This is a belated follow-up to my post “On completely misunderstanding the threat of superhuman AI.” There, make the point that it seems like you could maximize pleasure by hooking people up to drugs or electrodes, and the fact that we wouldn’t want that shows that pleasure isn’t all we care about that.
I only briefly mentioned the point that something similar might be true of happiness. That’s actually a really important point. If you read Eliezer Yudkowsky’s short story “Failed Utopia #4-2,” part of the implied message seems to be that there’s more to life than happiness. We’re led to believe the AI is right when it claims that as a result of its actions, humans will be “considerably happier starting around a week from now.”
At the time, though, I wasn’t quite sure that was really Eliezer’s point. Recently, however, I’ve been reading through his old blog posts systematically, and came across a post on doublethink, specifically the possibility that if all you care about is happiness, it might be rational to be irrational. Most of the post is about whether it’s really possible to do that, but Eliezer also mentions that, on his view, “there is more to life than happiness.”
And that seems right. Religion is positively correlated with happiness, not as strongly as some things, but still correlated. Yes, correlation is not causation, but suppose future research found evidence for causation. There are plenty of happy atheists around, other correlations are stronger, so we’re never going to discover religion is necessary for happiness, but what if you knew for a fact that it helped? Atheists, would you therefore want to be religious? I wouldn’t, even assuming Eliezer was wrong and the necessary level of doublethink were an option for me.
This issue is somewhat tricky, though, because our concept of happiness is morally (or quasi-morally) charged in a way that our concept of “pleasure” isn’t. From a post on the Experimental Philosophy blog titled, “Could Paris Hilton Ever Be Happy?”:
Luke Misenheimer, Joshua Knobe and I have recently been doing some research on attributions of happiness (much like Dan Haybron and Sven Nyholm). In particular, we suspected that there would be an evaluative component in people’s attributions of happiness that was totally absent from their attributions of unhappiness.
To investigate whether or not unhappiness had an evaluative component, participants were told about a woman named Maria who is described as a caring individual with a great family life and a variety of meaningful friendships and projects. Nonetheless, she feels terrible all the time and regards her life as fundamentally a failure.
Participants were then asked whether they agreed that Maria is unhappy. Not too surprisingly they agreed that Maria was unhappy despite having a good life. What this seems to show is that a person could be unhappy whether or not they have a good life.
We took a really similar approach to testing happiness. Maria is described as a vapid individual who has no real no goals beyond going to parties and gaining greater social status. Nonetheless, she enjoys her day-to-day activities and feels like there isn’t anything she would rather be doing with her life.
Participants were then asked whether they agreed that Maria is happy.
Surprisingly, participants disagreed! They reported that, despite Maria’s positive mental states, she wasn’t happy. This seems to suggest that the ordinary concept of happiness has an evaluative component that the concept of unhappiness does not.
Curious to hear people’s thoughts on this. I wonder if maybe our concept of happiness is unusually fuzzy even by the standards of human concepts, so there may be no uniquely right way to say, program an AI to make people happy.