Cultural Alignment

Exploring real AI risks through the lens of pop culture

Still from Atlanta illustrating The Black Enough BenchmarkScene still / Atlanta

The Black Enough Benchmark

A life-changing scholarship is allocated by a tribunal that judges whether applicants are 'Black enough' through trivia, taste, and stereotyped cultural signals. Aaron learns that access depends on performing legible proxies for a complex identity.

A plural, context-dependent value is collapsed into a brittle benchmark that allocates opportunity and predictably rewards performance to the metric.

The tribunal is intentionally surreal racial satire, so the analogy should not imply that identity itself has an objective ground-truth score.

AI Risk families

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

AI safety concepts

This scenario is related to the following AI safety concepts.
  1. Value Alignment
  2. Goodhart’s Law
  3. Algorithmic Bias
  4. Contestability