Rational Culture Choice

I’ve been on a culture kick the last few months, and I know many of you don’t get it, as I didn’t until recently. Here is why it matters so much: you inherit most of your values from your culture, via a process that looks pretty arbitrary, and is now broken, giving you maladaptive values. That should startle you, and make you ask: how could I adopt more reliable values? That is what I want to think about here.

Standard decision theory is our best account of ideal choices.This theory assumes a set of possible states of everything, and says to pick the act with the highest expected utility, which is a sum over states of state utility times act-conditional state probability. Beliefs are about the world, saying the chance of each state, while utilities (= preferences, values) are about a person, saying how much that person likes each state. Utilities never change, but beliefs update via info using Bayes’ rule. (Further complexities are re committing to limiting future choices, but I’ll ignore that here.)

We often take sets of actual and hypothetical choices, and try to find a closest fit standard decision theory model (SDTM). These almost always fit with noise, suggesting that we could improve our choices by cutting this noise, i.e., by making our choices more closely approximate a best fit model of them. And in fact, large literatures in rationality, statistics, operations research, computer science, etc., show many concrete ways to use such models to make more consistent choices.

Most of these methods make less noisy probabilities, but some make less noisy utilities. For example, we sometimes ask people re many random hypothetical choices, and fit utility estimates to those. Can something like this help for cultural values?

A key startling fact about culture is that different co-existing cultures teach different values, the values that each culture teaches change a lot over time, and most of these differences in space and time can’t be accounted for by differing context, like climate, disease, lifespans, or wealth. So if we made a best fit SDTM for a sample of choices for a particular place and time time, the best fit utilities from that SDTM would be different from those for other places and times. But as utilities aren’t supposed to change unless context changes in a related meaningful way, this makes culture-set values seem a poor approximation to SDTM.

However, we can often explain choices that look like different values as instead learning how to achieve constant underlying values. For example, If we track an individual’s ice cream choices, we might find that while last year they tended to pick vanilla, this year they tend to pick mango sherbet. While one way to describe this is via changing preferences, another way is as constant underlying preferences mixed with learning about new flavors. Maybe this person had just never tried mango sherbet before, and upon trying it they learned how it better matched their constant underlying dessert preferences. 

So can we do this for cultural changes? One approach would say that each person’s main underlying preference is really just to conform to their cultures norms, and rank high on their culture’s status markers. So that explains why they act like they share the values taught them by their particular culture, and change those values as their culture changes its values. That seems a rational account of an individual who is just trying to get along in their culture, and not trying to lead or change it, or switch between cultures.

What about someone willing and able to switch cultures? All else equal, they might seek to join the most adaptive cultures, the ones most likely to succeed in the long run. (Today they might try to join the Amish or Haredim.) Once moved, they’d try to adopt and track local values, to conform to local norms and achieve the local status markers.

What could be a rational account of a group of people large enough to have a chance of changing their culture? Such a group might see humanity as slowly learning which cultures work best, and so try to get their culture to change to adopt features of the most adaptive cultures on Earth so far. They might even try to estimate the features produced by future selection effects, and try to jump to those future features on purpose without waiting for selection to induce them. If, as I’ve suggested, our main world culture is suffering from cultural drift, they might try to jump to an adaptive culture that no longer suffers this problem. 

I expect there are even more possible ways to see typical culture-conforming behaviors, that appear on the surface to involve value changes, as actual expressing learning to achieve constant underlying preferences.