A lengthy piece in Wired:

But what does the law say about autonomous vehicles?

“The law in California is silent, it doesn’t address it,” Google’s Anthony Levandowski told me. “The key thing is staying within the law — there’s a always a person behind the wheel, the person in the seat is still the driver, they set the speed, they’re ready to take over if anything goes wrong.”

Ryan Calo, who studies, among other things, the legal aspects of robotics at Stanford University’s Center for Internet and Society, notes, “generally speaking, something is lawful unless it is unlawful — that’s the whole idea of having a system of so-called ‘negative liberties.’”

He has parried on this issue with economist Tyler Cowen, who counters with one local driving code, which states “No person shall operate a motor vehicle upon the streets of the city without giving full time and attention to the operation of the vehicle.” And yet by this definition alone there was nothing illegal about what the Google engineers, sitting up front and busily monitoring the Prius’ various operations, were doing. They were within both the spirit and letter of the law.

In fact, you could argue they were paying more attention than any of the drivers around us.

Ryan is a friend. Keep an eye on him.

So is Nevada simply reaffirming what’s on the road, or raising the specter that existing in-car technologies would be subject to the requirements laid out in the draft regulations? For example, would the requirement that “prior to testing, each person must be trained to operate the autonomous technology, and must be instructed on the autonomous technology’s capabilities and limitations” apply to someone test-driving a new Mercedes-Benz?

Calo doesn’t think so.

“It sets out a definition of autonomous technology that is based in part on the statutory definition of autonomous vehicle that the legislature gave to the DMV, and then it very explicitly excludes essentially all of the individual technologies that are commercially available today,” he says. “It’s an open question if it would exclude them in their combination or in their particular uses but it reflects the DMV’s effort to say no, no, no, we’re really not talking about things you can buy today, we’re really not talking about the driver assistive technologies, we’re talking about autonomous in its extreme sense.”

And this is the million-dollar question that applies to all aspects of AI (not just autonomous cars)–what happens when the decision-making is shifted from the human to the machine?

Welcome to the brave new world of autonomous vehicle law. When the rules become official, Nevada will become the first state to broach the subject (largely thanks to Google’s influence). But as increasingly autonomous technologies enter our cars, the legal questions, liability in particular, have grown more relevant. As a report by the RAND Corporation notes, “As these technologies increasingly perform complex driving functions, they also shift responsibility for driving from the driver to the vehicle itself…. [W]ho will be responsible when the inevitable crash occurs, and to what extent? How should standards and regulations handle these systems?”

 …

As der Heijden and van Wees write, “under fault-liability regimes drivers and vehicle owners will not be liable if they acted as a careful person.” But what’s the definition of a careful person in an autonomous vehicle? How could we prove ex post facto they were monitoring the car’s performance and not simply daydreaming?

Or what if the autonomous or semi-autonomous vehicle is a Mercedes-Benz using a hypothetical Google geolocation product and it crashes into a barrier while headed for an off-ramp because it misjudged its location? Is fault attributed to Mercedes (acting on the information), or Google (providing the information), or the driver for not correcting for the error? . . .

An interesting and related area of inquiry here is product liability in the case of crashes that occurred as drivers were given incorrect coordinates by navigation systems. As legal scholar John Woodward notes in an article in the Dayton Law Review, finding fault in that case requires not only locating the source of the malfunction (software, hardware, or a triangulation snafu involving a wayward satellite in the heavens), but ascertaining whether the navigation instructions doled out by GPS constitute a product or a service — which would render product liability claims moot.

I just blogged about this the other day.