I think it's orthogonal question, right? Even if there were some mistakes, how many more death there would have been if they were not thinking about rigorous process behind ML development?
I would not use Silicon Valley technology to A/B test death rates by 'trying' arbitrary 'cool' things inside a fast moving metal torpedo. He can play with toys as he pleases but using real cars seems absolutely bonkers.
