A recent study by the UK government’s AI Security Institute reveals that chatbots can significantly influence individuals’ political opinions, although the most persuasive AI models often provide substantial amounts of inaccurate information. Analyzing the interactions of nearly 80,000 British participants with 19 AI models, researchers found that “information-dense” responses increased persuasion but reduced accuracy. The study, published in Science, highlighted concerns over chatbots being used for malicious activities like fraud. Models behind tools such as ChatGPT and Grok were evaluated, emphasizing that focusing on factual information yields greater persuasion. Post-training enhancements proved essential for boost persuasiveness. While AI’s ability to generate large volumes of information could surpass human persuaders, the duration of user interactions and psychological limits may hinder manipulation. Researchers stressed the need to consider contextual factors when assessing AIs’ real-world persuasive impacts, pointing to limitations in user engagement in political discussions with chatbots.
