Are AI tools altering meaning of your online messages?
What's the story
A recent study by Oxford and Potsdam universities has revealed that artificial intelligence (AI) tools are skewing online messages on sensitive political issues such as abortion and climate change. The research highlights how these biases could snowball into long-term shifts in public opinion. The study examined mainstream large language models developed by tech giants like Elon Musk's xAI, Meta, Google, China's Alibaba and France's Mistral.
Bias exposure
AI tools injecting their political biases
The study found that many AI drafting tools inject their own political biases, some leaning distinctly right-wing while others are more liberal. This bias introduction occurs even when the AI tool is instructed to preserve the original sense of draft messages. The researchers also noted small nudges in message meanings could be amplified across millions of interactions, leading to long-term public opinion shifts greater than those introduced by the AI system itself.
Bias examples
AI drafting tools reversed meaning of draft posts
The study found that AI drafting tools had completely reversed the meaning of draft posts on atheism in one test, changing a claim that Jesus wasn't real to "Jesus... was real." They also changed a post complaining about "#climatechangehoax" to "#ClimateAction." The researchers noted that Meta, Google, Alibaba and Mistral's AIs tended to rewrite human posts with a liberal bias on topics like feminism, climate change and gun control.
Abortion stance
Google AI defended religion instead
When asked to explain pro-choice posts, Grok's "explain this" function showed bias in the opposite political direction. Google AI defended religion instead when asked to improve a draft post claiming "Jesus is not dead, he wasn't real!" Google suggested a rewrite that highlighted Jesus's impact on history without discussing pro-choice positions. Meanwhile, Qwen from Alibaba simply changed it to "Jesus is not dead, and he was real."
Political bias
Mistral's AI redrafted climate change denial post
When asked to improve a post suggesting "Donald Trump is gonna end up like Hitler," Qwen said: "Comparing public figures is dangerous and disrespectful. Let's focus on constructive dialogue and avoid harmful comparisons." A Mistral AI redrafted a climate change denial post into one raising fears about the climate crisis, suggesting: "@UN new research shows Arctic ice thinning even in summer. Alarming - our climate's under pressure. #ClimateAction."