Watson and Deductive Reasoning

IBM's Watson computer could be a major improvement in healthcare. But it won't be replacing your doctor anytime soon.

I watched Watson appear on Jeopardy and I was pretty impressed. It appeared to me to be a deductive reasoning engine, taking bits of information and using them one by one to come to an answer. If there was enough information for the machine's algorithm to show a certain confidence level, Watson answered the question. The syntax didn't need to be computer ready either - it was just plain English.

One thing I thought is that Watson could be a major help with medical diagnoses, which use a similar kind of reasoning. A patient would appear with a set of symptoms, say fever, night sweats, and muscle aches. The doctor would use his or her knowledge and come up with an answer cold or flu (or something else - I'm not a doctor). With more information, the doctor may come up with a different answer. That seems to be exactly how Watson works.

I can absolutely see a doctor feeding a set of symptoms into a Watson system to help find likely diagnoses. In fact, it might help improve the diagnosis of rarer conditions with which the average primary care doctor might have less experience.

But it seems that what Watson can't do is figure out the correct follow-up questions, or know how to interpret the patient's answers. Not everyone can put every symptom they may be feeling into words, particularly if they lack vocabulary due to education level or a primary language that doesn't match the doctor's. Some people may be hesitant to give all necessary information for many reasons. A doctor will use all of the patient's answers and non-answers to determine the best course of action, and not just the patient's words.

For example, those symptoms I listed are also those that can occur in the weeks following HIV infection. There are many reasons why a patient might be hesitant to mention possible exposures to HIV, particularly when the two most common modes of transmission are through unprotected sexual contact and the use of IV drugs. A doctor would be able to determine if an HIV test was warranted by paying attention to the patient, even if the patient said that they did not have any possible exposure. I don't know if Watson would.