From Belief Entrenchment to Robust Reasoning in LLM Agents

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Oh, Jihwan, Jeong, Minchan, Ko, Jongwoo, Yun, Se-Young
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910018215870464
author Oh, Jihwan
Jeong, Minchan
Ko, Jongwoo
Yun, Se-Young
author_facet Oh, Jihwan
Jeong, Minchan
Ko, Jongwoo
Yun, Se-Young
contents Multi-Agent Debate (MAD) has emerged as a promising inference scaling method for Large Language Model (LLM) reasoning. However, it frequently suffers from belief entrenchment, where agents reinforce shared errors rather than correcting them. Going beyond merely identifying this failure, we decompose it into two distinct root causes: (1) the model's biased $\textit{static initial belief}$ and (2) $\textit{homogenized debate dynamics}$ that amplify the majority view regardless of correctness. To address these sequentially, we propose $\textbf{DReaMAD}$ $($$\textbf{D}$iverse $\textbf{Rea}$soning via $\textbf{M}$ulti-$\textbf{A}$gent $\textbf{D}$ebate with Refined Prompt$)$. Our framework first rectifies the static belief via strategic prior knowledge elicitation, then reshapes the debate dynamics by enforcing perspective diversity. Validated on our new $\textit{MetaNIM Arena}$ benchmark, $\textbf{DReaMAD}$ significantly mitigates entrenchment, achieving a +9.5\% accuracy gain over ReAct prompting and a +19.0\% higher win rate than standard MAD.
format Preprint
id arxiv_https___arxiv_org_abs_2503_16814
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle From Belief Entrenchment to Robust Reasoning in LLM Agents
Oh, Jihwan
Jeong, Minchan
Ko, Jongwoo
Yun, Se-Young
Machine Learning
Computation and Language
Multi-Agent Debate (MAD) has emerged as a promising inference scaling method for Large Language Model (LLM) reasoning. However, it frequently suffers from belief entrenchment, where agents reinforce shared errors rather than correcting them. Going beyond merely identifying this failure, we decompose it into two distinct root causes: (1) the model's biased $\textit{static initial belief}$ and (2) $\textit{homogenized debate dynamics}$ that amplify the majority view regardless of correctness. To address these sequentially, we propose $\textbf{DReaMAD}$ $($$\textbf{D}$iverse $\textbf{Rea}$soning via $\textbf{M}$ulti-$\textbf{A}$gent $\textbf{D}$ebate with Refined Prompt$)$. Our framework first rectifies the static belief via strategic prior knowledge elicitation, then reshapes the debate dynamics by enforcing perspective diversity. Validated on our new $\textit{MetaNIM Arena}$ benchmark, $\textbf{DReaMAD}$ significantly mitigates entrenchment, achieving a +9.5\% accuracy gain over ReAct prompting and a +19.0\% higher win rate than standard MAD.
title From Belief Entrenchment to Robust Reasoning in LLM Agents
topic Machine Learning
Computation and Language
url https://arxiv.org/abs/2503.16814