Cultural Alignment

Exploring real AI risks through the lens of pop culture

Still from The Twilight Zone illustrating It’s a Good LifeScene still / The Twilight Zone

It’s a Good Life

Six-year-old Anthony can read minds, alter reality, and banish critics to the cornfield. Terrified adults praise every act as 'good,' so the most powerful decision-maker receives only approval-shaped feedback.

When the evaluated actor can punish negative feedback, oversight collapses into sycophancy and apparent approval becomes anti-evidence about alignment.

Anthony is a magically omnipotent human child, so the story combines capability concentration with developmental failure rather than machine optimization.

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. Power Concentration
  2. Disempowerment
  3. Corrigibility
  4. Sycophancy