In February, an artificial-intelligence supercomputer from IBM called Watson, named after the company's first president, easily beat two seasoned champions of the U.S. TV game show "Jeopardy." The game is a test of general knowledge, though that is not to trivialize such trivia games as "Jeopardy." We now know they need a much higher level of intelligence than, say, the game of chess. It was another IBM computer, Deep Blue, that demonstrated that fact in 1997 by beating Grandmaster Garry Kasparov. Unlike Kasparov, Deep Blue still couldn't tie its shoelaces the next morning.
Language, Concepts and the Relationships between the Concepts
The intelligence involved in "Jeopardy" requires such core attributes as:
- an extensive and unstructured textual knowledge base represented in a natural human language (English, in this case, but it could be any human language);
- the ability to parse that language sufficiently to identify all the concepts within it and all the relationships among the concepts and between the concepts and all the information in the knowledge base;
- the ability to search the knowledge base to find the answer to the parsed question.
There are some additional required attributes, such as the ability to structure a simple response, but they are less fundamental.
The "Jeopardy" event was just a demonstration of Watson's capabilities and a clever marketing promotion for what is coming next, and it is what's coming next that ought to be of great interest to those of us in health care.
Watson Goes to Med School
IBM is very frank about the next step for Watson: It is to become a physician's assistant, specifically for diagnosis. Considering that the Institute of Medicine and HealthGrades have reported between 400,000 and 1.2 million deaths from misdiagnosis and other human error from 1996 to 2006 in the United States, doctors (not to mention their patients) may need all the help they can get.
Diagnosis is essentially about determining relationships between a patient's clinical, demographic and psychographic signs and symptoms on one hand, and the complete, current base of medical knowledge on the other. Not only can Watson discover extremely complex and subtle relationships in enormous volumes of information, it can do so almost instantly. In other words, it can diagnose better and faster than humans can.
John Kelly III, IBM research director, says in a video on the IBM Watson website, "In seconds, doctors everywhere in the world are going to be able to find out what are the best treatments and how to ensure the best outcomes." That's great, but why stop at doctors?
One easily can envision Watson as an app available to anyone with a smart phone, which soon will mean almost everyone on the planet. Doctors and nurses might have an edge today in knowing how to pose the question—how to include all the patient's known symptoms, vital signs, demographics, history and so on—but it is not difficult to imagine the Dr. Watson app being able to ask and answer these questions all by itself and, indeed, to have better information through other health apps and sensors in the patient's smart phone and clothing.
The speed of Watson's responses on "Jeopardy" was as significant as their accuracy. In diagnosis, hitting the button fast does not just mean winning a game. It can mean saving a life.
Watson does not mark the end of human intellectual superiority any more than Deep Blue did, but it—or its close successor—might well mark the end of human superiority at diagnosing disease. Of course, Deep Blue did not stop people from playing chess; but then, chess is just a game.
David Ellis is a futurist, author and consultant, and publisher of Health Futures Digest, a monthly online review and commentary on technological innovations and their consequences and implications for health policy and practice. He is also a regular contributor to H&HN Daily and a member of Speakers Express.