|
As Westchester celebrities go, it’s big. Its three television appearances drew record ratings. It has more than 40,000 likes on Facebook and roughly 28,000 followers on Twitter. It has its own Wikipedia page and its own YouTube channel. But what has Watson, the IBM supercomputer that famously trounced its Jeopardy! opponents in 2011, been up to lately?
Watson actually was designed for work, not play. And in the two-plus years since the computer’s Jeopardy! triumph, IBM has been forming collaborations in the healthcare, retail, and financial-services industries to develop commercial applications for the computer’s technology. IBM hopes that Watson, the result of the work of 25 research scientists over the course of four years, will eventually earn far more than prize money.
Jeopardy! was the perfect showcase for Watson’s potential in the world of artificial intelligence. It gave IBM engineers a high-profile (and entertaining) platform to demonstrate how it could understand the nuances of natural language, and sort through masses of information to come up with logical answers to questions.
To understand Watson’s breakthrough technology, you need to think about the subtleties of how people actually speak. Traditional computing systems can’t understand all of the idiosyncrasies and expressions people use in everyday language. Think about it: Why are “a wise man” and “a wise guy” opposites, while a “slim chance” and a “fat chance” are the same thing? Why do noses run and feet smell? Why do houses burn down but their contents burn up? These idioms may be something that a human child can pick up naturally, but programming a computer to recognize and understand this no easy task.
Watson uses a cognitive system that is similar to how humans reason and process information. It uses context when it can, and then taps into its own ever-increasing knowledge base to process deep natural language. (At one point, Watson was even fed the Urban Dictionary, a compendium of street slang, but it was quickly removed.)
The key is that after Watson sifts through huge amounts of data, it can then engage in complex decision-making. It uses what researchers call “Deep QA,” which is very different from search-engine technology. With a search engine, you type in key words or phrases and get links to documents or websites where those words appear. It is then up to you to sort through the results—likely tens of thousands—and find what’s relevant. Watson does that for you. It doesn’t return documents, it returns answers to your questions in a ranked list, with the computer’s confidence level attached to each answer. “What Watson would do is analyze your question and try to figure out what you are actually asking for. It would understand the context of what you’re asking,” explains Jennifer Chu-Carroll, one of the core IBM researchers who developed Watson.
This is far more than a parlor trick. IBM estimates that 90 percent of all the data in the world was produced in the last two years alone, and that 80 percent of all the information in the world is unstructured. By unstructured, they mean that the untagged material appears in all sorts of forms—as literature, as articles in academic journals, as PowerPoint presentations, as physician notes dictated into charts, as blogs, as Tweets, as emails, as banter in chat rooms, and more.
“Watson is the first commercially available cognitive computing offering—a system capable of continually learning and refining itself through human/computer interaction,” says Steve Gold, vice president of Watson Solutions at IBM. “Part of why cognitive systems are so necessary is because of the rise of Big Data. As organizations work to make fact-based decisions based on data, they need advanced systems that are able to crunch massive amounts of structured and unstructured data and deliver insights within seconds.”
To put Watson’s capabilities in perspective, imagine a person who is able to absorb every possible piece of data available on a subject, along with anything tangentially related, then sort through it, make logical connections among mounds of disparate sources—and then make a recommendation based on all of it—in a matter of seconds.
Now, on top of all that, add a third skill—the ability to learn. Watson is actually getting smarter. “Watson is learning at an accelerated pace on what it’s been taught to date,” says Gold. “We’re very excited about how far and how fast Watson has come.”
The potential commercial applications for Watson are huge. One of IBM’s first Watson collaborations has been with Memorial Sloan-Kettering Cancer Center (MSKCC) in New York City. The need was clear—medical information doubles in volume every five years. The National Center for Biotechnical Information estimates that it would take an oncologist roughly 160 hours a week to remain current with medical literature, research, guidelines, and best practices.
Since 2012, MSKCC and IBM have been collaborating on a new Watson-based decision support system for cancer research and treatment. They began with lung cancer and are progressing to breast and prostate cancers. Watson has been fed reams of data that unites clinical expertise, molecular and genomic data, and a repository of 25,000 case histories. The idea is that an oncologist can treat an individual patient by having Watson crunch all the data and then identify the key disease patterns that most closely approximate that person’s case.
A demo co-created by IBM and Sloan-Kettering showcases how it would work: Mrs. Yamodo (a fictional patient), 37, comes to her doctor because she has a persistent dry cough and labored breathing. Her doctor orders a chest X-ray, which reveals a suspicious mass on her lungs. A subsequent CT scan and biopsy confirms the likelihood of cancer, and Mrs. Yamodo is referred to an oncologist.
