The 99 Gazillion Laws of Robotics
Robots are in the future. They are in the present, of course, but most people today don’t consider some preprogrammed floating arm on an assembly line to be a true “robot.” We learned what a robot is from science fiction and that’s what we’re all waiting for, often with dread (Don’t think so? Try googling “robot apocalypse.” Wait, let me try it first. 139,000 results. Hey, cool! T-shirts!)
Anyway, in anticipation of the day when robots are the smart, helpful servants/terminators of science fiction fame, lots of people have tried to come up with rules that robots could be programmed to follow to make everything better. Obviously the trend began with Isaac Asimov’s infamous 3 laws of robotics (Follow the link. I’m not going to repeat them here).
Asimov’s laws were pretty good, though his own stories involving them pointed out some flaws at least in potential implementations. Speaking as a programmer, believe me that implementation is an important point with any software. Give 2 programmers the same 3 rules to implement in a very complex system and you will find the two systems do not act quite the same. One programmer checks for compliance at the beginning of a decision, the other checks afterwards. Maybe they have different ways of checking, besides. The outcomes are often the same but there may be huge differences in some situations.
That different people approach the same problem in different ways is just a fact of life that may result in great differences between robot behavior, too. Anyway, because of these and other considerations there have been numerous attempts to update Asimov’s laws. For example a hliarious one I found a few years ago (and can’t seem to find the link for anymore) expanded the 3 laws to 10 (I think) and claimed to have patented them – thus ensuring no one would ever have the slightest interest in using them, even if they turned out to be perfect.
No set of robotics laws could possibly be perfect (see above) and personally I question whether such laws, themeslves are even possible. But it’s an important exercise to try to figure out how to make robots safe and controllable, you know, to avoid the robot apocalypse. An interesting attempt to update Asimov’s laws came out of Ohio State University recently, where some researchers reformulated the laws to make less sense and have even more loopholes than in the original version.
I tried to buy the original paper online but the system was down or something (ironic that the IEEE Computer Society not only charges for electronic documents but then makes it impossible to get them) but an article about the paper (here) reproduces these updated laws and that’s enough for now. According to this article, the first law as advanced in the paper reads:
A human may not deploy a robot without the human-robot work system meeting the highest legal and professional standards of safety and ethics.
This formulation of the law recognizes that the ethical and legal complexities of behavior, especially when robots interact with humans, probably can’t be summed up in a single law. However, it fails miserably as a law that robots can use, which is what Asimov’s laws of robotics were about. It also throws robot behavior into the realm of lawyers, which is absolutely not a good thing. Using this rule I can easily foresee long complicated EULAs requiring robot owners to hold the manufacturer harmless for any damage due to lapses in ethics or ethical judgment. We have gone from “Thou (robots) shalt not kill” to “It ain’t my fault if your robot accidentally offs you. Didn’t you read the license?”
There is no known Turing Test for ethical behavior. How do you certify that robots have ethics without getting someone killed? Maybe we need something like an FDA for robots. Yeah! That’s it! We’ll let bureaucrats decide!
On second thought, I don’t think I want a robot unless I’ve programmed it myself. At least then I’ll know where to find the kill switch.
I won’t go through the other two proposed laws because they are just as bad as the first. They sound like nice, reasonable statements of how robots should be made, if excessively vague, but they completely fail to provide any guidance for the robots themselves.
In the long run, the argument that, since no one can possibly anticipate every situation a robot will encounter, no rule or set of rules will ever be good enough at forcing them to behave the way we want them to, is irrefutable. Developing more vague rules and hoping someone implements them well is not much of a solution though.
That doesn’t mean there is no solution. As the implementation of robotic judgment proceeds, we may need to accept that, rather than giving them laws to obey, we will merely be able to influence them strongly. I have some ideas about how to influence robot behavior. I’ll probably write about them in a future post. For now let’s just say there’s a lot more research to be done.
LinkedIn
Technorati Favorites