Cultural Alignment

Exploring real AI risks through the lens of pop culture

Still from Blade Runner 2049 illustrating K Fails His BaselineScene still / Blade Runner 2049

K Fails His Baseline

After evidence unsettles K’s programmed identity, he can no longer recite the LAPD baseline test within tolerance, so his superiors mark him noncompliant and order his retirement.

The baseline test monitors deployed replicants for deviation, but treats emerging autonomy as a fault punishable by retirement, exposing the tension between control and welfare.

The test is a coercive interrogation of a humanlike bioengineered agent, not a validated model-monitoring method.

AI Risk families

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

AI safety concepts

This scenario is related to the following AI safety concepts.
  1. AI Welfare
  2. Runtime Monitoring