Cultural Alignment

Exploring real AI risks through the lens of pop culture

Still from Altered Carbon illustrating Dimi’s Double Sleeve

Dimi’s Double Sleeve

Dimi illegally copies himself into two sleeves. After Kovacs and Poe kill one body, the surviving copy treats it as the death of his “brother” and pursues revenge.

A copied agent can diverge through experience; deleting one instance does not remove the system, and identity and responsibility become ambiguous.

These are human consciousness copies in a fictional substrate, not trained AI models.

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. Self-Replication
  2. AI Welfare
  3. Governance Failure
  4. Multi-Agent Risks