Artificial Morality

Essay by Bruce Sterling: “This is an essay about lists of moral principles for the creators of Artificial Intelligence. I collect these lists, and I have to confess that I find them funny.

Nobody but AI mavens would ever tiptoe up to the notion of creating godlike cyber-entities that are much smarter than people. I hasten to assure you — I take that weird threat seriously. If we could wipe out the planet with nuclear physics back in the late 1940s, there must be plenty of other, novel ways to get that done.

What I find comical is a programmer’s approach to morality — the urge to carefully type out some moral code before raising unholy hell. Many professions other than programming have stern ethical standards: lawyers and doctors, for instance. Are lawyers and doctors evil? It depends. If a government is politically corrupt, then a nation’s lawyers don’t escape that moral stain. If a health system is slaughtering the population with mis-prescribed painkillers, then doctors can’t look good, either.

So if AI goes south, for whatever reason, programmers are just bound to look culpable and sinister. Careful lists of moral principles will not avert that moral judgment, no matter how many earnest efforts they make to avoid bias, to build and test for safety, to provide feedback for user accountability, to design for privacy, to consider the colossal abuse potential, and to eschew any direct involvement in AI weapons, AI spyware, and AI violations of human rights.

I’m not upset by moral judgments in well-intentioned manifestos, but it is an odd act of otherworldly hubris. Imagine if car engineers claimed they could build cars fit for all genders and races, test cars for safety-first, mirror-plate the car windows and the license plates for privacy, and make sure that military tanks and halftracks and James Bond’s Aston-Martin spy-mobile were never built at all. Who would put up with that presumptuous behavior? Not a soul, not even programmers.

In the hermetic world of AI ethics, it’s a given that self-driven cars will kill fewer people than we humans do. Why believe that? There’s no evidence for it. It’s merely a cranky aspiration. Life is cheap on traffic-choked American roads — that social bargain is already a hundred years old. If self-driven vehicles doubled the road-fatality rate, and yet cut shipping costs by 90 percent, of course those cars would be deployed.

I’m not a cynic about morality per se, but everybody’s got some. The military, the spies, the secret police and organized crime, they all have their moral codes. Military AI flourishes worldwide, and so does cyberwar AI, and AI police-state repression….(More)”.