IBM Watson may indeed be "Dr. Watson" one day

Two months ago, IBM announced the creation of a new business unit called IBM Watson Health. The idea was to take the artificial intelligence (AI) technology that powered IBM Watson to victory in the game show Jeopardy and apply it to problems in healthcare. In their blurb, IBM mentioned the goal of empowering "individuals to understand more about their health, while doctors, researchers, and insurers can make better, faster, and more cost-effective decisions." One area of interest is automating the process of disease diagnosis followed by treatment recommendation.

An informative article in IEEE Spectrum highlights the challenges of automated diagnosis, and starts by providing background on the past history of AI in this problem domain:

"There’s a rich history of medical AIs, from Internist-1—a 1970s-era program that encoded the expertise of internal-medicine guru Jack Myers and gave rise to the popular Quick Medical Reference program—to contemporary software like Isabel and DXplain, which can outperform human doctors in making diagnoses. Even taken-for-granted ubiquities like PubMed literature searches and automated patient-alert systems demonstrate forms of intelligence."

In theory, Watson could combine old and new technologies within its "cognitive computing" framework to create a powerful new medical informatics platform. However, one should not underestimate the challenge. How does a computer system make recommendations for 'a 68-year-old man with diabetes and heart palpitations: "Watson, given my medical record, which is hundreds of pages long, what is wrong with me?"'

Is it a chronic problem or an acute problem? Does it indicate a major health problem or a minor fright. What tests need to be ordered? Should medications be administered immediately. Will there be any adverse reactions to existing medications, and so on.

Watson's success on the show Jeopardy was the result of its ability to store and access a staggering amount of information. However, this body of information was factual in nature, unambiguous, and well-defined. Furthermore, Watson had only to provide a direct answer (in the form of a question) to a direct clue, and so a sophisticated decision-making process was not involved.

Medical diagnosis and recommendations are much more complicated, and these complications can be categorized into two general areas: Data processing (acquisition and organization), and decision-making. Data are a major issue because 1) The vast quantity of data needs to be prioritized and organized. 2) Data can be of mixed quality and accuracy and so filters and evidence-weighting need to be established. 3) It is necessary to integrate disparate data sources from journal articles to electronic health records (EHRs) to patient testimony.

The second even greater challenge is converting these data into knowledge and understanding which forms the basis of concrete recommendations. One can envision a combination of hand-crafted rules coupled to statistical machine learning methods that are trained on data. The key is identifying the "optimal" decision in the presence of significant uncertainty caused by partial information and novel individual circumstances:

"[T]he much more complicated challenge of also codifying how a good doctor thinks. To some extent those thought processes—weighing evidence, sifting through thousands of potentially important pieces of data and alighting on a few, handling uncertainty, employing the elusive but essential quality of insight—are amenable to machine learning, but much handcrafting is also involved."

To sum up, the ultimate challenge is reproducing a doctor's good judgment. Fortunately, IBM Watson is not alone in this endeavor. There are competing and complementary AI technologies that are addressing these and other challenges:

"Take the AI that Massachusetts General Hospital developed called QPID (Queriable Patient Inference Dossier), which analyzes medical records and was used in more than 3.5 million patient encounters last year. Diagnostic programs like DXplain and Isabel are already endorsed by the American Medical Association, and startup company Enlitic is working on its own diagnostics. The American Society of Clinical Oncology built its big-data-informed CancerLinQ program as a demonstration of what the Institute of Medicine, part of the U.S. National Academies, called a “learning health system.” Former Watson developer Marty Kohn is now at Sentrian, designing programs to analyze data generated from home-based health-monitoring apps."

This competition is good and bodes well for the future. However, one must take the long view because the problem is so hard and we are just starting:

"How will all this shake out? When will AI transform medicine, or at least help improve it in significant ways? It’s too soon to say. Medical AI is about where personal computers were in the 1970s, when IBM was just beginning to work on desktop computers, Bill Gates was writing Altair BASIC, and a couple of guys named Steve were messing around in a California garage. The application of artificial intelligence to health care will, similarly, take years to mature. But it could blossom into something big."

Yes the future payoff of a competent "Dr. Watson" can be huge in terms of better and more cost-effective medical care.

Figure 1. One use-case for IBM Watson is in the area of cancer diagnosis and treatment.