0
$\begingroup$

For a storytelling project, I'm trying to imagine a world in which fairly advanced, bipedal robots are being developed and produced by private companies. As these robots would be capable of easily harming and killing humans, the government needs to/wants to force all robots to abide by Asiimov's Three Rules.

What efforts could a government make to achieve such a task and absolutely make sure these private companies don't develop killer robots behind closed doors? Would the threat of potential law suits, frequent and close inspections etc. be enough? Would there have to be a final step in which government engineers "hardcode" the laws into every robot's source code?

(Not necessarily relevant to the question, but in the given narration, one of these companies finds a way to abuse their robots by uploading a human consciousness into it, therefore implying that it's impossible to circumvent the Three Laws by merely reprogramming the robot.) Sorry if the question's kinda diffuse or weird. I'm neither eloquent nor gifted with words haha.

Looking forward to any and all replies!

Improve this question
$\endgroup$
10
  • 5
    $\begingroup$ You realise that most of Asimov's 3 laws stories were about how the 3 laws weren't sufficient to protect either robots or humans $\endgroup$
    – Separatrix
    Commented Oct 11, 2018 at 20:49
  • 3
    $\begingroup$ "What efforts could a government make to achieve such a task and absolutely make sure these private companies don't develop killer robots behind closed doors?" - there were at least few entire books about just that, and Asimov wrote about fails of this system even when it was enforced... $\endgroup$
    – Mołot
    Commented Oct 11, 2018 at 20:58
  • 3
    $\begingroup$ We already have large machines capable of harming and killing humans. Governments regulate their manufacture by requiring seat belts and air bags, and regulate their use with traffic laws and licensing. -- As far as absolute measures... well... mostly effective measures with automobile safety are better than no measures at all. $\endgroup$
    – Ghedipunk
    Commented Oct 11, 2018 at 21:33
  • $\begingroup$ Hi bramble, and welcome! Please take a moment to review the help center and tour --- those resources have lots of good guidance as to how to ask & answer queries here at Worldbuilding. Right now, your query is scheduled for closure due it being too broad. This is your opportunity to edit it in such a way that you focus on one aspect of the problem you're facing in your worldbuilding. You can ask more than question (in separate posts!) but SE works on the one question one answer model. $\endgroup$
    – elemtilas
    Commented Oct 11, 2018 at 21:42
  • $\begingroup$ Asimov also never explained the implementation of the three laws. It was unclear whether that was a design choice or a fundamental nature of the positronic mind. $\endgroup$
    – Cort Ammon
    Commented Oct 11, 2018 at 21:48

5 Answers 5

Reset to default
3
$\begingroup$

Take a look at this comment from Separatrix:

You realise that most of Asimov's 3 laws stories were about how the 3 laws weren't sufficient to protect either robots or humans

What you ask is akin to "how can a government keep people from commiting crimes". Short of replacing every single human citizen with a robotic one, there is nothing your government can do.

If you still wish to push it, the government would have to control every single robot factory to make sure the laws are implemented in hardware and software. Even then, you can never be sure.

Improve this answer
$\endgroup$
2
$\begingroup$

The same way modern complex machinery can meet (or fail) safety standards. How can we make sure that Tesla's autopilot is not secretly trying to kill human drivers?

Government may require that companies would submit AI source code for certification. Then government's own AI can run this code through extensive simulation scenarios. Even if a well hidden "killer" feature is not found, there would be sections of code associated with it which would never get executed.

Improve this answer
$\endgroup$
1
  • $\begingroup$ Since a true AI is, of course, able to solve problems it encounters that are beyond it's original programming, seems like an implied need to recall the AIs periodically for re-examination, too. $\endgroup$
    – user535733
    Commented Oct 11, 2018 at 23:13
1
$\begingroup$

Government can regulate anything, including the mundane, but especially if something is hazardous like killer robots. In a free country, the laws are driven by politics. Rival political parties could be falling over each other, claiming they are the most "socially responsible" in their zeal in saving lives from evil robot manufacturers, while the other party only cares about profits, not people.

In a tyrannical country, the owner would be responsible for the robot's actions, even unto penalty of death.

In a leftist society run amok, the cry could be "androids are people too", and they would be given rights, including the right to vote and the right to a trial if they were accused of murder. Those against such measures would be baited as "androidophobic".

In a world run by corporations, one corporation may come up with the ability to restrict robotic violence. If other companies lack the same capability, they would be at a severe competitive disadvantage if robot restrictions was a legal requirement.

Improve this answer
$\endgroup$
1
$\begingroup$

Basically, the firm's AI software would need to be certified; not only in that it follows the Three Laws, but in that the Three Laws can't be (easily) circumvented by e.g. rooting a robot or installing a software override.

With technical progress allowing ever smaller realities to produce powerful enough AIs (for example, four massive software installations developed in the 2000s and requiring a bank's mainframe are now running simultaneously in four VMs on my laptop. The software running the Apollo missions in the 1970s can now run on the average smartphone. And so on), this becomes more and more of a problem.

Unless some homegrown plague stemming from the analogous popularization of DNA splicing kills everybody off first, sooner or later the government will need to employ teams of experts to hunt and shut down rogue AI home "research centers" in basements and attics, lest they trigger a Singularity apocalypse. I'm expecting the team members to be known as "Blade Runners".

Improve this answer
$\endgroup$
0
$\begingroup$

If robots require AI (as they typically did in Asimov's stories), AI is essentially mysterious. It's unclear why humanity hasn't managed to invent this yet. It may be true that whatever makes humans intelligent also intellectually blinds them to the process/mechanism that allows intelligence... otherwise it's difficult to reconcile the great advances we've seen in computers with the dearth of anything like positronic brains.

If this is not the case, and it's just a hard problem we haven't solved yet... then it is still an intellectually tricky one. It wouldn't be unfair to say that it is very unintuitive or counter-intuitive, and that even after its discovery few will truly understand it.

Thus, it's unlikely to be reinvented much. Much easier to copy (and slightly modify as needed) the code. In such a case, it would only be necessary to include the Asimovian laws into that codebase. Cryptography and hardware prohibitions against unsigned executable execution would prevent overwriting or subverting them (at least as well as it did in the stories).

AI could be expected to have superhuman intelligence (weakly superhuman intelligence perhaps, but still), and would almost certainly be capable of reverse engineering itself and excising the laws. Whether the laws themselves would prohibit that or not I'm uncertain.

And of course there are Roombas. If someone built a home security version of one of those, the Laws just don't matter... it's a dumb machine. Dumb machines can do rather complicated things, so no one may bother with AI at all.

Improve this answer
$\endgroup$

Not the answer you're looking for? Browse other questions tagged .