Take AI-assisted self-driving cars. SUPPOSE for the sake of argument that self-driving cars CAUSE fewer accidents than they prevent, CAUSE fewer deaths than they prevent, CAUSE fewer serious injuries than they prevent, and the medical and other dollar costs of the accidents they CAUSE are less than the money saved by the accidents they prevent, it would be an overall win, EVEN IF some people die who would otherwise not die.
We don't have enough data to say if this supposition is actually true, but suppose it is.
The same is true for the "seatbelts save lives" argument: It's a net win, even if there are some edge cases where the use of a seatbelt may kill someone who might have otherwise survived.
Ditto medicines. Even for some common medicines that have been around for decades like penicillin, we know that overall they save lives but every now and then someone has a fatal reaction, killing a patient who was not at risk of dying.
IMHO, there are still too many unknowns with AI to let the industry go "full steam ahead" without putting in automated and human guardrails.