Internal documents show that IBM Watson Health’s supercomputer, designed to supplement physicians’ clinical judgment, often recommended unorthodox and unsafe cancer treatment options. The problems were discovered within the Watson for Oncology program at Memorial Sloan Kettering Cancer Center (MSKCC), where operators input the system with synthetic cancer cases, rather than real patient data.
Watson Health uses IBM’s Watson supercomputer to analyze vast amounts of health-care data and make treatment recommendations. IBM’s acquisitions of health-care companies have added to the database available to Watson, potentially allowing for improved recommendations and patient outcomes. However, Watson Health has faced unexpected challenges integrating these new datasets, and its supercomputer has been criticized for its failure to live up to customers’ expectations.
Official IBM product information indicates that Watson for Oncology makes treatment recommendations based on “thousands of historical cases and insights gleaned from thousands of [MSKCC doctor] and analyst hours.” The internal Watson Health documents, though, show that doctors and engineers at MSKCC would often input synthetic data into the computers, which would then be used as the basis for Watson for Oncology’s recommendations. This process, IBM’s customers and rivals say, demonstrate that Watson Health has been dishonest about its methodology.
The company, however, defends its product. “We have learned and improved Watson Health based on continuous feedback from clients, new scientific evidence, and new cancers and treatment alternatives,” read a statement released by IBM. “This includes 11 software releases for even better functionality during the past year, including national guidelines for cancers ranging from colon to liver cancer.”
Others remain skeptical of IBM’s approach to data analytics and its transparency with its methods. “They’re making up personas of cancer patients, basically,” said CEO Jana Eggers of Nara Logics, an artificial intelligence company. “Why do you do that when you have the real people?”
Source: STAT News, July 25, 2018.