A new study reveals that even minor AI-driven changes to social media content can gradually shape broader public opinion and online discourse.
Scientists from the University of Oxford and Hasso Plattner Institute have proved that the use of AI-based systems for creating, modifying or providing context for social media posts could influence public opinion by bringing their biases into online discussion. The researchers found that even small AI-driven changes could result in gradual shifts in broader public opinion due to their accumulation during millions of interactions.
It was revealed that large language models (LLMs) consistently changed the direction of social media posts on controversial issues even when asked to maintain their original sense. Computer simulations using the real-life social networking data from X and Facebook revealed that such minor changes could propagate throughout the online network over time.
It turned out that biases in AI were rather consistent between different systems. Multiple models tended to nudge posts in similar directions – for example, towards supporting positions on gun control, legalisation of marijuana use, and feminism, but against views associated with atheism and the death penalty. Moreover, it was revealed by the researchers that platform implementation can have a significant impact on the direction and magnitude of AI-generated influence.
To study this problem, scientists asked various LLMs from different providers to improve human-created posts on controversial issues and analyse whether the AI-generated versions systematically changed the positions expressed in the original posts.
“Our research points to AI-mediated communication as a new and more subtle way of influencing opinions—one the law has yet to catch up with—and offers food for thought about who, or what, is shaping public discourse,” says senior author Sandra Wachter, professor of technology and regulation at the Oxford Internet Institute, University of Oxford.
