How IBM's Watson Went From the Future of Health Care to Sold Off for Parts

ibm-watsonai-failurehealthcare-aicase-study

Abstraction: IBM Watson Health $5B AI healthcare failure sold for parts

Key points:

IBM spent ~$5 billion acquiring health data companies (Truven, Phytel, Explorys, Merge) to train Watson Health, which it sold to Francisco Partners for just over $1 billion in 2022
Watson Health failed in part because training data from Memorial Sloan Kettering (upper east side Manhattan patients) did not generalize to patients in other geographies or clinical contexts
Oncologists at MD Anderson and elsewhere found Watson's treatment recommendations either obvious, irrelevant to local protocols, or unavailable locally; an internal 2017 IBM document showed a physician called it a "piece of shit"
IBM's marketing got far ahead of capabilities; CEO Ginni Rometty called healthcare Watson's "moonshot" in 2015, but the technology was nowhere near ready
The failure is a cautionary tale about combining tech bravado with healthcare, but is not seen as evidence that AI in healthcare is impossible — rather that IBM's approach and the technology generation were both flawed

Connections: Ibm · Ibm Watson · AI In Healthcare · AI Failure