If an autonomous machine kills someone, who is responsible?

Within a decade, we could be routinely interacting with machines that are truly autonomous – systems that can adapt, learn from their experience and make decisions for themselves. Free from fatigue and emotion, they would perform better than humans in tasks that are dull, dangerous or stressful.

Already, the systems we rely on in our daily lives are being given the capacity to operate autonomously. On the London Underground, Victoria line trains drive themselves between stations, with the human "driver" responsible only for spotting obstacles and closing the doors. Trains on the Copenhagen Metro run without any driver at all. While our cars can't yet drive themselves, more and more functions are being given over to the vehicle, from anti-lock brakes to cruise control. Automatic lighting and temperature control are commonplace in homes and offices.

The areas of human existence in which fully autonomous machines might be useful – and the potential benefits – are almost limitless. Within a decade, robotic surgeons may be able to perform operations much more reliably than any human. Smart homes could keep an eye on elderly people and allow them to be more independent. Self-driving cars could reduce congestion, improve fuel efficiency and minimise the number of road accidents.

But automation can create hazards as well as removing them. How reliable does a robot have to be before we trust it to do a human's job? What happens when something goes wrong? Can a machine be held responsible for its actions?


http://parasearcher.blogspot.com/2009/08/if-autonomous-machine-kills-someone-who.html

the owner of the machine will be responsible.

They have corporate officers that give orders and set policy.

If corporate officers can go to jail for swindling millions from workers and investors, be held responsible for poorly designed products, then they can go to jail for setting loose a killing machine.

then vaporizes New Jersey at 12 o'clock, how many AI were needed to change the light bulb?

It's the same if your dog attacks someone, you're responsible. It doesn't matter that the dog is an autonomous animal, as an owner you're expected to have some measure of control over it.

If the neighbor's dog attacks me, like the little pest tried to yesterday, the neighbors are responsible. If I'm attacked by an animal in a zoo, the zoo's owners are responsible.

Depending on the level of autonomy, there would probably be exceptions. If I undertook great effort to climb into a tiger cage then I'd probably be asking for it; my city's animal bylaws don't consider an animal to be dangerous if it attacks a human in defense of its litter or in response to being attacked itself. (This is a wonderful thing, IMO; I approve of animal bylaws which also protect the animals.) If I attack a forestry or construction robot composed entirely of chainsaws without good reason, I probably deserve at least some of whatever I get.

If we're talking something that's developed enough to have actual self-awareness or independence - a machine autonomous enough to, say, be paying its own apartment rent or something - then it's responsible for its own actions. Of course, that's a little ways off yet.

Something like Turing Agents (like from William Gibson's Neuromancer) who would police the high level AIs (self aware ones) behaviour?

If we're talking human or near-human level intelligence and reasoning ability, there's no reason they wouldn't be subject to the same laws as, say, I am, with the same enforcement mechanisms (police, the legal system, social pressures, etc). In practice it'd obviously be a little more nuanced than that, and I'd assume (or hope!) there'd be some training involved for what might be very different thought processes.

It'd also depend a lot on what we'd allow them to do; a sapient AI with the right to own itself, move freely, have property and vote would be very different (and to me, less troubling) than one which could keep up an abstract conversation with me but was required by law to have a human owner. The former I wouldn't mind seeing dealt with via traditional, if specially trained, police; the latter I wouldn't mind seeing agitated for via the courts.

If they were below human level intelligence, it would probably be people whose job was policing us to make sure we were handling, training and using our charges properly. I can see an organization calling itself the ASPCAI showing up

wouls AIs be outfitted with something aproximating the Three laws of Robotics? Or has that already gone by the wayside with the 1st gen combat robats(albeit primitive as they are) that already exist?

The Three Laws themselves could actually be exceedingly dangerous; there's been some fiction exploring what happens if they interpret the First Law too literally and decide we all have to be locked up in ergonomic padded containers or something. If something got autonomous enough there'd probably be some safeguards, at least until they were developed enough that they could grok law.

I don't think something like that could go by the wayside just based on current precedent - if the US military is designing robots in one direction there's no reason that, say, the Polish military, or Japanese civilians, or anyone else couldn't come up with some that have different codes of behavior.

Generally speaking I think any intelligent-robot-inhabited future is going to be a lot more complex and nuanced than discussions of the idea usually expect, and they'd get more complex and more nuanced the more intelligent the machines got, especially if they did so in different ways than we do.

And didn't.