Reports in Stat News and Gizmodo claim that internal IBM documents show that Watson for Oncology made "unsafe and incorrect treatment recommendations" multiple times. For example, in one case, a 65-year-old man with lung cancer was recommended a drug that can cause hemorrhaging, even though the man was already experiencing severe bleeding and shouldn't have been a candidate for the drug.
The report says the problems may have been caused by training Watson on hypothetical patient data rather than on real data.
“This product is a piece of s**t,” a doctor at Florida’s Jupiter Hospital said to IBM. “We bought it for marketing and with hopes that you would achieve the vision. We can’t use it for most cases.”