// ⛨ Safety, Evaluation & Security

AI Safety via Debate

Two AIs debate and a human judges, aiding oversight of hard questions.

AI 辩论 2018 Intermediate extended

IN PLAIN WORDS

Even when humans can't verify an answer directly, judging two AIs exposing each other's flaws can be more reliable — a key scalable-oversight proposal.

Used in: Scalable oversight research

TIMELINE · prerequisites → this → what builds on it

2018 AI Safety via Deb…2022 Reinforcement…

01Original paper

AI safety via debate

Geoffrey Irving, Paul Christiano, Dario Amodei · 2018

arXiv ↗ ✓ Verified · arXiv API

02Where it sits

Learn the left column first; the right column builds on this one.

Learn first

Reinforcement Learning from Human Feedback (RLHF)

AI Safety via Debate

Then you can learn

Nothing in the atlas builds on it yet

Left: prerequisites · right: builds on this · dashed: related. Full map →

03Learning resources

Official docs, explainers, courses and reference code. All links checked.

Books

Nathan Lambert：RLHF Book ↗EN

04Where the course teaches it

U93 Frontier Risk, Scalable Oversight & Governance Weak-to-strong generalisation, debate, Constitutional AI, risk frameworks and regulation (EU AI Act, NIST AI RMF). 6 h4 algorithms3 papers

For AI / data

Verification: verified, checked 2026-10-21.
JSON: /api/learn/algorithms.json (id: debate)