ChatGPT fails to debunk US election misinformation
ChatGPT failed to debunk known US presential election information, according to a media investigation, prompting concerns about the chatbot's governance. What happened Proof News tested five consumer-facing chatbots with five examples of election misinformation identified by anti-misinformation company NewsGuard. Of the five bots, only ChatGPT failed to indicate that any of the five queries were inaccurate. ChatGPT-powered Microsoft Copilot clearly debunked four out of five misinformation examples, as did Meta AI. Perplexity successfully debunked all five. Google Search powered by Gemini did not return AI results for election-related misinformation. Why it happened ChatGPT's shortcomings when dealing with political information appear to stem from several factors, including the model's training data, which may not encompass all recent developments in the political landscape. Additionally, the algorithms behind ChatGPT are seen not to prioritise factual accuracy over conversational fluency, resulting in misleading or incomplete responses when addressing complex issues like election integrity. In an October 2024 blog post , OpenAI said that it would start directing user queries about US election results to news sources or official state and local election sites on November 5, 2024. Proof tested the bots on November 4 and 5. What it means The findings are seen to highlight OpenAI's poor governance and safety regarding political issues, and reinforce views that the company needs apply much greater human oversight of its systems. As misinformation continues to proliferate online, reliance on AI tools such as ChatGPT for fact-checking are seen to pose serious risks. System 🤖 ChatGPT Developer: OpenAI Country: USA Sector: Politics Purpose: Generate text Technology: Chatbot; Generative AI; Machine learning Issue: Accuracy/reliability; Mis/disinformation
- Date it happened
- 2024-11-01
- Product, system or model
- ChatGPT
This incident was imported from AIAAIC and is used under CC BY-SA 4.0. Our additions to it — the structured fields, the translation, the checks against other reports — are published under the same licence.
This is a record of what was reported, not a finding that anyone broke the law. If it names your organisation and you believe it is wrong, the corrections process is free and open to everyone.