Watson is out of Jeopardy!

Imagine a doctor’s consulting room in the near future. A hard-pressed doctor is unsure of a difficult diagnosis. She is pretty sure what is wrong with her patient, but would like some support.

She picks up her iPhone6 from a stack of papers and asks calls a trusted colleague for a second opinion, describing the patient’s symptoms.

Her remote colleague listens carefully, asks some useful questions, and very quickly offers a view on the most likely diagnosis. A pretty routine second opinion; except that the remote clinical expert on the phone is the world’s smartest computer.

Better than the best at Jeopardy

Named after Thomas J Watson, the founder of IBM, rather than after Sherlock Holmes’ faithful medical chum, Watson shot to fame in February when - after intensive tuition and training - it took on and beat the all-time champions of the iconic US game show Jeopardy!

What really sets the computer apart is its ability to understand questions asked in natural language and respond quickly. Really quickly. Watson represents the state-of-the-art in expert knowledge systems.

To win Jeopardy! required Watson – and, of course, the IBM team of engineers and programmers - to achieve several things that most computers are notoriously bad at.

First to ‘listen’ to and to ‘understand’ natural language questions; and then to correctly answer them. Although this sounds simple, Jeopardy asks questions in a convoluted fashion. It requires contestants to both have vast knowledge, but also to make intuitive leaps of reasoning and deduce obscure connections.

Watson got the correct answers on Jeopardy! by building hypotheses and simultaneously evaluating and weighing up massive numbers of possible answers against a huge repository of knowledge, all held in its memory rather than search.

The number of hypotheses created and evaluated to answer any one question is astonishing. Watson processes the equivalent of 100m pages of data a second to work out the right answer to a question.

It is what IBM calls a "natural language processor" because it can take questions spoken in plain English, break them into three or four parts, and quickly correlate them with information in its databanks.

Rob Smith from the chief technology officer’s office at IBM UK told eHealth Insider: “It’s a massively parallel design; an idea that has been around since the 1970s and 80s but is now being delivered.

“In essence, when asked a question, Watson evaluates lots and lots of options, assigns probabilities and eliminates them until it reaches the right answer.”

To help refine and test its hypotheses Watson can also ask further questions to narrow down its deliberations. It can also give answers that are tagged with ratings representing its "confidence level" in a particular answer.

Watson gets ready to leave the labs

This ability to construct hypotheses and make sense from vast amounts of unstructured data has huge potential in many sectors. IBM is particularly interested in using it to improve the service delivered by online helpdesks and telephone advice services.

But it is also looking at medicine, where Watson, accessed over a smart phone, could provide an immediately available second opinion or expert guide to diagnosis.

Within healthcare, it is estimated that the volume of research data is doubling every five years; far outstripping the ability of a specialist to keep up with developments, even in their narrow field. Managing and making sense of this avalanche of data creates the opportunity for expert knowledge systems such as Watson.

“It only deals in evidence,” Smith explained. “Watson doesn’t deal in hunches; it follows pathways. We avoid the use of pre-built decision trees like Map of Medicine.”

He suggests that Watson could offer the way to make decision support tools far more acceptable to doctors and other clinicians.

To help show the potential of Watson in healthcare, IBM put the computer through medical school, spending four months feeding it the content of the New England Journal of Medicine.

The company has since announced that it is collaborating with Columbia University Medical Centre, the University of Maryland School of Medicine, to understand where Watson’s technology could best contribute to medicine, and how it could best provide medics with assistance.

It has begun working with US health insurer WellPoint to develop evidence-based care protocols. And it has announced an alliance with Nuance, to gain access to its speech recognition and clinical language understanding technologies.

The two companies say they expect the first commercial offerings from their relationship to be available in 18-24 months.

Coming to a cloud near you?

With the hardware including ten super-computers, we aren’t going to be seeing Watson sat on anyone’s desk any time soon.

But cloud-based access to a technology that can comb through medical histories, medical journals and clinical trials to come up with appropriate diagnoses and treatments? That could well available within a few years.