One problem that enters into this situation that does not apply to the case of human drivers, and that may hold this process up for some time is the diffusion of what is now individual responsibility into the corporate domain. This is a general problem with the deployment of artificial intelligence, and we will face many versions of it soon. But so far, the problem has always been mapped back onto humans. In the case of an autonomous vehicle, this would be quite difficult.
Corporations take responsibility the same way individuals do, but they are able to shift the parts that compose themselves around to evade apparent responsibility or to hide resources available for remediation from those to whom they would be owed. They have been known to dissolve completely, escaping liability by having no assets while a 'different' corporation is constituted with the exact same human members, free of historical ties to the past behavior of the same group of people.
As it is (at least in the U.S.) determining who is at fault in an accident is a legal process between insurance companies representing individuals. The facts are ascertained from eyewitnesses, if only in the form of the two drivers themselves. Approximately equal resources and equal process will be deployed for both drivers, and they are aggregated by the requirement of mandatory insurance into large enough groups that if a large settlement is determined, the individual will not be financially destroyed (though they may never again be able to get insurance and thus may no longer be allowed to drive.)
If a machine makes an error, and the passenger is not in a position to take responsibility for knowing what happened, we are left with a difficulty assigning blame. The contest of attempts to prove fault is vastly unfair, strangely it is so to either party, but in wholly different ways.
- The corporation and its machine can expect very little empathy from
humans, who will constitute the jury or magistracy making the decision; but
- The same resources cannot possibly be deployed on behalf of the
human that will automatically come forward to defend the machine.
We have no clue how to insert equality here. We don't know which party is more disadvantaged, and we don't have faith that if the car company is just put out of business the existing cars already produced will not become unfortunate burdens upon their owners.
(The German government raised the same issues with respect to open-source software back when it first became an important force in the computing industry: that it proposed a problem of responsibility that had no parallel in earlier law, and it was unclear what would constitute fairness.
Elaborate networks of proxies and escrow schemes were finally instituted to buffer the court system against the risks of getting this wrong. A similar situation might arise here, with auto-insurance companies basically changing in form in a way that simply removes the issue by convincing the courts that this not so bizarre, after all.