My view is that even if self-driving car corporations kill people with zero consequences (due to poor laws and poor oversight), as long as they're doing so at lower rates than human drivers are, that's still a net win. It's still not an ideal scenario, because those deaths could be further reduced under the threat of real consequences. But by and large, it'd still be better than what we have today.

If they're killing people during testing of experimental tech that hasn't been approved for general use (like the Uber case you mentioned), they need to be smacked down hard. If they're killing people at a higher rate than human drivers, they should not be approved for general use in the first place. If they're approved, but release an update that ends up killing more people, they again need to be smacked down hard, with the update immediately reverted.