About the SA Blog Network



Opinion, arguments & analyses from the editors of Scientific American
Observations HomeAboutContact

Paging Dr. Watson: IBM to apply Jeopardy! victor’s analytic skills to medical diagnoses

The views expressed are those of the author and are not necessarily those of Scientific American.

Email   PrintPrint

Sony, Jeopardy, IBM, computerThe answer is: For its next assignment, this Jeopardy! champion will have to work on its bedside manner.

If you replied, "What is Watson?" give yourself a round of applause.

With last night’s big game show victory under its belt, IBM has its sights set on applying the high-performance computer’s advanced analytics capabilities to the healthcare industry. IBM announced Thursday that it is teaming up with Burlington, Mass.-based Nuance Communications to integrate that company’s speech recognition and Clinical Language Understanding (CLU) technology with Watson’s "Deep Question Answering" natural language processing and machine learning capabilities. The result, which could be ready in as soon as 18 months, is expected to improve patient diagnosis and treatment.

IBM paints the following scenario: A doctor considering a patient’s diagnosis could literally ask Watson—running Nuance’s software—for help. Once the question is processed, Watson would quickly tap all the related texts, reference materials and prior cases as well as the latest findings in journals and medical literature to formulate an answer. A score indicating Watson’s level of confidence in its answer could even be provided. Columbia University Medical Center and the University of Maryland School of Medicine are helping IBM and Nuance by providing the medical expertise necessary to adapt the technology to a healthcare setting.

Nuance is perhaps best known for its work with Ford to provide voice-recognition software for drivers using the car company’s SYNC an in-car communications and entertainment system.

IBM is also in discussions with an unnamed consumer electronics retailer to develop a version of Watson that could interact with consumers on a variety of subjects such as buying decisions and technical support, The New York Times reports.

The Jeopardy! episodes this week served as a way for IBM to measure its progress in developing technology that would come up with answers to an incredibly wide array of questions covering many topics. IBM wanted to size up Watson’s performance along five metrics, according to David Ferrucci, the IBM researcher who led Watson’s development. These included the ability to respond to a broad and open domain (Jeopardy! questions come from any number of categories), questions formed in complex language with a variety of grammatical structures (often including wordplay that helps make the show more entertaining), high precision (there is a penalty for wrong answers), accurate confidence (betting depends on faith in one’s answer) and high speed (the first player to buzz in gets first crack at answering a question).

Watson got off to a slow start during the three-day trivia tournament staged at IBM’s Thomas J. Watson Research Center in Yorktown Heights, N.Y. After Monday night’s round, Watson was tied with former Jeopardy! champ Brad Rutter at $5,000, with another former Jeopardy! winner, Ken Jennings, sitting in second place at $2,000. Watson opened up a big lead during the second round and finished Wednesday night with the strangely precise amount of $77,147 to Jennings’ $24,000 and Rutter’s $21,600.

In a brief, informal, unscientific poll Scientific American took on Monday of 26 print and online staffers to determine who would win and by how much, half chose Watson. The predicted margin of victory ranged between $5,000 and $25,000—no one foresaw Watson running up the score the way it did.

IBM plans to donate its $1 million Jeopardy! prize to World Vision, a worldwide relief and development organization, and World Community Grid, whose mission is to create the world’s largest public computing grid. Half of Jennings’s $300,000 winnings and Rutter’s $200,000 purse will go to Seattle-based nonprofit VillageReach and Pennsylvania’s Lancaster County Community Foundation, respectively.

Gracious in defeat, Jennings commented (via an IBM press release), "I’ve studied artificial intelligence, and I can say that this is an impressive and unprecedented moment. A few years ago, I didn’t think it was possible for a computer to play at this level. IBM knocked my socks off with how fast they’ve caught up."

Image of [left to right] Jennings, Watson and Rutter, courtesy of Sony Pictures Entertainment

Tags: , , ,

Rights & Permissions

Comments 9 Comments

Add Comment
  1. 1. wolfkiss 4:11 pm 02/17/2011

    This is an impressive technical achievement. It’s interesting to note that expert systems, which Watson essentially is, have been promised for decades. It’s good to a really useful tool on the threshold of real world implementation. Probably, the most impressive aspect of this system isn’t its quick access to a library of facts, but its natural language processing. One-to-one mapping of trillions of facts is commonplace with search progress over the years, but nuanced language parsing is a much more involved problem that begs what is being meant in a question. Kudos for that.

