Can Artificial Intelligence Be Ethical?
project-syndicate.org
project-syndicate.org
... but then ex-hypothesis, shouldn't a human-plus intelligent machine be teaching us ethics, rather than the other way around?
To suppose otherwise is to assume there's something about ethics that's antithetical to machine intelligence, whereas (without a specific reason to the contrary) it might be reasonably held to be not unlike any other social/psychological knowledge domain which should be mastered to human+ level by the sort of AGI being talked about.
On the other hand, if by "teaching ethics to a machine" the article really means "ensuring the machine behaves in a way that we regard as meeting our ethical norms" (which isn't the same thing at all, obviously) then all bets are off: the machine's intelligence exceeding ours strongly implies (to me, anyway) our inability to predict its behaviour, so I'd agree that strong guarantees are at least problematic in this respect.
Cognition and affect/motivation get conflated a lot in AI discussions. Motivations don't magically arise from intelligence. They are not just acquired in upbringing either. Affect is a separate and innate system that requires its own explicit hardwiring. There is some imprinting, conditioning and other fine tuning with experiential feedback, but the basic drives and reward rules (and morals) have to be there from the beginning. Otherwise the person, artificial or not, will be catatonic.
But is hard to think in defining ethics in an universal timeless manner. I think the main issue is the unpredictability of the future.
The closest thing I can think of is an hypotetical matrix (or bunch of equations) that maximizes the outcome in the evolutionary path for each individual (having in mind all their interactions between one another). The kind of calculus that once was attributed to be only solvable by The Great Matematician.
1. You can still do unethical things that are too small or too big for laws to handle. It's perfectly legal to cheat on your partner, or to mock a bullied child, or to form a political party that takes power and repeals laws you don't like.
2. Even for the things that are covered by laws, a superhuman AI would be able to find bizarre loopholes that obey the letter of the law but would be considered unethical by most people.
In which case, we might define "act according to the law" as something like "act in a way which, if it were described in a court of law, would not lead to any prosecution". Predicting the outcome of such a hypothetical court hearing requires anticipating, thinking through and weighing up the possible arguments on both sides. That's basically a description of ethics, so we're back where we started :(
The law defines what is permitted, not what is ethical.
Frederic Bastiat wrote a short book "The Law" where he elaborates on this and the purpose of law.
Thomas Jefferson was also very aware about questioning inmoral laws.
Criminal law always tries to formalize the average ethical landscape of citizens. But there are lots of ways how you can try to do it. And success varies greatly. Anyhow it's the best thing anybody has.