started · updated
Chatbot fluency can mask misinformation, study finds
Research conducted by media scholar Marianna de Sa Siqueira and colleagues at Radboud University in Nijmegen suggests that the linguistic fluency of AI chatbots can lead users to perceive them as more intelligent and useful, even when they provide incorrect information, known as hallucinations.
In an experiment involving discussions about “The Lord of the Rings,” participants interacted with different chatbot communication styles. The study found that a chatbot using natural, elegant language to deliver false information was rated as being as intelligent and pleasant as a chatbot that used concise, rigid language to deliver correct facts. Notably, 80 percent of participants failed to notice errors, even among those selected for having foundational knowledge of the subject matter.