We've seen this happen with airplanes. Where the autopilots result in the human pilots no longer being fully aware of their environment and context in that environment. So, when the autopilot alarms, the human pilot doesn't have the context of the problem and often makes a naive mistake as a result (a naive mistake that has fatal consequences).

There was a good article discussing this recently on here, a quick search should find it for more details.

So, I agree with the parent that semi-auto pilot like features are probably quite risky when intervention is required. At least current generations, but I think we can also make better designs that address these concerns.