[bumby]

There are plenty of resources that demonstrate why those statistics can be misleading. Chief among them, not all miles are created equal. It’s like claiming autopilot in airplanes is significantly safer than ape-controlled aircraft. It’s partly true because apes control the hard parts (takeoff and landing) and leave the more easy parts to software

[judge2020]

> It’s like claiming autopilot in airplanes is significantly safer than ape-controlled aircraft. It’s partly true because apes control the hard parts (takeoff and landing) and leave the more easy parts to software

That's because the term "autopilot" is badly named, both in cars and in aircraft. People just thing autopilot in aircraft means "plane flies itself" because they get on a jetliner and never see the pilot actually control it, leaving the impression that some combination of magic and electricity-infused rocks got them there instead of a human with assistive software.

[grumple]

Those statistics are pretty relevant when it comes to fatalities, which are more likely to occur at highway speeds. Total accidents, yes, self driving cars aren't generally operating in city traffic yet. Those same resources will note that the most common accident for Teslas is being rear-ended by another vehicle, which is also relevant.

Let's not lose sight of the fact that this technology is under active development, with a theoretical target being eliminating the vast majority of car-related deaths. Nobody is arguing that it's already superior, though it may already be close in certain circumstances.

[bumby]

That’s the thing though. Those statistics aren’t showing that level of rigor. They aren’t even saying “miles driven at highway speeds”, just “miles with autopilot engaged”.

They don’t control for things like vehicle age. They don’t control for safety features, or even for driver assisting software control. They aren’t comparing driving conditions. They aren’t comparing driving duration. There aren’t comparing driving speed. They aren’t even from the same datasets. Etc etc.

It’s not a rigorous study but used as if it’s evidence when it’s only slightly better than anecdotal