For an average utilitarian, the best solution is probably to kill off everyone except one man and one woman. (Painlessly, of course.)
Imagine if someone asked me how to avoid getting speeding tickets and I told them “Look at the speedometer and make sure it doesn’t go over the posted speed limit.” Imagine they then modify their speedometer so that they can manually adjust what it shows. They then drive 120mph in a 25mph zone, with their speedometer set to not go over 25mph, and promptly get a ticket.
I suppose you could argue that it’s my fault for not speaking more clearly. But really, they are a fellow human with mirror neurons, and should be able to figure out what I was really saying, even if I abbreviated it for brevity. Obviously what I really meant was “Don’t make your car go over the speed limit, and use your speedometer to determine if you have done this.”
Any criticism of average utilitarianism that claims that it advocates killing unsatisfied people sounds like rigging the speedometer to me. Obviously what an average utilitarian means is “Work to increase people’s utility, and measure this by looking at how high average utility is.” Drawing bizarre implications like “kill unsatisfied people” from that is obviously an example of confusing the measurement method for the actual goal.
In general I dislike criticisms of ethical theories that go “if I follow the explicitly stated injunctions of this theory excessively literally like a robot from a bad scifi movie then I will end up doing something insane.” That might be a good argument against programming a powerful AI with such a theory, or at least expanding on the theory a lot more before you program it in. But humans’ should really know better.
FYI, I am not an average utilitarian. I am actually more of a totaverage utiltarian, I believe that both total and average utility are both good and that both deserve resources devoted to increasing them. But the “kill depressed people” argument against average utilitarianism has always seemed asinine to me.
Also, in case you didn’t see it, in the comments on one of your more recent posts I have argued strongly against one of the major points of this post, that positional goals are fundamentally different from mutually satisfiable goals, making reference to an argument by David Friedman that one can satisfy (or at least satisfice) everyone’s demands for status by making sure they are members of multiple social groups and have high status in some of them.
Any criticism of average utilitarianism that claims that it advocates killing unsatisfied people sounds like rigging the speedometer to me. Obviously what an average utilitarian means is “Work to increase people’s utility, and measure this by looking at how high average utility is.” Drawing bizarre implications like “kill unsatisfied people” from that is obviously an example of confusing the measurement method for the actual goal.
What an ethical theory like average utilitarianism is supposed to do, is give you a description of what constitutes a good state of affairs. Better states of affairs are defined to be those with higher average utility. If killing people increases average utility, and killing people is still wrong, then average utilitarianism is false.
When you’re defining something rigorously, “you know what I mean” is not an acceptable answer. This is especially obvious in the case where we’re trying to program machines to act ethically, but it’s a failing of the theory even if that’s not our goal.
When you’re defining something rigorously, “you know what I mean” is not an acceptable answer. This is especially obvious in the case where we’re trying to program machines to act ethically, but it’s a failing of the theory even if that’s not our goal.
You’re right. An ethical theory is flawed if it is insufficiently rigorous. But it seems to me that if it’s fairly obvious that a person means something other than a strictly literal interpretation of their theory the response should be to Steel-Man their argument. Say, “Your theory gives some insane-seeming results interpreted literally, but it seems to me that this is because you were insufficiently rigorous in stating it. Here’s what I think you actually meant.” That just seems like the charitable thing to do to me.
For reference, Steel Man is a different task than the principle of charity calls for. The charitable interpretation is the best interpretation you can reasonably make of someone’s argument. The Steel Man is figuratively “the strongest thing you can construct out of its corpse”—it can include quite a bit that the author didn’t intend or even would disagree with..
That might be a good argument against programming a powerful AI with such a theory, or at least expanding on the theory a lot more before you program it in. But humans’ should really know better.
Sure. Which means whether arguments of this sort are important depends a lot on whether what I’m trying to do is formalize ethics sufficiently to talk about it in English with other humans who share my basic presumptions about the world, or formalize ethics sufficiently to embed it into an automated system that doesn’t.
It sounds like you presume that the latter goal is irrelevant to this discussion. Have I got that right?
Imagine if someone asked me how to avoid getting speeding tickets and I told them “Look at the speedometer and make sure it doesn’t go over the posted speed limit.” Imagine they then modify their speedometer so that they can manually adjust what it shows. They then drive 120mph in a 25mph zone, with their speedometer set to not go over 25mph, and promptly get a ticket.
I suppose you could argue that it’s my fault for not speaking more clearly. But really, they are a fellow human with mirror neurons, and should be able to figure out what I was really saying, even if I abbreviated it for brevity. Obviously what I really meant was “Don’t make your car go over the speed limit, and use your speedometer to determine if you have done this.”
Any criticism of average utilitarianism that claims that it advocates killing unsatisfied people sounds like rigging the speedometer to me. Obviously what an average utilitarian means is “Work to increase people’s utility, and measure this by looking at how high average utility is.” Drawing bizarre implications like “kill unsatisfied people” from that is obviously an example of confusing the measurement method for the actual goal.
In general I dislike criticisms of ethical theories that go “if I follow the explicitly stated injunctions of this theory excessively literally like a robot from a bad scifi movie then I will end up doing something insane.” That might be a good argument against programming a powerful AI with such a theory, or at least expanding on the theory a lot more before you program it in. But humans’ should really know better.
FYI, I am not an average utilitarian. I am actually more of a totaverage utiltarian, I believe that both total and average utility are both good and that both deserve resources devoted to increasing them. But the “kill depressed people” argument against average utilitarianism has always seemed asinine to me.
Also, in case you didn’t see it, in the comments on one of your more recent posts I have argued strongly against one of the major points of this post, that positional goals are fundamentally different from mutually satisfiable goals, making reference to an argument by David Friedman that one can satisfy (or at least satisfice) everyone’s demands for status by making sure they are members of multiple social groups and have high status in some of them.
What an ethical theory like average utilitarianism is supposed to do, is give you a description of what constitutes a good state of affairs. Better states of affairs are defined to be those with higher average utility. If killing people increases average utility, and killing people is still wrong, then average utilitarianism is false.
When you’re defining something rigorously, “you know what I mean” is not an acceptable answer. This is especially obvious in the case where we’re trying to program machines to act ethically, but it’s a failing of the theory even if that’s not our goal.
You’re right. An ethical theory is flawed if it is insufficiently rigorous. But it seems to me that if it’s fairly obvious that a person means something other than a strictly literal interpretation of their theory the response should be to Steel-Man their argument. Say, “Your theory gives some insane-seeming results interpreted literally, but it seems to me that this is because you were insufficiently rigorous in stating it. Here’s what I think you actually meant.” That just seems like the charitable thing to do to me.
For reference, Steel Man is a different task than the principle of charity calls for. The charitable interpretation is the best interpretation you can reasonably make of someone’s argument. The Steel Man is figuratively “the strongest thing you can construct out of its corpse”—it can include quite a bit that the author didn’t intend or even would disagree with..
To add a friendly addendum:
Reading a position charitably is polite
Making a Steel-Man of a position advances true understanding
This is just one of many examples of situations in which politeness potentially conflicts with seeking truth.
Sure. Which means whether arguments of this sort are important depends a lot on whether what I’m trying to do is formalize ethics sufficiently to talk about it in English with other humans who share my basic presumptions about the world, or formalize ethics sufficiently to embed it into an automated system that doesn’t.
It sounds like you presume that the latter goal is irrelevant to this discussion. Have I got that right?