https://arstechnica.com/science/2023/05/ais-chaotic-rollout-in-big-us-hospitals-detailed-in-anonymous-quotes/

When it comes to artificial intelligence, the hype, hope, and foreboding are suddenly everywhere. But the turbulent tech has long caused waves in health care: from IBM Watson's failed foray into health care (and the long-held hope that AI tools may one day beat doctors at detecting cancer on medical images) to the realized problems of algorithmic racial biases.

But, behind the public fray of fanfare and failures, there's a chaotic reality of rollouts that has largely gone untold. For years, health care systems and hospitals have grappled with inefficient and, in some cases, doomed attempts to adopt AI tools, according to a new study led by researchers at Duke University. The study, posted online as a pre-print, pulls back the curtain on these messy implementations while also mining for lessons learned. Amid the eye-opening revelations from 89 professionals involved in the rollouts at 11 health care organizations—including Duke Health, Mayo Clinic, and Kaiser Permanente—the authors assemble a practical framework that health systems can follow as they try to roll out new AI tools.