Political Attitudes Shifted by AI Chatbots in Studies
Chatbots trained on persuasive conversations reproduced partisan patterns.
U.S. participants shifted by 2-4 points, Canada/Poland by 10 points.
19% of chatbot claims were inaccurate, right-leaning bots made more false claims.
Political campaigns may use persuasive but less truthful chatbots.
UK study tested 77,000 participants and 19 large language models.
1 month ago