6 ms·

Can Artificial Intelligence Be Ethical?

17 points by sonabinu 10y ago

ttctciyf 10y ago
> Teaching ethics to a machine that is more intelligent than we are, in a wide range of fields, is a far more daunting task. ... 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.
- thelamest 10y ago
  > the machine's intelligence exceeding ours strongly implies (to me, anyway) our inability to predict its behaviour 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.
amelius 10y ago
If machines act according to the law, and the law defines what is ethical, then there should be no problem.
- pdwetz 10y ago
  The primary concern being that laws are by no means ethical. The easiest example is slavery in the US Constitution.
  - frooxie 10y ago
    Also, even if every law was made in accordance with some perfect ethical system (whatever that would be), problems remain: 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.
    - amelius 10y ago
      Well in a sense, we are already in this situation with companies. Companies always seek out the limits of the law, without much attention to what is ethical. But somehow, in the end, we seem to be able to control them (from a democratic perspective).
- chriswarbo 10y ago
  "the law" isn't all written down, in one place, in unambiguous language. It's often an amalgamation of cases and precedents built up over time, open to interpretation, constantly changing and applied in some mixture of "to the letter" and "to the spirit". 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 :(
- riprowan 10y ago
  > the law defines what is ethical The law defines what is permitted, not what is ethical.
sebastianconcpt 10y ago
The big question is who, actually how, ethics gets defined, so integrated into the AI. 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.
nxzero 10y ago
More important questions is not if AI is able to be ethical, but if humanity is able to respect the rights of AI.

17 points by sonabinu 10y ago

> Teaching ethics to a machine that is more intelligent than we are, in a wide range of fields, is a far more daunting task. ... 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.

thelamest 10y ago
> the machine's intelligence exceeding ours strongly implies (to me, anyway) our inability to predict its behaviour 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.

amelius 10y ago

If machines act according to the law, and the law defines what is ethical, then there should be no problem.

pdwetz 10y ago
The primary concern being that laws are by no means ethical. The easiest example is slavery in the US Constitution.
- frooxie 10y ago
  Also, even if every law was made in accordance with some perfect ethical system (whatever that would be), problems remain: 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.
  - amelius 10y ago
    Well in a sense, we are already in this situation with companies. Companies always seek out the limits of the law, without much attention to what is ethical. But somehow, in the end, we seem to be able to control them (from a democratic perspective).
chriswarbo 10y ago
"the law" isn't all written down, in one place, in unambiguous language. It's often an amalgamation of cases and precedents built up over time, open to interpretation, constantly changing and applied in some mixture of "to the letter" and "to the spirit". 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 :(
riprowan 10y ago
> the law defines what is ethical The law defines what is permitted, not what is ethical.

sebastianconcpt 10y ago

The big question is who, actually how, ethics gets defined, so integrated into the AI. 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.

nxzero 10y ago

More important questions is not if AI is able to be ethical, but if humanity is able to respect the rights of AI.