    I still argue (interested in people’s thoughts), however, that statistical systems like Watson are far from being the general problem solvers humans are. There’s a reason we call it trivia in games like Jeopardy!. It’s because such constrained and isolated facts are useful only to a very isolated problem. This is the nature of a quiz show, mapping tight problem parameters to a coherent solution set, and this particular skill will most probably aid in similar scenarios like expert systems have promised. Never the less, most solutions to every day decisions do not converge so neatly on one solution and the problem parameters are diverse and often quite vague until a process of integration is initiated. For all its power consumption, it’s not designed to integrate the disparate facts it gathers based on disembodied questions into a higher-order conception. This is what every person does, mostly as a matter of course with magnitudes less power. I imagine that’s a completely different architecture.


    Link to this
  2. 2. Gregory CH 5:18 pm 02/17/2011

    IBM does it again! We need more IBM’s.

    Link to this
  3. 3. jtdwyer 7:34 pm 02/17/2011

    I agree. Here’s my earlier response in another article to a commentator referring this great achievement in relation to the HAL 9000 computer portrayed in "2001: A Space Odyssey":

    I haven’t checked, or even watched the show, but I expect that the Watson computer system includes highly optimized software and perhaps even hardware specifically designed to play the Jeopardy game, or some set of games, perhaps while seeming to be intelligent.

    If the Watson software were dynamically adaptable enough to perform some significant fraction of the myriad tasks continuously required of human intellect I’d be much more impressed. If I recall HAL was supposed to be capable of performing psychological evaluations of the spacecraft’s crew in addition to monitoring spacecraft operations and playing chess.

    I’ve been periodically reading about "intermediate steps" in computer and AI technical journals since the 1980s. I wonder when a system might be capable of learning a brand new game through English language instruction – you know, like an intelligent person.

    I’d personally be much more impressed with a system that tracked and fully evaluated the success rate of futurists’ predictions – that might be truly useful!

    Link to this
  4. 4. jtdwyer 7:42 pm 02/17/2011

    BTW, enough cute references to Dr. Watson, MD, sidekick of Sherlock Holmes. IBM’s Watson computer more directly refers to the Thomas J. Watson Research Center, headquarters of IBM’s Research Division where it was likely developed. Thomas J. Watson Sr. & Jr., respectively, were the president & CEO of IBM from 1915 (when it was the Computing Tabulating Recording Corporation) to 1971.

    Link to this
  5. 5. Dr. Strangelove 12:28 am 02/18/2011

    Playing games is child’s play. You want real AI? Pit Watson against a 6-yr old child in a Turing test. I bet $10k the child would win.

    A less intelligent test. Put all that computing power of Watson in a small robot via remote control and see if it can beat a cockroach in mobility and survival in the natural world. I predict the robot would be dead in one day.

    Link to this
  6. 6. elenasakman 8:35 am 02/21/2011

    I agree, that type of logic is still far from complexity of the human thinking. And still I am impressed with how much comfort such "search – simple logic and statistic" machine can bring to scientific research – imagine each one of us could have software like that? and just wait it will be definitely available soon – awesome achievement! In my work (programming) it would bring such conveniences! Because this software basically can do everything we do – understand humans and create a machine code. How exciting!

    Link to this
  7. 7. elenasakman 8:39 am 02/21/2011

    awww… I wish IBM would hire me – what an absolutely exciting project to create an application of this software for mainframe banking tasks

    Link to this
  8. 8. SFMalerich 8:41 am 02/22/2011

    Some interesting comments here. It’s so easy to criticize someone for failing to achieve what they didn’t even attempt. Watson isn’t about machine intelligence surpassing human intelligence. It’s about doing specific types of analyses faster and sometimes better than most of us can do. When I first heard about Watson and the Jeopardy challange, the first thing to come to my mind was what a wonderful tool this could be in medical diagnosis. It’s role is not to replace a human diagnostician, but to search the wealth of medical data that exists in digital form, correlate it with a patient’s symptoms and history, and inform the physician of various possibilities, along with likelihoods, and then present further testing and treatment options. Watson would not be in competition with the physician, but in consultation. Each would bring to the task their respective skills to solve a problem faster and more effectively than either could do alone.

    Link to this
  9. 9. jacobR 1:10 am 05/26/2011

    IBM’s supercomputer Watson has proven its problem-solving capability by stumping “Jeopardy!” winners. That problem-solving ability is now getting used to aid in medical diagnostics. Estimations of possible diagnosis with Watson will include some information doctors simply do not have time to consider. Personal blogs, off-label uses of drugs, and emerging research is all going to be used in calculating the result of diagnostic questions. Here is the proof: <a title="Every doctor gets additional aide with Watson supercomputer" href="">Watson Supercomputer turning to medical diagnostics,</a>.

    Link to this

Add a Comment
You must sign in or register as a member to submit a comment.

More from Scientific American

Email this Article