of 2006. By winter of 2011, Watson had competed and won.

A computer will never understand language the way humans do, but it will get better at predicting language.

“The method which we used was communicated across the world,” Ferrucci adds.

He says he sees many uses for the methodology, including in the health care industry. For example, written symptoms and patient histories need to be easily shared and interpreted among medical professionals to find possible diagnoses and treatments.

“Watson enables computers to communicate more the way humans communicate with each other,” Ferrucci says. “Any application where the challenge is more effectively understanding and analyzing is one where Watson can try to help.”

From Bio Labs to the Big Leagues

As a biology major and a computer science minor at Manhattan, Ferrucci took a strong interest in combining the two fields.

“I developed programs for the Biology department and built tools used in the bio lab and physiology lab at Manhattan and the College of Mount Saint Vincent,” he says.

Even though computers were in their infancy during Ferrucci’s college years, his programs are still being used years later.

After graduation, he worked at IBM for a couple of years before pursuing his doctorate at Rensselaer Polytechnic Institute. He returned to IBM in 1995 to work in the research department of the Watson Research Center, named in honor of Thomas Watson, the founder of IBM.

After their Jeopardy victory, the Watson team stayed together to work on making the system even smarter.

“Language is so much about context. I sometimes have to tell myself to step back and be more flexible in interpreting words,” Ferrucci says, recognizing personal similarities to his computer creation. “A computer will never understand language the way humans do, but it will get better at predicting language.”