---
aliases:
  en:
  - corrigible
  ja:
  - 修正可能性
concept_id: corrigibility
preferred:
  en: corrigibility
  ja: 訂正可能性
relations:
  relations:
    broader:
    - ai_alignment
    prerequisite:
    - instrumental_convergence
    related:
    - oversight
    - power_seeking
status: active
---

# corrigibility

A property of an AI system that allows and does not strategically resist appropriate correction, modification, shutdown, or other interventions by its operators, despite incentives that ordinary goal pursuit might create to prevent such interventions.

## broader

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

## prerequisite

- [instrumental convergence](../terminology/instrumental_convergence.txt)

## related

- [oversight](../terminology/oversight.txt)
- [power-seeking](../terminology/power_seeking.txt)

## Documents: about

- [CEV 型 Sovereign と訂正可能な AGI は異なる安全化戦略である](../documents/yudkowsky.cev-sovereign-vs-corrigible-agi.txt)
- [訂正可能性は一般的な目標追求から自然には生じない](../documents/yudkowsky.corrigibility-is-anti-natural.txt)

## Documents: mentions

- [自己保存は最終目標でなくてもシャットダウンを妨げうる](../documents/akira.instrumental-self-preservation-defeats-shutdown.txt)
- [能力が訓練分布の外へ急速に汎化する局面で、アラインメント特性は同程度には汎化しない可能性がある](../documents/soares.sharp-left-turn.txt)

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