If it's terminal brain fart by an active driver, that sux. If it's an example of auto-driving having issues, or worse, a Darwin award from somebody uninterested in paying attention, that's a datapoint for the argument as to if/when we'll see auto-drive.

While it's possible to design fault-tolerant systems that are quite capable, I feel they're at a level of cost and effort that's not being used for current-day programming in pretty much any form. I know they've been working on computer vision since the 60's so I assume that's solved (enough) to trust its data.