Watson Learns to DebateBy
IBM Watson is a computer system that can answer questions posed to it in natural language. It might be most famous for its defeat of the two Jeopardy! champions in 2011, but the number of Watson’s commercial applications from 2013 has grown into a division (IBM Watson products and solutions) at the company.
For many years now, IBM and others have sponsored competitions between human experts and computers. While some might dismiss the value of these events, they do serve as milestones for the level of sophistication and power of AI in our machines. In 1997, IBM’s Deep Blue defeated Garry Kasparov, the highest-rated chess champion, in a tournament that Kasparov thought he would win. Chess, after all, had reassured humanity for many years as the ultimate Turing test for human vs. machine intelligence.
Before the 2011 Jeopardy! contest, Ken Jennings was also very confident. He later said, “I had taken some Artificial Intelligence classes, and I knew there were no computers that could do what you need to do to win on Jeopardy.” Yet it was the machine that walked away (rolled actually) with the million-dollar prize. In 2017, Ke Jie, the world champion of the most complicated of games, Go, was defeated by Google’s AlphaGo. After the match, Jie said he was considering retiring from his lifelong involvement with the game because there was now a champion that likely could only be defeated by another machine.
Another, somewhat less-heralded, head-to-head with machines took place in February 2019 at IBM’s Think conference in San Francisco, Calif. Like the other competitions, it was televised. The contestants were IBM Debater (the machine) and Harish Natarajan. According to IBM’s project manager, Natarajan was “the most decorated debater in the history of university debate competitions with the world record in the number of victories.” The test was which contestant would prevail in a formal debate of the question: Should governments subsidize preschools?
The debate offered a secondary, more important question: How far along was IBM in its development of AI applications? The previous chess and Go contests were on a level as abstract as mathematics, and Jeopardy! essentially tested recall and intelligent guessing. The debate was a conversation about a problem that went beyond information. The contestants had to persuade or, as Leonardo da Vinci explained 500 years ago, “Whoever in debate quotes authority uses not intellect, but memory.” Instead of taking pieces off a board, the computer was asked to move opinions within the minds of other human beings with logic, emotion, prejudice, and charm.
The moderator of the debate commented after the contest, “She was surprisingly charming and human-sounding.” In actuality, “she,” IBM Debater, was running on a powerful server with 28 processing cores and 768GB of memory, giving it 50 times the power of a high-end laptop. “She” was supported by four servers with 64GB of memory each and 2TB hard drives loaded with text.
It was close, but Natarajan prevailed, and the IBM team wasn’t surprised. This was the second public debate involving IBM Debater. (In June 2018, Project Debater participated in two contests, losing the first and winning the second.) Natarajan was in a class by himself. Nevertheless, for IBM the contest was a step forward.
When Project Debater began in 2011, researcher Noam Slonim had suggested a human vs. computer debate for IBM Research’s global AI team, and it became one of 30 different projects for the team. Aya Soffer, who runs the AI team, explained the value of the goal, “From our perspective, the debate format is the means and not the end. It’s a way to push the technology forward and part of our bigger strategy of mastering language. In general, computers are lagging significantly in understanding and being able to express themselves. If we expect AI to be useful, being able to communicate with people is critical.”
After the match, Natarajan commented on the quality of the exchanges and the persuasive reach of Debater. “What really struck me is the potential value of IBM Debater when combined with a human being.” He explained that he was impressed with the program’s ability to dig through mountains of information and offer useful context for that knowledge—context that da Vinci said would imply the involvement of intellect, not just memory, digital or otherwise.
In March 2020, IBM announced a number of Debater technologies now incorporated in Watson’s Business Solutions and the Watson Speech to Text service for browsers, including clustering for analyzing raw data, summarizing information, and sentiment analysis. The Watson products and solutions page on the IBM website has information on all the commercial offerings.