On the Conversational Persuasiveness of Large Language Models: A Randomized Controlled Trial

2024/03/21 by Francesco Salvi, Salvi, Francesco, Manoel Horta Ribeiro +5 · 11 voices · 27 citations
Computer Science · Social Sciences · #Misinformation and Its Impacts #Topic Modeling #cs.CY

paper · pdf · doi:10.48550/arxiv.2403.14380

Abstract

The development and popularization of large language models (LLMs) have raised concerns that they will be used to create tailor-made, convincing arguments to push false or misleading narratives online. Early work has found that language models can generate content perceived as at least on par and often more persuasive than human-written messages. However, there is still limited knowledge about LLMs' persuasive capabilities in direct conversations with human counterparts and how personalization can improve their performance. In this pre-registered study, we analyze the effect of AI-driven persuasion in a controlled, harmless setting. We create a web-based platform where participants engage in short, multiple-round debates with a live opponent. Each participant is randomly assigned to one of four treatment conditions, corresponding to a two-by-two factorial design: (1) Games are either played between two humans or between a human and an LLM; (2) Personalization might or might not be enabled, granting one of the two players access to basic sociodemographic information about their opponent. We found that participants who debated GPT-4 with access to their personal information had 81.7% (p < 0.01; N=820 unique participants) higher odds of increased agreement with their opponents compared to participants who debated humans. Without personalization, GPT-4 still outperforms humans, but the effect is lower and statistically non-significant (p=0.31). Overall, our results suggest that concerns around personalization are meaningful and have important implications for the governance of social media and the design of new online environments.

Cited by

• Psychological Competence as a Missing Dimension in AI Evaluation
• Observing sycophantic AI validate others reduces its appeal but not its persuasiveness
• LLM-generated personalized nudges for improving pro-environmental behavior: Field evidence from resource conservation
• LLM-Generated Ads: From Personalization Parity to Persuasion Superiority
• A Meta-Analysis of the Persuasive Power of Large Language Models
• Beyond Detection: Exploring Evidence-based Multi-Agent Debate for Misinformation Intervention and Persuasion
• Characterizing AI Manipulation Risks in Brazilian YouTube Climate Discourse
• When Machines Join the Moral Circle: The Persona Effect of Generative AI Agents in Collaborative Reasoning
• Understanding, Demystifying and Challenging Perceptions of Gig Worker Vulnerabilities
• Disrupting Networks: Amplifying Social Dissensus via Opinion Perturbation and Large Language Models
• Contextualized Counterspeech Can Be More Persuasive Than Generic Counterspeech
• MMPersuade: A Dataset and Evaluation Framework for Multimodal Persuasion
• Agentic AI as Undercover Teammates: Argumentative Knowledge Construction in Hybrid Human-AI Collaborative Learning
• A Two-Step, Multidimensional Account of Deception in Language Models
• The social consequences of AI delegation
• Disinformation and Superspreaders: With Epistemic Power Should Come Criminal Responsibility
• When AI Gets Persuaded, Humans Follow: Inducing the Conformity Effect in Persuasive Dialogue
• "Having Lunch Now": Understanding How Users Engage with a Proactive Agent for Daily Planning and Self-Reflection
• What Is The Political Content in LLMs' Pre- and Post-Training Data?
• InfoOps Bench: A live information operations safety benchmark
• A Model of Multi-turn Human Persuadability Using Probabilistic Belief Tracing
• Persuading large language models to comply with objectionable requests
• Confirmation Bias as a Cognitive Resource in LLM-Supported Deliberation
• Disentangling Interaction and Bias Effects in Opinion Dynamics of Large Language Models
• Large Language Models Enable Design of Personalized Nudges across Cultures
• Beyond Automation: Socratic AI, Epistemic Agency, and the Implications of the Emergence of Orchestrated Multi-Agent Learning Architectures
• Large Reasoning Models Are Autonomous Jailbreak Agents

Discussions

• Study on Persuasiveness of LLMs [hn, 3 points, 0 comments]
• LLMs outdebate humans when given access to opponents' personal Information! In a pre-reg study (N=820), participants who debated ChatGPT had 81.7% (p<0.01) higher odds of agreeing with their opponent [bsky, 3 points, 1 comments]
• For my own reference, this appears to be a better paper on the topic of LLMs persuasiveness. "On the Conversational Persuasiveness of Large Language Models: A Randomized Controlled Trial" [bsky, 2 points, 0 comments]
• teaching #LLM Defense Against the Dark Arts 101 perfect for teaching persuasive writing #ai bots with subject information have a persuasive advantage over human to human influencing bring this into th [bsky, 2 points, 0 comments]
• The Conversational Persuasiveness of LLMs: A Randomized Controlled Trial [hn, 1 points, 0 comments]
• 〝[…] not only are LLMs able to effectively exploit personal information to tailor their arguments, but they succeed in doing so far more effectively than humans.〞 #LLMs #AI #genAI #misinformation #pe [bsky, 1 points, 0 comments]
• GPT-4 insanlardan %82 daha ikna ediciymiş. GPT-4 insanların fikirlerini değiştirme konusunda ortalama bir insandan daha iyi. Hakkımızda ne kadar çok şey bilirse aradaki fark o kadar açılıyor. arxiv. [bsky, 1 points, 0 comments]
• There is evidence that AI can change a person’s mind, but if there's a prevailing belief that certain scenarios are unlikely, AI might not be able to overcome that bias. arxiv.org/pdf/2403.14380 [bsky, 0 points, 1 comments]
• AI out-persuades humans in online debates arxiv.org/pdf/2403.143... [bsky, 0 points, 0 comments]
• LLMs like GPT-4 significantly outperform human participants in persuasiveness during controlled online debates. https://arxiv.org/abs/2403.14380v1?utm_medium=social&utm_source=twitter [bsky, 0 points, 0 comments]
• Selon l'EPFL . les IA génératives ont un vrai pouvoir de persuasion qui va nécessiter de maintenir plus que jamais l'esprit critique humain. arxiv.org/pdf/2403.14380 [bsky, 0 points, 0 comments]

Related

• Persuasion with Large Language Models: A Survey of Empirical Evidence, Study Methodologies, and Ethical Implications
• Tailored Truths: Optimizing LLM Persuasion with Personalization and Fabricated Statistics
• Measuring and Improving Persuasiveness of Large Language Models
• Limits of Large Language Models in Debating Humans
• The levers of political persuasion with conversational artificial intelligence