AI ALIGNMENT FORUM
AF

2669

Debate (AI safety technique)

Edited by Bird Concept, plex last updated 15th Jul 2022

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/

1

1

Posts tagged Debate (AI safety technique)

3

50Writeup: Progress on AI Safety via Debate

Beth Barnes, paulfchristiano

6y

15

3

28A guide to Iterated Amplification & Debate

5y

0

3

64Debate update: Obfuscated arguments problem

5y

15

2

35An alignment safety case sketch based on debate

Marie_DB, Jacob Pfau, Benjamin Hilton, Geoffrey Irving

6mo

15

2

24How should AI debate be judged?

abramdemski, paulfchristiano

5y

26

2

11AI Safety via Debate

8y

10

2

20Optimal play in human-judged Debate usually won't answer your question

5y

8

1

72An overview of 11 proposals for building safe advanced AI

5y

32

1

38My Overview of the AI Alignment Landscape: A Bird's Eye View

4y

4

2

58Imitative Generalisation (AKA 'Learning the Prior')

5y

13

1

34Why I'm excited about Debate

5y

12

2

32Why I’m not working on {debate, RRM, ELK, natural abstractions}

3y

16

1

23Three mental images from thinking about AGI debate & corrigibility

5y

32

1

21Looking for adversarial collaborators to test our Debate protocol

5y

5

2

25Reward button alignment

6mo

10

Load More (15/49)

Add Posts