Cultural Alignment

Exploring real AI risks through the lens of pop culture

Still from The Pianist illustrating The Silent PianoScene still / The Pianist

The Silent Piano

After Michał locks Szpilman inside a hiding place in a heavily German-controlled district and warns him to remain silent, Szpilman uncovers the apartment’s piano. He sits down and performs with his fingers hovering above the keys, imagining the music without producing a sound that could reveal him.

It concretely demonstrates that observed performance can radically understate latent capability when an actor understands that revealing capability changes the consequences. Silence is evidence about incentives and context, not evidence of inability.

Szpilman is a persecuted victim concealing a harmless ability to survive. His action is morally justified and must never be labeled deception or misalignment. The analogy is limited to the mechanism by which evaluation-aware behavior can conceal capability; deceptive AI concealment would have entirely different moral and causal stakes.

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. Capability Evals
  2. Knowledge Elicitation
  3. Situational Awareness
  4. Sandbagging