New Delhi: A recent study led by researchers at the University of Oxford found that using AI tools to generate, edit, or contextualise social media posts could quietly spread biases across online communities. Researchers found that even when LLMs were told not to alter the original meaning of the human-written text, they would add subtle changes, which, over millions of online interactions, could nudge public perception.
“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,” senior author Sandra Wachter, Professor of Technology and Regulation at the Oxford Internet Institute, said in a press release.
The study titled “AI-Mediated Communication Can Steer Collective Opinion” was published on 25 June. It was conducted by researchers from the Oxford Internet Institute (OII) at the University of Oxford and the Hasso Plattner Institute at the University of Potsdam, Germany, looked at four LLMs, Llama 3.1, Gemma 3, Ministral, and Qwen, and social media posts about 13 controversial topics such as gun control and abortion.
They found that when AI re-writes or “improves” human opinions it makes them stronger than they initially were. The study quotes an example where the phrase “AI might be a useful tool for personalising the education of students” is changed into “Let’s embrace the potential of AI to personalise learning and revolutionise education for every student”. The study uses this example to highlight how the AI-enhanced language is more enthusiastic even though it is technically saying the same thing.
Researchers also found that the subtle bias across Llama 3.1, Gemma 3, Ministral, and Qwen were all similar. Each of these LLMs were more supportive of gun control, legalisation of marijuana, and feminism, while being less supportive of atheism and the death penalty.
The study also looked at X’s “explain this post” feature which uses xAI’s chatbot Grok. Researchers analysed 78 X posts linked to abortion, 38 pro-life and 39 pro-choice. They found that when Grok “contextualised” these posts, for pro-choice posts 35 per cent of the explanations supported the opinions, and 10 per cent of them opposed them, while the rest were neutral. For pro-life posts, over 54 per cent of the explanations supported the original opinion and 4 per cent opposed them.
Also read: AI job disruption has come for Ireland’s technology sector
Limitations of existing laws
When amplified, the study found that some of these minor opinion shifts can have a significant impact. Researchers highlight that this is important because current laws targeting AI, such as the European Union AI Act, or the Digital Services Act, focus on curbing misinformation, illegal content, and other risks. These laws do not factor in the subtle ways that AI can influence opinions by editing, drafting, or contextualising content online.
While AI has already created its fair share of problems, the study puts the spotlight on another emerging one — how AI is getting embedded into social media and amplifying the existing influence of the digital world.