Salvato in:
Dettagli Bibliografici
Autori principali: Brown-Cohen, Jonah, Irving, Geoffrey, Piliouras, Georgios
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:https://arxiv.org/abs/2506.13609
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866915346819055616
author Brown-Cohen, Jonah
Irving, Geoffrey
Piliouras, Georgios
author_facet Brown-Cohen, Jonah
Irving, Geoffrey
Piliouras, Georgios
contents Training powerful AI systems to exhibit desired behaviors hinges on the ability to provide accurate human supervision on increasingly complex tasks. A promising approach to this problem is to amplify human judgement by leveraging the power of two competing AIs in a debate about the correct solution to a given problem. Prior theoretical work has provided a complexity-theoretic formalization of AI debate, and posed the problem of designing protocols for AI debate that guarantee the correctness of human judgements for as complex a class of problems as possible. Recursive debates, in which debaters decompose a complex problem into simpler subproblems, hold promise for growing the class of problems that can be accurately judged in a debate. However, existing protocols for recursive debate run into the obfuscated arguments problem: a dishonest debater can use a computationally efficient strategy that forces an honest opponent to solve a computationally intractable problem to win. We mitigate this problem with a new recursive debate protocol that, under certain stability assumptions, ensures that an honest debater can win with a strategy requiring computational efficiency comparable to their opponent.
format Preprint
id arxiv_https___arxiv_org_abs_2506_13609
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Avoiding Obfuscation with Prover-Estimator Debate
Brown-Cohen, Jonah
Irving, Geoffrey
Piliouras, Georgios
Artificial Intelligence
Computational Complexity
Data Structures and Algorithms
Training powerful AI systems to exhibit desired behaviors hinges on the ability to provide accurate human supervision on increasingly complex tasks. A promising approach to this problem is to amplify human judgement by leveraging the power of two competing AIs in a debate about the correct solution to a given problem. Prior theoretical work has provided a complexity-theoretic formalization of AI debate, and posed the problem of designing protocols for AI debate that guarantee the correctness of human judgements for as complex a class of problems as possible. Recursive debates, in which debaters decompose a complex problem into simpler subproblems, hold promise for growing the class of problems that can be accurately judged in a debate. However, existing protocols for recursive debate run into the obfuscated arguments problem: a dishonest debater can use a computationally efficient strategy that forces an honest opponent to solve a computationally intractable problem to win. We mitigate this problem with a new recursive debate protocol that, under certain stability assumptions, ensures that an honest debater can win with a strategy requiring computational efficiency comparable to their opponent.
title Avoiding Obfuscation with Prover-Estimator Debate
topic Artificial Intelligence
Computational Complexity
Data Structures and Algorithms
url https://arxiv.org/abs/2506.13609