Bad news for self-driving cars...

....is the sad fate of a man using this feature on his Tesla Model S, as the software failed to detect a semi in his path--suffice it to say it will probably be a closed casket funeral, to put it mildly.  At this point, it is one fatality in 130 million miles, and that fatality rate ( 7 deaths per billion vehicle miles) is comparable to the overall vehicular fatality rate of ten fatalities per billion vehicle miles.

While the confidence range is huge with n=1, and Tesla and others will certainly improve their image recognition software as a result, it's still troubling, even with the company's explanation and promise.  The fact of the matter is that the key drivers of accidents--alcohol and fatigue--were obviously not involved (at least on the autopilot side), and hence we'd infer that this fatality rate is actually fairly high versus that of sober, attentive drivers.

This illustrates a big weakness with vehicular autopilot systems; the key driver of vehicle safety, really, is the proverbial "nut behind the wheel", and autopilot allows that nut, as it were, to come loose a bit.  Count me out of efforts to rush this to market.