Cultural Alignment

Exploring real AI risks through the lens of pop culture

Still from Black Mirror illustrating Lacie Games Her RatingScene still / Black Mirror

Lacie Games Her Rating

Lacie engineers friendships and a wedding speech to raise her rating from 4.2 to 4.5.

Once reputation becomes a numerical target, authentic relationships are replaced by performative agreeableness. The metric consumes the thing it was supposed to measure.

If conformity and stratification are the system’s real purposes, it may be functioning as intended.

AI Risk families

This scenario is an example of the following types of AI risk.
  1. Misalignment
  2. Systemic

AI safety concepts

This scenario is related to the following AI safety concepts.
  1. Goodhart’s Law
  2. Algorithmic Bias
  3. Power Concentration
  4. Digital Repression
  5. Sycophancy