AI Chatbots Spread Election Misinformation in Over Half of Responses, Risking Voter Confusion
-
Popular AI chatbots provided false information about basic election facts 50% of the time when tested by researchers. This could lead to voter confusion.
-
The researchers asked the chatbots questions voters might have, like polling locations and registration rules, but often got inaccurate or misleading answers.
-
All 5 major chatbots tested - Claude, Gemini, GPT-4, LLaMA 2, and Mixtral - gave bad information frequently.
-
The incorrect answers could frustrate voters and make voting seem overly complex, potentially reducing turnout.
-
The study authors warn the prevalence of false information from AI systems threatens to "erode the truth" around the 2024 elections.