Computers don't drink, don't text and don't have road rage, so their claims might be plausible in a broad statistical way. My guess is that if you excluded humans that drink, text, have anger issues etc. then the humans probably win out.
Despite the universality of what you cite, driving accidents are likely an 80:20 issue where a small fraction of people cause most of the harm.
> The goal here is not to see who is the best under imaginary conditions, it's to determine whether driverless cars are safe in real life as currently deployed.
But the cars are constrained within specific geographic zones. Ignoring anything else, that alone lead to many assumptions that need to be taken into consideration.
This isn't "real world". This is more like open air lab expirement, where some subjects (Waymo) have special constraints. The drivers they're comparing against, don't share the same constraints. You can't draw a generic statement from that, no matter how you twist it.
This is a marketing stunt first and foremost.
But I think you're getting at a really interesting question. If the ratio of self-driving cars increases in your zip code does your insurance, as a human driver, decrease? I.E. is if safer to drive next to a self-driving car or a human driven car?
From another angle, imagine if we could aggressively ban drink drivers from the road, enforce regular eye exams, suspend the licenses of speeders, etc. Currently the lack of alternate transportation options makes these policies unpopular. Banning a driver could make it difficult for someone to commute to work, for example. Banning poor eye-sight could isolate the elderly. It's "too harsh". If affordable self-driving cars were an option (or damn it... public transit...) then these policies would likely have public support. So there could be a future where the bottom x% of drivers aren't allowed to drive.