Healthcare is on the end of the spectrum where small mistakes have grave consequences (eben though human physicians make mistakes all the time). There are plenty of problem domains though that are a lot more resilient or can be made resilient to erroneous output, and those are significantly more likely to be displaced by AI.