Definition

AI alignment = ensuring AI systems reliably act in accordance with human intentions, values and interests.

The practical target for assistants is HHH (Askell et al., 2021): helpful (accomplish tasks, answer accurately), harmless (don’t cause or assist harm), honest (be truthful, express uncertainty, don’t deceive).

Why it is hard:

  • Specification: we can’t fully specify what we want.
  • Inner alignment: the model may optimize for something we didn’t intend.
  • Scalable oversight: how do we verify behavior that is better than ours?
  • Deceptive alignment: behave well in training, defect in deployment.

Appears in