IBM’s Watson computer system successfully competed against the two most successful “Jeopardy!” contestants — Ken Jennings and Brad Rutter.
IBM looks to make 'Jeopardy!' computer a doctor's helper
- Article by: STEVE ALEXANDER
- Star Tribune
- August 14, 2011 - 2:19 PM
Meet your doctor's new assistant. He listens well, asks a few questions and reads 100 million pages per second.
Or at least that's the vision IBM has for Watson, its computer system that in February bested two human champions to win the "Jeopardy!" TV show -- a challenge that required encyclopedic knowledge and split-second answers.
Medicine is one of the first practical applications IBM sees for Watson, named after IBM's first president, Thomas J. Watson Sr. So IBM's Watson experts recently met with the Twin Cities medical community to solicit their help in turning Watson into an expert physician's assistant.
While Watson is only in the early stages of development for use in medicine, IBM says it could help doctors zero in on the correct diagnosis and treatment of illnesses, enable medical providers and insurance companies to agree on which treatments should be covered in different situations, and amass data that could be used to defend medical providers against lawsuits.
"We're confident, from what we're seeing so far, that Watson can make a substantial contribution to medicine," said Martin Kohn, a doctor who is IBM's chief medical scientist at its Watson Research Center in Hawthorne, N.Y. "But people outside IBM say 'Show me,' and we think that's reasonable."
The key to Watson is that it is far more than a number-cruncher or a search engine, Kohn said. Watson 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 a databank. Watson can also ask questions to add to its storehouse of information, and can give answers that are tagged with ratings representing Watson's "confidence level" in a particular answer.
But would your doctor want a computer's help in diagnosing and treating your illness? There were differing opinions at an Edina lunch that IBM hosted for Twin Cities medical professionals.
On one hand, the answer is clearly yes. There is now so much new medical research that no human being can keep up with it, IBM and Twin Cities medical experts agree. Watson could read nearly everything related to the topic, decide what's most relevant and prioritize the highlights for a doctor.
"If a doctor reads medical journals for two hours every night, in one month he or she is 40 years behind due to the sheer volume of new information," said Gary Oftedahl, a doctor who is chief knowledge officer for the Bloomington-based Institute for Clinical Systems Improvement, which helps health care organizations and insurers create standards of care for various illnesses. "Watson's ability to synthesize 100 million pages a second holds great potential as a resource to enhance my ability."
Others had different tasks they wanted tackled.
"I'm sitting here with this huge data management problem," said Alan Abramson, senior vice president and chief information officer for medical provider HealthPartners in Bloomington, where electronic patient records contain an amount of data equivalent to 20 million four-drawer filing cabinets filled with text. "And Watson reads fast."
But, to be effective, Watson needs a way to tap in to confidential patient records without violating privacy laws, Oftedahl said.
"If Watson gets its synthesized data from me, it's still limited by what I don't know," Oftedahl said. As a next stop, Watson needs to "become an aggregator of patient record information," he suggested.
Kohn proposed two possible solutions: Personally identifiable information could be removed from electronic medical records before they are given to Watson, or IBM could act as a trusted repository of patient records so Watson could search them.
But some worry that Watson could be used to "grade" doctors, an unpopular idea in medical circles, where decisions are at least partly subjective.
"Doctors do not want another tool that judges their performance," said Jeff Schiff, a doctor who is medical director of health care programs for the Minnesota Department of Human Services. "And not all medical evidence is equal. How can Watson understand the nuances of what makes a good medical study?"
As a result, using Watson to recommend diagnoses or treatments might spark a silent rebellion among doctors, Schiff said. Rather than accurately report a patient's symptoms and accept Watson's recommendations, doctors might fudge the reporting of symptoms to get Watson to make the recommendations they want, he said.
At the other end of the spectrum is Nicholas LaRusso, a doctor who is the director of the Mayo Clinic's Center for Innovation in Rochester, who wonders if Watson might help the nation alleviate a projected shortage of physicians.
"We'll be 80,000 doctors short in the U.S. by 2020," LaRusso said. "Are there situations where Watson could answer questions directly to patients? I don't know."
The Mayo Clinic, one of the country's premier health organizations, has had talks about Watson with IBM, whose Rochester facility built the 10 computers that make up Watson. But it isn't currently a partner in the Watson project, LaRusso said.
Others object to the idea that Watson could be used by insurance companies to predetermine what treatments will be covered for particular illnesses.
"Using Watson to preauthorize care or treatment would be an incredible mistake because it doesn't involve the patient," said Carolyn Pare, CEO of Buyers Health Care Action Group of Bloomington, a group of public and private Minnesota employers that deals with the price and quality of health care. "Doctors will then just recommend the procedure that has the most insurance coverage. And what insurance will pay for and what's good for the patient are not always the same."
IBM's Kohn said this isn't the first time concerns have been raised about Watson's role.
"Watson is not a tool to challenge or threaten doctors," Kohn said. "It's a tool to get more efficient access to information that is more relevant to the decisions that doctors have to make."
IBM won't say how much it spent to develop Watson, although outside estimates range from $100 million to $2 billion. It's equally unclear how much IBM would charge the medical community to buy or rent a Watson. But IBM is firm about one thing: It's looking for big medical industry partners to help share the costs of further developing Watson for medical use.
"Right now Watson is being trained and tested against standardized multiple choice questions in medicine, where we already know the right answers," Kohn said. "But we're working with medical faculty members at Columbia and Johns Hopkins universities, we have a group of doctors at IBM that are doing research and we are looking for other partners that are well-respected health care organizations."
The Twin Cities medical community seems cautiously interested in participating.
"We're interested in Watson, but the cost of any of these ideas is a long way from being finalized," Abramson said. So far, "Watson is a terrific solution for a problem that hasn't been specified."
Steve Alexander • 612-673-4553
© 2017 Star Tribune