Way back in 2010 I did some paper reading at university on AI in healthcare, and even back then there were dedicated AI systems that could outperform many healthcare workers in the US and Europe.

Where many of the issues came were not in performance, but in liability. If a single person is liable, that’s fine, but what if a computer program provides an incorrect dosage to an infant, or a procedure with two possible options goes wrong and a human would choose the other?

The problems were also painted as observational. Often, the AI would get things with a clear solution right far more, but would observe things far less. It basically had the same conclusions that many other industries have - AI can produce some useful tools to help humans, but using it to replace humans results in fuck-ups that make the hospital (more notably, it’s leaders) liable.