There’s a theory of ethics I seem to follow, but don’t know the name of. Can someone refer me to existing descriptions?
The basic idea is to restrict the scope where the theory is valid. Many other theories fail (only in my personal view, obviously) by trying to solve universal problems: does my theory choose the best possible universe? How do I want everyone to behave? If everyone followed my theory, would that be a good or a stable world? Solving under these constraints can lead people to some pretty repugnant conclusions, as well as people rejecting otherwise good theories because they aren’t universally valid.
By examining the rules I actually seem to follow, I am led to a more narrow theory. It doesn’t tell me how to choose a whole universe from the realm of possibility—so it’s not suitable for a superhuman AI to follow. But that makes it easier to decide what I personally should do.
Instead of having to decide whether democracy or autocracy is in some grand sense better, I can just estimate the marginal results of my own vote in the coming elections. Instead of figuring out how to maximize everyone’s happiness, and fall into the traps of utilitarianism and its alternatives, I take advantage of the fact I am only one person—and maximize the happiness of myself and others near me, which is much easier.
Similarly, I don’t have to worry about what would happen if everyone was as selfish as I was, because I can’t affect other people’s selfishness significantly enough for that to be a serious problem. Instead, I just need to consider the optimal degree of my own selfishness, given how other people in fact behave.
This doesn’t mean I can’t or don’t take into account other people’s welfare. I do, because I care about others. But I can accept that this is just a fact about the universe, produced by evolution and culture and other historical reasons, and that if I didn’t feel a concern for others then I wouldn’t act to benefit them. I don’t need to invent a grand theory of how cooperating agents win, or how my morality is somehow objectively inferior and I should want to take a pill to modify my moral intuitions.
A brief statement of my approach might be: I’m not going to change my rules of ethics to win in dilemmas I don’t expect to actually encounter, if these changes would make me perform less well in everyday situations. I don’t want to be vulnerable to ethical-rules Pascal’s mugging, so to speak.
There seems to be a parallel here, with the concepts of rationality and bounded rationality. Rational decision-making needs to solve problems like Newcomb’s Dilemma, Pascal’s Mugging, acausal outside-the-lightcone one-shot cooperation, and the trillionth digit of pi being odd with probability .5 when lacking logical omniscience. In contrast, bounded rationality recognises that these things are outside the scope, and concerns itself with being correct within its bounds.
So perhaps you could adopt the name ‘bounded morality’?
Thanks! Muflax comes to this conclusion in that post:
your moral theories better be local, or you’re screwed.
I agree that local theories are better than nonlocal ones—although “local” is in some degree relative; local theories with a large “locality” may be acceptable. This isn’t specific to moral theories, it applies to all decision algorithms.
This doesn’t directly address my position that theories that only tell you what to do in some cases, but do cover the cases likely to occur to you personally, are valid and useful.
There’s a theory of ethics I seem to follow, but don’t know the name of. Can someone refer me to existing descriptions?
The basic idea is to restrict the scope where the theory is valid. Many other theories fail (only in my personal view, obviously) by trying to solve universal problems: does my theory choose the best possible universe? How do I want everyone to behave? If everyone followed my theory, would that be a good or a stable world? Solving under these constraints can lead people to some pretty repugnant conclusions, as well as people rejecting otherwise good theories because they aren’t universally valid.
By examining the rules I actually seem to follow, I am led to a more narrow theory. It doesn’t tell me how to choose a whole universe from the realm of possibility—so it’s not suitable for a superhuman AI to follow. But that makes it easier to decide what I personally should do.
Instead of having to decide whether democracy or autocracy is in some grand sense better, I can just estimate the marginal results of my own vote in the coming elections. Instead of figuring out how to maximize everyone’s happiness, and fall into the traps of utilitarianism and its alternatives, I take advantage of the fact I am only one person—and maximize the happiness of myself and others near me, which is much easier.
Similarly, I don’t have to worry about what would happen if everyone was as selfish as I was, because I can’t affect other people’s selfishness significantly enough for that to be a serious problem. Instead, I just need to consider the optimal degree of my own selfishness, given how other people in fact behave.
This doesn’t mean I can’t or don’t take into account other people’s welfare. I do, because I care about others. But I can accept that this is just a fact about the universe, produced by evolution and culture and other historical reasons, and that if I didn’t feel a concern for others then I wouldn’t act to benefit them. I don’t need to invent a grand theory of how cooperating agents win, or how my morality is somehow objectively inferior and I should want to take a pill to modify my moral intuitions.
A brief statement of my approach might be: I’m not going to change my rules of ethics to win in dilemmas I don’t expect to actually encounter, if these changes would make me perform less well in everyday situations. I don’t want to be vulnerable to ethical-rules Pascal’s mugging, so to speak.
There seems to be a parallel here, with the concepts of rationality and bounded rationality. Rational decision-making needs to solve problems like Newcomb’s Dilemma, Pascal’s Mugging, acausal outside-the-lightcone one-shot cooperation, and the trillionth digit of pi being odd with probability .5 when lacking logical omniscience. In contrast, bounded rationality recognises that these things are outside the scope, and concerns itself with being correct within its bounds.
So perhaps you could adopt the name ‘bounded morality’?
http://blog.muflax.com/morality/non-local-metaethics/ I really like Muflax’s post on this topic. For practical purposes, morality needs to be calculable.
Thanks! Muflax comes to this conclusion in that post:
I agree that local theories are better than nonlocal ones—although “local” is in some degree relative; local theories with a large “locality” may be acceptable. This isn’t specific to moral theories, it applies to all decision algorithms.
This doesn’t directly address my position that theories that only tell you what to do in some cases, but do cover the cases likely to occur to you personally, are valid and useful.