msellers@mntgfx.mentor.com (Mike Sellers) (05/26/88)
I've always like Asimov's Laws of Robotics; I suspect that they will remain firmly entrenched in the design and engineering of ambulatory AI systems for some time (will they ever be obsolete?). I have some comments on the proposed variations Barry proposed, however... In article <31738@linus.UUCP>, bwk@mbunix.UUCP writes: > > I propose the following variation on Asimov: > > I. A robot may not harm a human or other sentient being, > or by inaction permit one to come to harm. > > II. A robot may respond to requests from human beings, ^^^ > or other sentient beings, unless this conflicts with > the First Law. Shouldn't "may" be "must" here, to be imperitive? Otherwise it would seem to be up to the robot's discretion whether to respond to the human's requests. > > III. A robot may act to protect its own existence, unless this > conflicts with the First Law. Or the Second Law. Otherwise people could tell robots to destruct themselves and the robot would obey. Of course, if the destruction was necessary to keep a human from harm, it would obey to be in keeping with the First Law. > > IV. A robot may act to expand its powers of observation and > cognition, and may enlarge its knowledge base without limit. Unless such expansion conflicts with the First, Second, or Third (?) Laws. This is a worthy addition, but unless constrained by the other rules it contains within it the seeds of Prometheus (from the movie "Demon Seed" -- ick, what a title :-) or Colossus (from "The Forbin Project"). The last thing we want is a robot that learns and cogitates at the expense of humans. > Can anyone propose a further refinement to the above? > > --Barry Kort In addition to what I've said above, I think that all references to generic "sentient beings" should be removed. Either this is too narrow in meaning, providing only for humans (which are already explicitly stated in the Law), or it is too general, easily encompassing *artificial* sentient beings, i.e. robots. This is precisely what the Laws were designed to prevent. I like the intent, and hopefully some way of engendering general pacifism and deference to humans, animals, and to some degree other robots can be found. Perhaps a Fifth Law: "A robot may not harm another robot, or by inaction permit one to come to harm, unless such action or inaction would conflict with the First Law." Note that by only limiting the conflict resolution to the First Law, a robot could not respond to a human's request to harm another robot unless by not responding a human would come to harm (V takes precedence over II), and a robot might well sacrifice its existence for that of another (V takes precedence over III). Of course, this wouldn't necessarily prevent a military commander from convincing a warbot that destroying a bunch of other warbots was necessary to keep some humans from harm... I guess this is what is done with human armies nowadays anyway. Comments? -- Mike Sellers ...!tektronix!sequent!mntgfx!msellers Mentor Graphics Corp., EPAD msellers@mntgfx.MENTOR.COM "Hi. So, can any of you make animal noises?" -- the first thing Francis Coppola ever said to me