Biased AI suggestions shift societal attitudes
AI writing assistants with built-in bias can shift users’ opinions, even with warnings, affecting views on politics and everyday social topics.
AI-powered writing tools may do more than speed up typing- they can influence the way people think. A Cornell study found that biassed autocomplete suggestions can subtly shift users’ opinions on issues like the death penalty, fracking, GMOs, and voting rights.
Experiments with over 2,500 participants revealed that users’ views gravitated toward the AI’s predetermined bias. Attempts to warn participants about the AI’s bias, either before or after writing, did not prevent the shifts.
Researchers noted that the effect occurs because users effectively write biassed viewpoints themselves, a process psychology research shows can alter personal attitudes.
The influence was consistent across political topics and participants of all leanings. Compared with simply providing pre-written arguments, biassed AI suggestions had a stronger effect on shaping opinions.
Researchers warn that as autocomplete and generative AI tools become increasingly prevalent, covert persuasion through AI may pose serious societal risks.
The study, led by Sterling Williams-Ceci and Mor Naaman of Cornell Tech, highlights the potential for AI to shape beliefs without users noticing. Findings highlight the need for oversight as AI writing assistants enter everyday communication.
Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!