---
aliases:
  en:
  - outer alignment problem
  - reward misspecification problem
concept_id: outer_alignment
preferred:
  en: outer alignment
  ja: アウターアラインメント
relations:
  relations:
    broader:
    - ai_alignment
    prerequisite:
    - ai_alignment
    - alignment_target
    - training_objective
    related:
    - reinforcement_learning_from_human_feedback
status: active
---

# outer alignment

The problem or property of specifying a training objective that adequately captures the designers' intended goal or broader alignment target, assuming the learned system then robustly pursues that specification. The training objective may itself be implemented through a loss, reward-derived return, preference feedback, or another criterion. The inner/outer distinction is useful but its exact boundary varies across authors and settings.

## broader

- [AI alignment](../terminology/ai_alignment.txt)

## prerequisite

- [AI alignment](../terminology/ai_alignment.txt)
- [alignment target](../terminology/alignment_target.txt)
- [training objective](../terminology/training_objective.txt)

## related

- [reinforcement learning from human feedback](../terminology/reinforcement_learning_from_human_feedback.txt)

## Documents: about

- [ベース目的関数とメサ目的関数は一致するとは限らない](../documents/hubinger_et_al.base-and-mesa-objectives-can-diverge.txt)

[Corpus index](../index.txt)