While IBM promoted Watson for Oncology to hospitals around the world as a tool that could help recommend cancer treatments, internal company documents told a more troubling story. In an investigation published on July 25, 2018, STAT reporters Casey Ross and Ike Swetlitz described slides from internal presentations given in 2017 by Watson Health’s deputy chief health officer, an oncologist, summarizing criticism the product had received.
The documents stated that Watson for Oncology had produced “multiple examples of unsafe and incorrect treatment recommendations.” In one cited case the system recommended a drug that could cause severe or fatal bleeding for a patient who already had serious bleeding. The investigation also reported that the software had been trained not on the records and outcomes of real patients but on a relatively small number of “synthetic,” hypothetical cases, with treatment recommendations reflecting the preferences of a few specialists rather than broad clinical guidelines or evidence. Some physicians at hospitals that were helping promote Watson reportedly told IBM privately that it was not useful for treating patients.
This reporting was an early, concrete crack in the Watson-revolutionizes-medicine narrative that IBM had built since Watson won Jeopardy in 2011, and it foreshadowed IBM’s eventual divestiture of Watson Health in 2022.
The lesson is that marketing momentum can run far ahead of clinical reality. A system trained on hypothetical cases and a handful of experts’ opinions, then sold as authoritative medical guidance, is dangerous precisely because its confident output invites trust it has not earned.