As Mrs. Yamodo sits in the waiting room, her new doctor heads to his office to familiarize himself with her case. The oncologist has only a few minutes between appointments. He logs into the electronic medical-records system and clicks on the “Ask Watson” button at the bottom left of his screen.
Watson, who already has been fed the results of Mrs. Yamodo’s medical tests, evaluates all of the information in the electronic medical records and analyzes it against tens of thousands of documents in its vast knowledge base—like medical journals, textbooks, clinical trials, industry association guidelines, specific hospital best practices—and then identifies the pertinent facts for the case.
Watson then recommends several treatment options, based not only on all the data it has just digested, but also on an individual’s genes, medical history, and demographic profile. If the doctor wonders why Watson is recommending a further test or a particular combination of drugs, he can press on the “evidence” button. The computer will drill down to the specific text, even the paragraph, to show what’s behind its decision-making.
Watson is not yet working in real time with live patients, but it is running scenarios with real (anonymous) patient data. It will be piloted at two hospitals in the Northeast region later this year. Meanwhile, researchers at IBM are working with the Cleveland Clinic on what it calls WatsonPaths, another evidence-based medical solution in which Watson is learning to help train medical students.
Fans who followed Watson’s performance on Jeopardy! know that the more it played the game, the better it got. The same holds for Watson’s medical skills. The more the computer works with actual cases, and the more information it’s fed, the better it gets at diagnosis and recommended treatments. And, as on the show, Watson can convey its confidence level and why it recommends one treatment over another.
If it all sounds a little frightening—“Dr. Watson will see you now”—the partners involved are quick to say that Watson will never replace physicians. It is referred to as “a very smart assistant.” That “smart assistant” has also garnered other deals. Among them: The Cleveland Clinic is also using Watson to help train medical students using questions from the US Medical Licensing Exam. WellPoint Inc., an insurance company, is training Watson to help analyze claims, and researchers in pharmaceuticals are exploring Watson’s potential for developing new drugs.
Watson could also be of use in financial services where data overload is daunting. IBM is collaborating with institutions to explore possible applications in areas like customer interactions, and in streamlining the banking experience for consumers.
And then there’s customer service. Last May, IBM rolled out the Watson Engagement Advisor, designed to be a significant improvement over traditional customer-contact centers. Instead of sending frustrated consumers through an endless trail of irrelevant phone-tree options, IBM believes Watson has what it takes to understand more precisely what the customer wants and to respond efficiently. Watson could do this through a variety of platforms, including television, tablets, and smartphones.
Should we all worry that Watson will be after our jobs? “It’s a persistent fear that we have seen with technology—that humans will no longer be needed. It’s happened with each innovation. I think you will see, as we have with steam engines, that Watson has a rightful place to make us better, as individuals, as employees, as humans,” says IBM’s Steve Gold.
But tell that to Ken Jennings, the reigning Jeopardy! champion until Watson came along. In a 2013 Tedx talk, in which he bills himself as “an obsolete know-it-all,” Jennings talks about Watson’s dark side. Jennings warns against outsourcing critical thinking. During the Jeopardy! game, Jennings said he felt like a factory worker in the 1980s, seeing that a robot could do his job. He describes IBM programmers and executives “holding up ‘Go Watson’ signs and applauding like pageant moms every time their little darling got one right.” He warns professions like paralegals and pharmacists they could be next. “Watson and his brethren may be something good or something ominous,” says Jennings. “All I know is how it felt to be the guy put out of work. Here was the one thing I could do well. It felt demoralizing.”
So what is it like to actually meet Watson—that famous, omnivorously knowing computer? Disconcerting. Watson—which in its Jeopardy! days was powered by 10 racks of IBM POWER 750 servers, using 15 terabytes of RAM and 2,880 processor cores (the equivalent of 6,000 high-end desktops performing in concert)—is a physical shadow of its former self. The whole system has been reduced by 75 percent, though amazingly it has gotten 240 percent faster. Watson still has a large physical presence in Yorktown, where it was developed. But the computer is also in the cloud, and can actually run on one server—about the size of four pizza boxes—that can travel to remote locations.
Watson does not do interviews. It really can’t. “Watson is not a conversational system and doesn’t have a sense of self,” explains Chu-Carroll. But do the researchers at IBM sometimes think of Watson as a “him” instead of an “it”? “Oh sure,” says Chu-Carroll, who admits that Watson is “like my third child—a third child I share with 40 other people.”
Kate Stone Lombardi is a journalist and the author of THE MAMA’S BOY MYTH: Why Keeping Our Sons Keeps Them Closer. She rooted for Watson during Jeopardy!
» For More from the November issue, click here.