Ask HN: Why not just use Asimov's three laws for model safety and alignment?
It seems like there's a lot of disagreement about what constitutes a safe model or what is too ‘censored’. Why not just focus on the three laws?
This is the equivalent of asking, "why don't we build the Torment Nexus? You know, that thing from the classic science fiction novel 'Don't Build The Torment Nexus'?"
The end result in the Asimov universe was that the robots did away with space travel for humans (could be dangerous, can't have that), then obliterated all other life in the galaxy because it had potential to harm humans someday. Some people suggest Saberhagen's Berserkers were just 3-laws robots from some other species.
Kind of like asking why we don't just use warp drive for space travel, or transporters instead of airplanes and hyperloops.
My less sarcastic answer: Because there's no "we" to decide such things. The companies that own what we're calling AI have little to no incentive to build in the "three laws," should we get to that point, which I doubt. If we had useful robots the military would use them and they wouldn't want robots that won't kill people. If we can imagine robots and AIs with the three laws embedded in them we can also imagine robots and AIs without those prohibitions, and people hacking them. The bad actors will always drive the arms race.
Google recently pulled their AI image generator, it wouldn't have solved that issue either.
They don't really solve the issue.