An interesting look at a question that was once hypothetical, but is now a very real consideration for policy makers considering the future of autonomous vehicles. 

How many people could self-driving cars kill before we would no longer tolerate them?

It's a consideration which goes to the heart of the development of autonomous vehicles and is not one which is based on statistical safety, but one based on public perception and public acceptance. The question for policy makers as they try to develop legislation which keeps pace with the progress of technology, which is at present 'unavoidable imperfect', is whether improving statistical safety alone is a sufficient barometer for legislative purposes.