I think you're right. I still think it's the right direction to go because when a human kills a child we can't<li><p>pull their logs</p></li><li><p>simulate the event</p></li><li><p>understand precisely the sequence of decisions made by the person that resulted in a death</p></li><li><p>push an OTA update to every single driver on the road so it never happens again</p></li>
... but people like to *feel* far more in control than they de-facto *are*, and I think the feeling of "robot killed my kid" and "stone-cold sober driver who had a single moment of inattention killed my kid" (much less "drunk driver" or "driver having a heart attack") is viscerally different in a way that is likely counter-productive to maximizing vehicle safety.