Well... I think what they'll probably do is continue testing - and they probably won't be widely deployed until they're as safe as human drivers (on average, they'll probably be safer in some ways and less safe in others). Soon after that, they'll be safer than humans (because they can share knowledge, are easy to upgrade, and once there's lots of them they'll be able to communicate in ways humans can't).. well, that is, if we keep going.
I say "if", because the more likely problem is Luddites who will want them banned after the first death, even if their overall safety record is better than humans. An enormous number of extra people will die because of how slowly we'll adopt self-driving cars. This is because people are dumb and ruled by emotional reactions: when people cause collisions (which they do thousands of times a day) it's just an accident, but the first time a self-driving car runs over a kid it's going to be pandemonium - and a good percentage of people will want to go back to the old higher death rates.
As to your argument, it's difficult to compare a computer to an ant or a person on some single scale of intelligence. An ant is very good at some things, but completely incapable at most everything else. Computers exceed humans at many tasks, while lagging behind in others. No computer today could learn how to drive well by itself, or have much conception of what driving is - but we've demonstrated that computers, designed and refined over time by people, can get very good at complex tasks. I think we're still a ways off from having safe computer drivers, but it's not in any way impossible or far distant; computers are already much closer to "humans" than "ants" on the "ability to drive" standard, and there's no reason they couldn't be better than humans at driving within the next 10-20 years.