Abstraction Worsens Drift

My Ph.D. was in formal political theory, and my job is tenured professor of economics. And the primary theory of choice in both of these areas is optimization. More specifically, game theory built on standard decision theory. Yes, these fields include many ways to approximately optimize, and often consider choice heuristics and tendencies from psychology. But optimization is still the main framework.

Recently I’ve been focused on cultural evolution, wherein choice is modeled quite differently. Here they assume people simply inherit values, norms, choices and heuristics from their cultural ancestors, choosing such ancestors according to a few simple known “biases”. Then natural selection applies to the individuals and communities who have made such choices.

To study human behavior, I want to merge these frameworks. And one simple way to do that is to pick an abstraction border level that separates the two. More detailed choices are made via optimization to implement heuristics, obey norms, and achieve subgoals that are set most abstractly by cultural evolution. For example, planning a route from A to B consists of searching for a path between them that minimizes a culture-chosen cost goal, prefers culture-chosen tech, avoids violating culture-set norms, and follows culture-inherited heuristic subpaths when available.

These abstract border goals, which we implement via optimization but do not choose via optimization, can vary with context and culture. And it seems that a big trend of the modern era has been to move this border to higher levels of abstraction. For example, we are less willing to follow rituals where we do not see rationales in terms of deeper goals. And we are replacing religions that ask us to accept specific claims with weak evidential support, with ideologies that make fewer factual claims and embrace more abstractly described goals.

Having our goals be specified at more abstract levels makes sense in a world that is changing more rapidly. More specific goals and heuristics might go stale and out of date faster than we could inherit new ones.

Having more abstract goals also puts a higher premium on intelligence, which helps us better reason abstractly. And a rise in the status of intelligence is also a big mark of the modern world. Together with credible signs of intelligence, like vocabulary and education. And together with much faster actual rates of change.

While higher abstraction can better deal with rapid environmental change, it has a problem of being harder to reliably copy. We often copy abstractions by inferring them from specific behaviors of others, and then we each make somewhat different inferences from the particular behaviors we observe. So norms and values that are culturally inherited do so with more noise, and so are more subject to drift. Yes, such noise may cancel itself when many independently infer and copy diverse prior examples, but culture is often transmitted via more central canonical cultural icons like specific movies, songs, and big historical events.

Increasing abstraction of cultural inheritance thus seems to contribute to cultural drift. Yes, drift might be even worse if we had not gone more abstractly, due to the very rapid rates of change in our world. But even so, culture is plausibly changing in random non-adaptive ways due to our difficulties with inferring abstractions from the social models from whom we inherit our cultures.

Added 11p: Another implication of this mixed view is that once you dig into a person to a sufficient depth, their coherence will degrade a lot. Culturally inherited goals and habits are just going to be a lot less coherent with each other, especially in a world where the environment is changing fast and culture is drifting. Once you get past the culture abstraction level, don’t expect people to make that much sense.