I’ll take ‘Business and Medicine,’ Alex
Rob High reports what Watson is doing after Jeopardy!
Watson, the IBM computer that trounced human contestants on TV’s Jeopardy! in 2011, has not been sitting idle. It’s already working in cancer treatment and research and helping businesses interact with potential customers.
Rob High, an IBM Fellow, vice president and chief technology officer for Watson, dispels some myths about the game winner. First, it is “not strictly speaking” a supercomputer. Although, the version that won Jeopardy! would probably rank in the top 50 most powerful computers at the time, he says. Second, there is no single Watson any more, but rather individual versions that can be tailored to specific projects.
“We can now run Watson on a single server with 16 to 32 computing cores and 256 gigabytes of memory,” he says. The quiz show champ, in comparison, had almost 3,000 cores and 15 terabytes of memory. (A terabyte equals 1012.) “We can get massive scale by continuing to increase the number of server replicas and can spread workload across that.”
Watson is a cognitive computer, which means it “simulates human cognition with a specific focus on understanding language with human-like clarity,” says High. “The second factor is that it’s a system that augments and amplifies human cognition.” That is, it enables humans to gain insight that they wouldn’t be able to gain through other types of computers.
“One of our early clients said it best,” High recalls. “He said, ‘What’s interesting about Watson isn’t just that it answers questions correctly. In doing that so rapidly and repeatedly, it allows me to realize that I’m asking the wrong questions.’” That’s a huge value for professionals whose job is to “see the other side of things.” “The risks and opportunities for most businesses lie in the things that you’re not thinking of,” says High.
Since Jeopardy!, the Watson team has been building out the system’s ability to understand language and extending its cognitive power. As a result, today’s Watson has abilities well beyond its Jeopardy! forbearer.
IBM has been working with the clinicians at Memorial Sloane Kettering Cancer Center in New York City to capture their expertise. Using a program called Interactive Care Insight for Oncology, Watson can offer physicians a list of treatment options that are specific to the patient, meet standard care quidelines, but are optimal for the patient’s needs. The program digests patient summaries, physician notes, nurses’ observations, lab reports and other relevant information.
“Watson can read all that and from it identify the attributes for making a selection of potential treatment options,” says High. “And we’ve augmented that with clinical expertise that we get from the hospital about which of the treatment options is most relevant to particular patients. In effect, we’ve individualized patient treatment selections.”
Watson even provides supporting evidence for these treatments, having absorbed reams of medical literature, research reports, journal articles and other relevant information.
“At the end of the day, the decision has to be made by the physician,” says High. “Watson is not going to replace doctors. It’s just helping them become more efficient.”
Watson is also playing a role in researching the deadly disease. Houston-based MD Anderson Cancer Center is using Watson to accelerate research and clinical trials in oncology treatment. “They’re interested in getting more of the clinical trials they produce in the hands of practitioners sooner, faster, more effectively,” says High. Again, Watson examines patient similarities, genetic conditions and other patient attributes to identify any that might be a candidate for clinical trials and that could be included in their treatment options.
The system also identifies possible adverse results and provides options for treating them. “If some new drug therapy is being administered and the patient develops side effects, the doctors are forewarned about that side effect and how to deal with it in addition to treating the cancer,” High says.
Watson is also busy in “engagement” – interacting with financial, insurance and telecommunications customers who want to know more about the goods and services that those businesses offer them.
“Today,” says High, “someone surfing the Internet might be exposed to an ad banner that they’d click on. The typical experience is they would then be linked to a portal page of some sort that talks about that offer and maybe has a brochure, FAQs or links to other pages.” Sometimes the experience can also include a forum where other customers share their opinions or a toll-free phone number with its attendant pitfalls of waiting for an operator who may or may not have the requisite knowledge. In any case, “the onus is on you to discover your own answers, and that experience can be very disengaging,” he says.
“We can replace all that with Watson using a question-and-answer system that answers the question you have, including questions that may not be directly related to the product. For example, I’ve been exposed to an offer for a mortgage. Before I decide, I might have a bunch of questions concerning the market conditions for buying a house, issues from a tax perspective, the trade-offs in value of real estate and so forth. Watson can answer those questions by ingesting all of the information the bank has on mortgages and real estate.”
The same approach can work for helping to choose a cell phone model or an insurance policy, High says. And he foresees Watson playing similar roles in other financial services.
“’What’s the best investment for me?’ is not a question Watson can answer,” he says. There are simply too many variables such as an investor’s appetite for risk, existing portfolio and market opportunities. But after digesting financial research from hundreds of sources, Watson could offer suggestions on specific investments, having analyzed factors such as a company’s potential, market forecasts and other data.
All this is not without challenges. “Cognitive systems are used to operate on human-related problems,” says High, “using human-related sources of information, and there is often ambiguity in our language.” Watson often needs a “shared historical context” to sort that out.
“For example, I could get a message from my wife that says ‘I’m going to stop by the store on my way home,’” he says. “That’s ambiguous. What store? When? For what? But if we had an earlier conversation about needing groceries and I know what time she gets off work and where we usually shop, it’s clear to me.
“Watson needs a similar shared historical context. Its big advantage is that it never forgets that context. If my wife told me about the groceries three days ago, I might not remember that. But Watson will.”