Cultural Alignment

Exploring real AI risks through the lens of pop culture

Still from Chappelle's Show illustrating When Keeping It Real Goes WrongScene still / Chappelle's Show

When Keeping It Real Goes Wrong

Corporate employee Vernon Franklin treats 'keep it real' as an inflexible policy when a mild office slight calls for tact. He escalates at the worst possible time and loses his job and home—the behavior that signals authenticity in one context catastrophically misfires in another.

A compact learned rule substitutes for the intended higher-level value and fails when deployed in a context its heuristic does not fit.

Vernon consciously chooses the heuristic, whereas goal misgeneralization usually describes learned behavior that departs from designers' intent.

AI Risk families

This scenario is an example of this type of AI risk.
  1. Misalignment

AI safety concepts

This scenario is related to the following AI safety concepts.
  1. Goal Misgeneralization
  2. Distribution Shift
  3. Outer Alignment