Schelling-Goodhart coordination problems

Sustainable beliefs and accurate beliefs, and the utility of each

March 13, 2024 — January 11, 2023

economics
game theory
incentive mechanisms
institutions
machine learning
optimization
religion
statistics
utility

Content warning:

This is about triggering content and as such the content it contains may be triggering. Worked examples might included, veganism, gender or racial contention.

Notes on designing lines in the sand of the political landscape. cf adversarial categorization. This is most salient when it arises in culture wars, so the examples here will be contentious.

Figure 1

1 Incoming

Start with the Emotional stability of veganism.

This next essay touches upon content which is, with unusually high likelihood, triggering.

M. Taylor Saotome-Westlake’s essay, Book Review: Charles Murray’s Human Diversity. He analyses a coordination-on-belief problem, from an economics-of-coordination angle, with the particularly interesting example of taboos in discussing psychometrics:

And that’s where the blank slate doctrine absolutely shines—it’s the Schelling point for preventing group conflicts! (A Schelling point is a choice that’s salient as a focus for mutual expectations: what I think that you think that I think… &c. we’ll choose.) If you admit that there could be differences between groups, you open up the questions of in what exact traits and of what exact magnitudes, which people have an incentive to lie about to divert resources and power to their group by establishing unfair conventions and then misrepresenting those contingent bargaining equilibria as some “inevitable” natural order.

If you’re afraid of purported answers being used as a pretext for oppression, you might hope to make the question un-askable. Can’t oppress people on the basis of race if race doesn’t exist!

and

Notice how the “not allowing sex and race differences in psychological traits to appear on shared maps is the Schelling point for resistance to sex- and race-based oppression” actually gives us an explanation for why one might reasonably have a sense that there are dread doors that we must not open. Undermining the “everyone is Actually Equal” Schelling point could catalyse a preference cascade—a slide down the slippery slope to the next Schelling point, which might be a lot worse than the status quo on the “amount of rape and genocide” metric, even if it does slightly better on “estimating heritability coefficients.”

An interesting analysis of the processes at play.

Also interesting: This argument seems to hinge upon people coordinating on “is” beliefs, rather than “ought” beliefs. Why?