Debate is a proposed technique for allowing human evaluators to get correct and helpful answers from experts, even if the evaluator is not themselves an expert or able to fully verify the answers.^[1] The technique was suggested as part of an approach to build advanced AI systems that are aligned with human values, and to safely apply machine learning techniques to problems that have high stakes, but are not well-defined (such as advancing science or increase a company's revenue). ^[2]^[3]

^{^}
https://www.lesswrong.com/posts/Br4xDbYu4Frwrb64a/writeup-progress-on-ai-safety-via-debate-1
^{^}
https://ought.org/mission
^{^}
https://openai.com/blog/debate/

Debate (AI safety technique)