In 2011, the world witnessed perhaps the most important game of Jeopardy! that has ever been played. The contestants were two Jeopardy!-made millionaires, Ken Jennings and Brad Rutter, and an untested newcomer—a supercomputer named Watson. Through the course of two matches, Watson completely dismantled his creative and brilliant human counterparts. His two-day total came out to $77,147, while his competitors garnered a paltry $45,600 combined.
It is difficult to express how remarkable Watson’s victory was. Computers have bested humans in tasks ranging from chess to poker to simple arithmetic. But one thing has always existed just outside of their grasp: puns. And yet, Watson was able to navigate and excel at a game that relies heavily on wordplay, a game that requires the contestant to understand an esoteric statement in the context of a given category and provide a question that can be answered with the statement.
This entire column could be devoted to Watson’s achievement, but it seems that task is best left to the tech column. What I’m more interested in is what Watson is doing now, two years removed from his triumph—he’s an oncologist. That is, he’s acquired the knowledge base of a cancer doctor.
Because Watson possesses the ability to quickly process “unstructured” data, an ability known to humans as “reading,” he has been able to digest and mine through an astoundingly large volume of medical information: some 25,000 case studies, 605,000 statistics and facts, and over 2 million pages of text. He’s also been shrunk from a room-sized data center to a single web-connected server and become 240 percent faster.
These vast improvements and data acquisition have Watson humming at a new home, the Memorial Sloan-Kettering Cancer Center, where he assists real oncologists on the job. If you don’t think physicians need any help with their jobs, consider one study published by Saul J. Weiner et. al in Annals of Internal Medicine that found that doctors made errors in judgment in 30 percent of uncomplicated cases and 60 to 80 percent in “biomedically [or] contextually complicated cases.”
How is this possible? The main reason is that physicians make decisions based partly on what they’ve learned in school and partly on what they’ve seen in their years of practice. But in a rapidly changing medical landscape, this technique can quickly fall behind, especially when some doctors practice for 30 to 50 years. It is estimated that in order to stay afloat in the sea of published medical literature, an average physician would have to dedicate 160 hours per week to reading scientific papers, according to the article “Doctor Watson” published by The Economist. This doesn’t leave much time for actually using their knowledge, considering a week only contains 168 hours. Watson, on the other hand, can process hundreds of pages a minute, doesn’t need to sleep or see patients, and makes a diagnostic error only once in every ten cases.
At this point, Watson is a just a tool for human doctors to use to make informed decisions, like a really expensive stethoscope or a tongue depressor. He provides differential diagnoses, treatment courses, and percentages that relay his confidence in each possibility. It’s up to the doctor and the patient working together to make the final decisions.
But Watson’s success so far begs the question: How much longer will we trust human doctors? Even the brightest and most careful humans make errors. A lot of errors. We often use our humanity as an excuse when we mess up: I’m only human. There’s no way for any person to hold all published medical knowledge in their head or to remember the outcomes and complexities of every case they’ve encountered. But a computer can do it all. A computer could, in the foreseeable future, keep up with the literature, take into account every intricacy before it, recall every patient it has ever seen, and provide confidence based on statistical percentages instead of gut feelings.
Yet it feels as though there’s something intuitively scary about having a computer as a doctor. Wouldn’t it lack empathy? Wouldn’t it miss the small but necessary contextual details? Undoubtedly, there are physician jobs that could not be filled by a computer. Nobody wants a machine telling them they have terminal cancer. Nobody wants a physical examination completed by the cold hands of a robot. It seems that at least some of the jobs of a physician, at least for a while, must be done by humans.
That being said, the case for replacing the decision-makers in medicine with ultra-powerful computers is compelling. In the five minutes it has taken you to read this article, 15 new scientific articles were published. A human cannot digest or retain even close to that amount of new information on a daily basis. A human makes mistakes in judgment, and while a computer can technically make mistakes, they should only occur when the medical evidence pointed them in the wrong direction. In an evidence-based profession, the evidence is pointing to a new conception of how we view healthcare. Hopefully, this will be one decision we mere humans can get right.