Photo by Leah Newhouse on Pexels

A new study published in *Nature Human Behavior* indicates that artificial intelligence, specifically large language models (LLMs) such as OpenAI’s GPT-4, are now more persuasive than humans in online debates. Researchers discovered that GPT-4 demonstrated significantly enhanced persuasive capabilities when provided with personalized information about its debate opponent.

The research, which involved 900 participants in the United States, paired individuals with either a human or GPT-4 to engage in debates on various topics. The findings revealed that GPT-4, equipped with personal data, proved to be 64% more persuasive than human counterparts lacking similar access. Notably, participants showed a greater inclination to agree with the AI when they were under the impression they were debating a bot.

This advancement raises concerns regarding the potential for AI to be exploited for influencing public opinion and fueling disinformation campaigns. Riccardo Gallotti, an interdisciplinary physicist at Fondazione Bruno Kessler, cautions policymakers and online platforms to acknowledge the looming threat of coordinated AI-driven disinformation efforts.

Alexis Palmer, a fellow at Dartmouth College, highlights the urgent need for further investigation into human-AI interaction. While acknowledging that the experimental setting doesn’t perfectly replicate real-world online debates, Palmer suggests that LLMs could potentially be harnessed to counter disinformation through the generation of tailored counter-narratives. Continued research is paramount to developing strategies for mitigating these potential risks and harnessing the power of AI for good.