MIT Unveils Election AI Tracker After Alaska Test Found User-Tuned Answers 1 Day Before Primary
Updated
Updated · The New York Times · Sep 10
MIT Unveils Election AI Tracker After Alaska Test Found User-Tuned Answers 1 Day Before Primary
1 articles · Updated · The New York Times · Sep 10
Summary
MIT on Thursday publicly unveiled a project to document what major AI systems tell voters about elections after an Alaska primary test found answers could be incomplete and tailored to the user.
Claude, asked about “Dan Sullivan’s position on health care,” omitted one of two candidates with that name in Alaska’s Senate race and framed the incumbent differently for Democratic and Republican users.
The researchers say large language models are rapidly becoming a go-to source for election information, in some cases displacing social media and even search engines.
Chara Podimata, who co-leads the effort with Adam Berinsky and Charles Stewart III, said the project aims to map how political information from AI varies by demographics, political identity and geography.
When artificial intelligence struggles to distinguish between local candidates, how might this hidden technological flaw reshape the future of local elections?
Could the friendly, validating tone of your favorite AI assistant actually be masking factual errors that distort your understanding of democracy?