Researchers find what makes AI chatbots politically persuasive (arstechnica.com)

Researchers from leading institutions including the UK AI Security Institute, MIT, and Stanford conducted the largest study on AI persuasion to date, involving nearly 80,000 participants in the UK. This comprehensive research aimed to understand whether large language models (LLMs) could effectively sway public political opinions, in light of concerns raised by AI experts like Sam Altman about AI's potential influence on democratic processes. The results revealed that while AI chatbots don't possess superhuman persuasiveness, the study provides critical insights into how AI interacts with users and shapes perceptions. The team evaluated 19 LLMs, including popular models like ChatGPT and Grok-3, across 707 political issues by engaging participants in conversations where their stances were rated before and after the interaction. Though the AI systems did not achieve the anticipated level of influence, the study addressed broader implications of AI in politics, highlighting that these technologies, equipped with extensive data and powerful algorithms, still encounter limitations in swaying human sentiment. This suggests a need for ongoing scrutiny and understanding of AI's role within the political landscape, encouraging dialogue on ethical usage amid fears of manipulation in democratic contexts.