Can a Hallucinating Model help in Reducing Human "Hallucination"?

TL;DR

This work investigates whether a hallucinating model can help reduce human 'hallucination' by evaluating LLMs against psychometric measures of unwarranted beliefs using the PEUBI inventory. It analyzes human and LLM belief formation through doxastic logic and dual-process theory, and proposes a framework for using LLMs as personalized misinformation debunkers based on cognitive dissonance and elaboration likelihood theory. Empirically, three LLMs (GPT-3.5, GPT-4, Gemini) outperform average humans on PEUBI and exhibit an 'unstable rationality' that is sensitive to language and context. The study suggests promising directions for debunking misinformation with guard-railed LLM-based agents, while acknowledging limitations and the need for further human-centered evaluation.

Abstract

The prevalence of unwarranted beliefs, spanning pseudoscience, logical fallacies, and conspiracy theories, presents substantial societal hurdles and the risk of disseminating misinformation. Utilizing established psychometric assessments, this study explores the capabilities of large language models (LLMs) vis-a-vis the average human in detecting prevalent logical pitfalls. We undertake a philosophical inquiry, juxtaposing the rationality of humans against that of LLMs. Furthermore, we propose methodologies for harnessing LLMs to counter misconceptions, drawing upon psychological models of persuasion such as cognitive dissonance theory and elaboration likelihood theory. Through this endeavor, we highlight the potential of LLMs as personalized misinformation debunking agents.