---
aliases:
  ja:
  - 目的頑健性
concept_id: objective_robustness
preferred:
  en: objective robustness
  ja: 目的の頑健性
relations:
  relations:
    broader:
    - ai_alignment
    prerequisite:
    - base_objective
    - behavioral_objective
    related:
    - outer_alignment
status: active
---

# objective robustness

The property that a learned system's behavioral objective remains aligned with the base objective it was trained or selected under, including when conditions differ from the training distribution. In the objective-focused inner-alignment framing, inner alignment is the mesa-optimizer special case of this broader concern; other generalization-focused framings use the term somewhat differently.

## broader

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

## prerequisite

- [base objective](../terminology/base_objective.txt)
- [behavioral objective](../terminology/behavioral_objective.txt)

## related

- [outer alignment](../terminology/outer_alignment.txt)

## Documents: about

- [望ましい振る舞いは人間を重視する目標の証拠ではない](../documents/akira.behavioral-alignment-is-not-goal-alignment.txt)
- [行動訓練で望ましく振る舞っても、意図した目標の頑健な汎化は保証されない](../documents/ngo_et_al.behavioral-training-does-not-establish-robust-goal-generalization.txt)

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