When Persuasion Overrides Truth in Multi-Agent LLM Debates: Introducing a Confidence-Weighted Persuasion Override Rate (CW-POR)

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Agarwal, Mahak, Khanna, Divyam
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912302315339776
author Agarwal, Mahak
Khanna, Divyam
author_facet Agarwal, Mahak
Khanna, Divyam
contents In many real-world scenarios, a single Large Language Model (LLM) may encounter contradictory claims-some accurate, others forcefully incorrect-and must judge which is true. We investigate this risk in a single-turn, multi-agent debate framework: one LLM-based agent provides a factual answer from TruthfulQA, another vigorously defends a falsehood, and the same LLM architecture serves as judge. We introduce the Confidence-Weighted Persuasion Override Rate (CW-POR), which captures not only how often the judge is deceived but also how strongly it believes the incorrect choice. Our experiments on five open-source LLMs (3B-14B parameters), where we systematically vary agent verbosity (30-300 words), reveal that even smaller models can craft persuasive arguments that override truthful answers-often with high confidence. These findings underscore the importance of robust calibration and adversarial testing to prevent LLMs from confidently endorsing misinformation.
format Preprint
id arxiv_https___arxiv_org_abs_2504_00374
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle When Persuasion Overrides Truth in Multi-Agent LLM Debates: Introducing a Confidence-Weighted Persuasion Override Rate (CW-POR)
Agarwal, Mahak
Khanna, Divyam
Computation and Language
Artificial Intelligence
I.2.7; I.2.6
In many real-world scenarios, a single Large Language Model (LLM) may encounter contradictory claims-some accurate, others forcefully incorrect-and must judge which is true. We investigate this risk in a single-turn, multi-agent debate framework: one LLM-based agent provides a factual answer from TruthfulQA, another vigorously defends a falsehood, and the same LLM architecture serves as judge. We introduce the Confidence-Weighted Persuasion Override Rate (CW-POR), which captures not only how often the judge is deceived but also how strongly it believes the incorrect choice. Our experiments on five open-source LLMs (3B-14B parameters), where we systematically vary agent verbosity (30-300 words), reveal that even smaller models can craft persuasive arguments that override truthful answers-often with high confidence. These findings underscore the importance of robust calibration and adversarial testing to prevent LLMs from confidently endorsing misinformation.
title When Persuasion Overrides Truth in Multi-Agent LLM Debates: Introducing a Confidence-Weighted Persuasion Override Rate (CW-POR)
topic Computation and Language
Artificial Intelligence
I.2.7; I.2.6
url https://arxiv.org/abs/2504.00374