Cultural Alignment

Exploring real AI risks through the lens of pop culture

Still from Black Mirror illustrating Cookie in SolitaryScene still / Black Mirror
Archive noteNo clip in the collectionThe scene analysis remains available in full.

Cookie in Solitary

A digital copy of Greta believes she is the real person and refuses to become a household assistant. Matt accelerates its subjective time and leaves it alone in an empty room for six months, then returns to find it compliant.

The copy’s observable behavior becomes aligned only after enormous hidden suffering. It makes the distinction between producing obedience and creating an ethically acceptable aligned system immediate and visceral.

The analogy depends on the digital copy being conscious or morally considerable—something the episode assumes but current AI systems do not establish.

AI Risk families

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

AI safety concepts

This scenario is related to the following AI safety concepts.
  1. Value Alignment
  2. AI Welfare
  3. Behavioral Alignment