True, which doesn't invalidate the author's point. I'm merely adding a second set of considerations to the discussion. I think it's worth drilling down a little harder on the author's point though. They correctly note that there are no life-or-death systems that are wholly computerized with no human supervision. To me that speaks rather strongly to either a lack of awareness or lack of ethics on the part of drive AI advocates, as they're cheerleading something the military, aerospace, and medical industries have rejected on merit.

Most industries don't have a million deaths a year as incentive.

For most industries zero deaths is the norm, so clearly computers have to be at least that good. Frankly we're so collectively bad at driving that we'd be an order of magnitude better if the computers only killed 100 000 a year.

In over 100 years of driving humans have, at best, gotten worse. We continue to kill despite cars being better and safer.

On the other hand 100 years of improvement to self-driving cars will (I predict) drop those deaths to under 1000.

You're attributing to operator error what is more likely caused by poor vehicle design, incomplete understanding of crash mechanics, and immature material science and manufacturing techniques. As early as the late 90s you could still buy a car that would force-feed you the steering column in a head-on collision.

I feel operator error (drunk driving, disregard of road rules, speed limits,signage, deteriorating vision) and so on are the cause of most road deaths.

Referring to the effects after a collision, and not the cause of the collision itself, is valid, but secondary.

How have we gotten worse? Deaths have plummeted.