The mastermind behind Watson predicts computers will offer more personal information in the years to come
Tomorrow’s computers will constantly improve their understanding of the data they work with, which in turn will help them provide users with more appropriate information, predicted the software mastermind behind IBM’s Watson system.
Computers in the future “will learn through interacting with us. They will not necessarily require us to sit down and explicitly program them, but through continuous interaction with humans they will start to understand the kind of data and the kind of computation we need,” said IBM (NYSE: IBM)(Fellow David Ferrucci, who was IBM’s principal investigator for Watson technologies. Ferrucci spoke at the IBM Smarter Computing Executive Forum, held Wednesday in New York.
“This notion of learning through collaboration and interaction is where we think computing is going,” he said.
IBM’s Watson project was an exercise for the company in how to build machines that can better anticipate user needs.
IBM researchers spent four years developing Watson, a supercomputer designed specifically to compete in the TV quiz show “Jeopardy,” a contest that took place last year. On “Jeopardy,” contestants are asked a range of questions across a wide variety of topic areas.
“Jeopardy” proved to be a formidable challenge for IBM, even more of a challenge than its work on building a chess-playing computer that could beat chess master Garry Kasparov, which IBM’s Deep Blue did in 1997.
Chess is a finite mathematical problem — albeit a very large math problem — whereas succeeding on “Jeopardy” requires a deeper understanding of language, Ferrucci explained. With “Jeopardy,” “we don’t even know what the questions [are that] we will get,” Ferrucci said. The information in a database cannot be “carefully aligned” ahead of time to the questions to be asked. Of course, Watson was loaded with many sources of information, such as encyclopedias and dictionaries. But Watson also needed to map the questions, which were often worded in ambiguous ways, to the data it had.
Complicating the task even further is that words, unlike chess pieces, can change their meaning depending on how they are used. “The usage of words really defines their meaning. And the usage of words happens in human context. This is not a mathematic, well-defined search space. Computers have to do a lot more analysis to get a handle of what these words mean,” Ferrucci said.
The research team looked at 20,000 “Jeopardy” questions to determine their structures. They found that the vast majority of questions were too unpredictable to easily model. The only way to generate a plausible answer was to analyze the question in multiple ways, generate multiple answers, and then rank the probability that each answer is correct. And this is what Watson did.
Watson did win at its “Jeopardy” match. Now IBM thinks the Watson computing model can have a wide range of uses.
“There is a lot more work to do, but [Watson] is something we can adapt and customize to new domains. The Watson technology is not about question-in and answer-out, but rather it is understanding a problem,” Ferrucci said.
The key to this technology, Ferrucci said, is that it queries both itself and its users for feedback on the answers it generates. “As you use the system, it will follow up with you and ask you questions that will help improve its confidence of its answer. In its work with you it will capture new information it can use,” he said.
One field IBM is investigating is medicine. The company is working with medical researchers and doctors from Columbia University to adapt Watson so it can offer medical diagnosis and treatment.
“Medicine is the ultimate information-based profession,” said Herbert Chase, a Columbia professor of clinical medicine, who also spoke at the conference. Watson could serve as a diagnostic assistant and even offer treatment plans, he said.
A big problem that doctors face today is having too much information. The scientific body of knowledge about how the human body works has expanded beyond any one person’s ability to understand. For doctors, “there is all this information out there, [but] we can’t access it,” Chase said. “We can’t even find the information.” In many cases, doctors make decisions not knowing what other doctors have learned about a particular condition or set of conditions. Wrong and missed diagnostics account for 30 per cent of medical errors, he said.
Chase displayed a demo about how a doctor could use Watson. A person comes into the emergency room complaining of pain in the eyes, a problem that the attendee hasn’t seen before. Watson provides a list of differential diagnoses, or a range of possible answers, ranked by probability. “We want to be prompted to get ideas,” he said. The most obvious answer could be uveitis. A doctor can look at the list and think of a few easy tests to distinguish which problem it may be. The patient may also have a rash and a fever. A doctor can then ask Watson about these additional symptoms in light of the diagnosis. Watson can provide an answer that may not be intuitive, such as Lyme disease.
Watson can also provide a list of ways to treat a patient, drawing from guidelines from sources such as the Centers for Disease Control. If the patient has complicating factors, such as being pregnant and having a penicillin allergy, Watson can provide an entirely new set of guidelines. “The machine can match the user’s need and drill down and find the right information,” Chase said.
Watson could even go as far as to read electronic health records and make a preliminary diagnosis. It can even find clinical trials in which the patient could participate. “It is often the case that clues of a patient’s illness can be found in the medical record–it’s just the doctors haven’t found them yet,” Chase said.
Chase recalled a tricky case he came across when he had just graduated from medical school–one that Watson could have easily helped solve. A woman had been bedridden because of weakening muscles. Chase spent the better part of a month querying colleagues and investigating the literature, and finally found the right diagnosis: rickets. It was a tricky diagnosis because rickets is a bone disease, not a muscle disease. She had a very rare form of the disease.
Fast-forward 35 years later. Chase thought of this case and fed the symptoms into Watson. It provided the answer almost instantaneously. If Watson had been in place all those years ago, it could have saved the patient a month’s worth of hospital time. “Watson has bridged the information gap, and its potential for improving health care and reducing costs is immense,” Chase said.