> And all this is to say that these "replace human activity" approaches wind-up with the problem of doing an "90%" of the activity right and then foundering on

> And all this is to say that these "replace human activity" approaches wind-up with the problem of doing an "90%" of the activity right and then foundering on corner cases - like self-driving cars that are easy-yet-impossible AI tasks. I think this is the crux of the issue.