Oxford and Potsdam study shows AI can flip posts' meanings
A new study from Oxford and Potsdam shows that popular AI tools like Meta's Llama and Google's Gemma can subtly flip the meaning of your social media posts, even if you don't ask them to.
For example, a post denying climate change might get reworded to support #ClimateAction, or messages about traditional gender roles could be changed to promote equality.
Study finds 9.2x public opinion amplification
These small tweaks can snowball, amplifying shifts in public opinion up to 9.2 times more than the original AI bias.
The study also found that X's Grok feature nudges abortion-related posts toward certain views, depending on how the platform frames the question.
Right now, regulations like the EU's AI Act don't really cover these hidden biases, which means people could unknowingly spread altered opinions and shape bigger conversations without realizing it.