BOSTON — IBM’s Jeopardy-playing artificial brainiac came to Beantown. It saw the pinnacles of East Coast higher education. And it conquered.
On Monday, at the Harvard Business School in Boston, Big Blue’s Watson supercomputer took on a team from the prestigious B School and a second from the MIT Sloan School of Management. Harvard’s Jeopardy team held its own and briefly had the lead. The MIT team, not so much. And in the end, Watson took home the trophy, just as it did when playing two human Jeopardy champs on national television earlier this year.
Following that famous TV victory, IBM is taking Watson on the road to multiple universities to expose students to its data analytic prowess. It’s just as important to get business school students thinking about the future of technology as it is to get scientists thinking about it, said David Ferrucci, IBM Fellow and Principal Investigator for Watson Technologies.
Many of the grad students in the audience will start up businesses or hold management positions in the coming years when IBM is likely to roll out Watson-related data analytics products — products that go beyond simple information retrieval, seeking to analyze many possible answers and return the most relevant information.
Harvard Men and Women Versus Machine
In Boston, the Harvard team boasted two Jeopardy veterans. Team member Genevieve Sheehan was on Jeopardy in 2009, and Jayanth Iyengar was a finalist in the Jeopardy College Championship in 2005. The experience showed. The MIT team frequently forgot to phrase answers in the form of a question.
At the end of the first round, Watson had $8,600, Harvard had $5,200, and MIT had minus $200. MIT was fast out of the gate in the second round and got to $3,400, but Watson and Harvard quickly widened the gap. The most dramatic moment of the match came when Harvard hit a Daily Double while trailing Watson $11,200 to $19,400. Harvard bet everything and won, pulling into the lead with $22,400. The crowd of mostly Harvard students erupted.
Watson regained the lead in the second round, but not by much. The round ended at Watson $28,600, Harvard $26,800, and MIT $4,600. The final question answer: “Finding the spot for this memorial caused its creator to say, ‘America will march along that skyline’.” The MIT team’s response was incorrect but appropriately geeky: “What is the Wright Brothers Memorial?” They finished with $100. Both Watson and the Harvard team answered correctly: “What is Mt. Rushmore?”
Harvard bet conservatively but it wouldn’t have mattered. Watson bet enough to win even if Harvard had gone all in. Harvard finished with $42,399 and Watson had $53,601.
Watson was clearly superior on geography and trivia — questions with clear if obscure answers — but much less effective on questions that involved wordplay. Watson was also more effective than the humans at timing the buzzer.
But it should be noted that Watson wasn’t all there. The roomful of 10 refrigerator-size racks of computer processors and memory chips that runs Watson stayed at home. IBM had Watson calculate answers to all of the questions beforehand and brought the answers and the game-playing portion of the system on the road.
Watson made real-time decisions about which questions to answer, which categories to choose and how much money to bet, said Ferrucci. The effect was the same as if the game took place at the IBM Research Center in Yorktown Heights, New York and Watson processed answers in real time, he said.
More Than a Game
Games are great for generating publicity, but IBM is out to make money, which means putting Watson to work. Big Blue is working with health insurance company Wellpoint, Inc. to develop a healthcare application of the Watson technology. IBM is also looking at the financial services industry.
IBM could eventually bring the technology to a range of products, said Ferrucci. “I could see different kinds of deployments, different sizes.”
There might also be a web-scale Watson deployment someday, he added. Just don’t expect it to resemble Apple’s Siri. “What you see in Jeopardy is question in, answer out,” he said. “But what Watson is doing behind the scenes is looking at many possible answers, gathering evidence, scoring the evidence. [Think] questions that really matter and you’ve got a human who’s got to weigh different possibilities and see the evidence for those possibilities and make decisions — that’s where Watson brings unique value.”
The worry is that systems like Watson will end up replacing jobs currently held by good old fashioned humans. Prior to the Jeopardy match, there was a mini-conference where various economists and business professors discussed the hollowing out of the middle class, the role of technology in income inequality, and the coming transformation of the global economy.
One of the panelists was Erik Brynjolfsson, who co-authored the recently published Race Against the Machine, which examines why — with technology boosting productivity — the bulk of US workers have not seen real gains in decades. “We have accelerating innovation, yet stagnating income,” said Brynjolfsson.
Humans Will Adjust. Eventually
The economists found that corporate after-tax profits rose from 1990 to 2010, and so did the portion of gross domestic product that went to corporate profits. Over the same period, labor’s share of GDP went gone down. The problem is, technology is beginning to take jobs away from workers in the middle of the labor market whose jobs involve clearly defined procedures. The result: a growing divide between educated, highly skilled workers and middle-tier workers, said Brynjolfsson.
There have been four waves of technological innovation that disrupted the labor market over the last two and a half centuries starting with the Industrial Revolution, and we’re beginning the fifth, said IBM Chief Economist Martin Fleming. “We’re now beginning to enter into, in my view, a period where the economy is beginning to open up opportunities for the deployment of very significant innovation … We’re going to see many new industries get created, radical new technologies being deployed, but being deployed in the context of new business models,” he said.
“This will have significant implications from an income and income distribution point of view.”
The MIT economists generally agree that we’re at the beginning of a technology-driven shift in the economy and ultimately the labor market will adjust. But no one had any good news for workers in the middle of economy during the transition. “The future is already here in many ways, in terms of what technology can do,” Brynjolfsson said. “But right now the benefits are not very evenly distributed.”
The Jeopardy team representing the MIT Sloan School of Management may realize this better than most.