'Bad' passengers and 'bad' drivers are a significant problem with public transit? Huh?
It's the same reason a web forum with firm moderation and anti-troll policies will have higher-quality discourse.
("Get a bike" is not a universal solution; bikes cost money, and the kind of winter wear you need to ride across town in a cold climate without getting frostbite costs more money, and a lot of people don't get paid enough to have that much squirreled away even before their life went to shit.)
As you can tell, this bus analogy irks me.
{1} On-demand vs. scheduled
{2} The given number of cars in the field sets a harder upper limit on number of riders. With buses, it's basically always "pack more people onto the bus, no matter how crowded" (i.e. the bus is forced to approach closer to "theoretical" carrying capacity of vehicle's physical volume - compare this to how much Uber is used for single riders.)
{3} Anti-"troll" filtering (I mentioned elsewhere that this works for the same reason it works on web forums - you eliminate the irresponsible elements.)
{4} No routes with Uber: go where you want. Normally this flexibility costs you at the rate of a taxi.
Etc.
1: Pool/Lines add a lag to search for a match, so it's a bit less on-demand and more favorable others' schedules.
2: Pool/Lines increase the riders/vehicle.
4: Lyft Line hotspots reward you for picking start points at special locations, and ending within a zone, so it's somewhat route-constrained in return for a lower price.
The article's prediction (and Uber's upcoming initiatives) simply go further in that direction: more constrained routes, more constrained times, lower prices. It's not the sharp boundary you suggest.
If they're partly on-demand and on a more convenient route, even a higher cost may be worth it. And ultimately anyone is free to ignore the offer if costs are too high.