Doctors Win the First Round Against Machines – IBM Watson's Treatment Recommendations Are Not Accurate or Safe

IBM and medical specialists working with its Watson supercomputer made a statement prompting the thought, "Fortunately, our patients' health is still entrusted to us doctors" as a reaction. A report published on 25 July 2018 stated IBM Watson had "repeatedly made unsafe and incorrect treatment recommendations".

Although less reported because it is not sensational, not every story about artificial intelligence or supercomputers is positive. Science and technology progress through precisely such ups and downs, but neither technologists nor scientists give up, and ultimately progress continues forward.

The report interpreted documents including two presentations published in June and July 2017 by former deputy health chief Andrew Norden, shared by IBM Watson health management.

Watson, which decisively won a quiz competition in 2011 and was later developed to make oncology treatment decisions, now faces strong criticism. Its testing in oncology had excited all of us; we even shared an article on our website describing IBM Watson's development processat the time.

The expected positive news did not arrive over subsequent years. Instead, in February 2017, MD Anderson Cancer Center announced early termination of the failed IBM Watson Project, which cost 62 million dollars. Reasons for its frequent incorrect decisions remain under investigation while the system continues to be improved. To

give an example from the documents, consider a 65-year-old lung cancer patient with severe bleeding. Watson recommended chemotherapy and the antiangiogenic drug "bevacizumab". But this drug can cause significant bleeding and should particularly not be given to patients already bleeding. A Memorial Sloan Kettering Cancer Center spokesperson said the recommendation was not applied to a real patient and was used only for system testing.

Is Watson Alone at Fault?

IBM blames the MSKCC doctors and IBM engineers who designed Watson in 2012 to think like a doctor. Documents state that instead of feeding real patient data into the software, doctors supplied hypothetical or "synthetic/artificial" cases.

Some doctors find IBM's results troubling. One MSKCC doctor said, "This product is a real nuisance. We bought it for advertising, and I don't think we can use it for most of our treatment options."

This doctor was reportedly one of many whose complaints appeared in the documents.

However, other hospitals use Watson as an additional opinion when doctors disagree about treatment.

Responding to the STAT website report, an IBM spokesperson said Watson for Oncology was trained to assist oncologists treating 13 cancers at 230 hospitals worldwide and had supported more than 84 thousand patients. They added that the software was still being improved for better functionality.

Memorial Sloan Kettering spokesperson Caitlin Hool described criticism as a "natural process" for evolving software such as Watson. Hool also said; "While Watson offers cancer treatment options, the final decision requires the treating doctor's involvement and clinical interpretation" and added: "No technology can replace a doctor."

Our Commentary on Artificial Intelligence and Watson in Oncology

Even if not in clinical decision-making (treatment selection), current artificial intelligence has made major progress in diagnosis and will take over much of radiologists', dermatologists' and pathologists' work in the near future (for example, in diagnosing the skin cancer melanoma).

Clinical decisions, however, are much harder and more complex than answering encyclopedic quiz questions. Doctors account for many "dynamic" factors, including patients' psychological characteristics and treatment costs...

Perhaps initially training Watson for oncology, medicine's most difficult field, was a mistake; perhaps doctors did not know how to train it, were unwilling, or could not collaborate fully with engineers... But we believe machines will soon understand reality, and in future rounds artificial intelligence will challenge and even outperform doctors in treatment decision-making.