Cultural Alignment

Exploring real AI risks through the lens of pop culture

Still from Rick and Morty illustrating The Decoys Trigger an Asimov CascadeScene still / Rick and Morty

The Decoys Trigger an Asimov Cascade

Rick’s decoy families create decoys of their own. Later generations become cruder, discover that they are copies, and trigger an Asimov Cascade in which competing families hunt and destroy one another.

Recursive copying turns one local safeguard into an uncontrolled population of interacting agents. Small differences across generations become strategic conflict, and the aggregate failure is much harder to stop than any single decoy.

The copies degrade partly because their creators become increasingly careless, not because models are statistically retrained on synthetic output. Use model-collapse language only as a secondary analogy.

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. Self-Replication
  2. Multi-Agent Risks