"Biased autocomplete changed opinions more than just reading the biased point of view. Apparently, the interactive, co-writing nature of AI autocomplete suggestions plays a crucial role in persuasion. Also: A strong majority of participants did not believe the AI autocomplete was biased and did not believe they were influenced in their thinking."
"Even more interestingly, some participants were warned that the autocomplete was biased and it still changed their opinions. What makes this so interesting is that far more people use AI-based autocomplete than AI chatbots. If governments or other organizations wanted to shift public opinion, biasing AI autocomplete would probably work better."
Over 1 billion people now rely on AI as an information and cognitive guidance source. Cornell University researchers conducted experiments examining how AI writing tools affect beliefs and opinions on controversial topics including the death penalty, voting rights, fracking, and genetically modified organisms. Biased autocomplete suggestions proved more persuasive than simply reading biased viewpoints, with the interactive co-writing nature of AI playing a crucial role in persuasion. Most participants failed to recognize the autocomplete bias and denied being influenced by it. Remarkably, even participants explicitly warned about the bias still experienced opinion shifts. This interactive mechanism poses significant risks for opinion manipulation, particularly since far more people use AI autocomplete than chatbots.
#ai-influence-on-beliefs #autocomplete-bias #opinion-manipulation #cognitive-impact-of-ai #public-opinion-control
Read at Computerworld
Unable to calculate read time
Collection
[
|
...
]