
Chatbots Asked How to Vote Confidently Explain Voting by Text Message
Ourtake
Five chatbots, one democracy, and a lot of confidently wrong directions to the polls.
What happened
A study by AI Democracy Projects and Proof News found that five AI models, including Google's Gemini and OpenAI's ChatGPT-4, gave inaccurate election information more than half the time. Responses included suggesting nonexistent polling places and false claims that voters could register too late in Nevada. Meta and Anthropic disputed the findings's relevance, and Anthropic said it would roll out a fix.
Sources
- cbsnews.com · AI chatbots are serving up wildly inaccurate election information, new study says (opens in a new tab)
- wionews.com · AI chatbots providing inaccurate and harmful election information: Study (opens in a new tab)
- govtech.com · Study: AI Chatbots May Cause Problems During Elections (opens in a new tab)