Such a system would still be much more inefficient than fixed-route transit because of the winding routes required and insufficient concentration of demand.
I would like to see the math on that. It's inefficient in terms of bus-miles driven, but assuming that we value people's time, it's not clear to me it would be inefficient in dollar terms.
Yes, insufficient concentration is a problem, which means that living in the suburbs would still be more expensive in transport costs. I still expect (or perhaps hope) that we'd see a contraction back to more dense urban cores. But automated transportation could provide a bridge between the suburban model and whatever comes after.
For the old sort of public transit, I think that's true. But our current notion of public transit is based on fixed schedules, fixed routes, and the high cost of human driver labor.
But think what a new public transit would mean. If everybody has a smartphone and vehicles are automated, then public transit could mean vans and small buses that dynamically route to come near you.