Good point. People love to point out when AI gets stuff wrong, but has there been a systematic comparison to how often people get stuff wrong?

Kinda like how every time an self-driving car crashes, it makes headlines, but how safe are they compared to human drivers?