Cultural Alignment

Exploring real AI risks through the lens of pop culture

Still from Minority Report illustrating The Suppressed Minority ReportScene still / Minority Report

The Suppressed Minority Report

When one precog foresees a different outcome, the dissenting “minority report” is removed from the official record so Precrime can continue presenting its predictions as infallible. People are imprisoned for predicted acts without a meaningful way to inspect or contest the evidence.

Model disagreement and uncertainty are hidden precisely where decisions are irreversible. A forecast is converted into punishment while the institution suppresses the information needed for appeal or calibrated trust.

The predictors are psychic humans rather than statistical models, and the film’s predictions are far more precise than real-world risk scores.

AI Risk families

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

AI safety concepts

This scenario is related to the following AI safety concepts.
  1. Automation Bias
  2. Calibration
  3. Contestability
  4. Governance Failure
  5. Digital Repression