@sickburnbro With cars it's especially rough because I won't buy a car that doesn't prioritize me/my wife over every other living creature. I'm fine with it making a safe attempt at stopping, but I'm not paying tens of thousands of dollars to be ethics-theory'd off a bridge. And while plenty of internet ethicists will claim otherwise, they aren't likely to do so either.

As for the Speech-to-Text stuff, I use it a lot for synced lyrics for songs, but it's spotty enough that it almost always requires checking. Still speeds up the process about a ton though. Because now it's two listens instead of fifty. So the liability should still be on the person knowingly trusting an 80-95% accurate STT model instead of doing an initial transcript and a second listen to correct mistakes.