I know its bad having a machine-controlled car killing someone, but.. it will happen and we must accept it and my guess is that it will happen far less than with human operators in the long-term.
> The NHTSA reports a fatality rate of 1.25 deaths per 100 million miles[2], twenty five times the [4 million miles] Uber has driven.
An automatic emergency brake doesn't prevent a computer or human driver from braking when it wants to. If the autonomous system can't even avoid interaction with emergency brake, the prototype belongs in a lab, not on a public road.
This isn't a situation where laissez-faire is a good idea. If we want autonomous vehicles to be safer than human drivers then we need to ensure that is the case with proper regulation, audits, standards of testing, and liability.
We who work in the software industry should know better. Yes, you can make software to a very high standard, even high enough to bet your life on it. Is all software made that way? Will software be made that way without some sort of pressure to do so? No, of course not. Every shortcut will be made, technical debt will be piled like cord wood until it reaches the sky. And people will die.
If autonomous vehicle technology is real, and if it can improve on human drivers, then let it be set to that standard from the start.