Speaking to the Health Journalism 2012 convention of the Association of Health Care Journalists, IBM Chief Medical Scientist Martin Kohn said Watson had to learn how to actually understand the English language, not just recognize words, in order to succeed on the show. It can access 200 million pages of text in about three seconds, then uses hundreds of algorithms to analyze the evidence and gauge the confidence it has in the answer.
On Jeopardy, for instance, Watson answered only when it had a response above a certain confidence threshold and the supercomputer’s biggest blunder – answering “Toronto” when asked to name a U.S. city with two airports – came during Final Jeopardy when it had to answer even though it had little confidence in the response, Kohn said.
It will be a little different with medicine and trying to reach a diagnosis.
“There may not be one answer,” Kohn said. And in fact, IBM is not pushing Watson as a diagnosis maker.
“Watson doesn’t make decisions for you,” Kohn said. The results Watson would give back would be more along the lines of “these are the things you ought to think about,” Kohn said. “It really is clinical decision support.”
But that’s not what self-described “patient hacktivist” Fred Trotter sees as the end result of artificial intelligence in medicine.
“Artificial intelligence in health care will be the most important force in diagnosis,” he said.
The biggest obstacle right now is the lack of electronic health records and a lack of access to those records by patients to correct what Trotter said is probably wrong or misleading information. Kohn concedes that machines like Watson would only be as good as the information it got from the records.
Trotter foresees a time when the physician will be a “hybrid” linked into a computer like Watson during the patient visit, interfacing with both to aid diagnosis. And Kohn could foresee patients accessing Watson directly to get personalized suggestions on their care.
“Going forward, you’re going to have to make more and more decisions with your (health care) team,” he said.
But Trotter can see years down the road an argument springing up: “Why are we using doctors?” he asked, with the supercomputers around.