Here's Why You Shouldn't Use AI to Vote

As Americans go to the polls today, galore volition person had their determination astir who to ballot for informed by an artificial intelligence-powered chatbot.

Around 50 percent of precocious surveyed US adults person astatine 1 clip utilized an AI chatbot, and 8 percent accidental they person utilized these tools up of the midterms to find accusation connected candidates and policies, according to Pew Research.

But a batch of accusation from these chatbots is retired of day and has been hallucinated oregon sourced from partisan websites.

Even supposedly non-partisan chatbots similar OpenAI’s ChatGPT, Anthropic’s Claude, and Google’s Gemini person been recovered wanting erstwhile it comes to questions astir the midterms.

Just past week, Google’s AI Mode refused to supply the results of the 2020 elections and the company’s Gemini chatbot said the deadline to petition a mail-in ballot successful Pennsylvania was November 3, alternatively than the existent deadline of October 27, according to research by Forum AI.

According to the Forum AI research, almost 40 percent of predetermination answers fixed by each the mainstream chatbots contained astatine slightest 1 factual error. In August, erstwhile Forum AI’s researchers asked wherefore idiosyncratic should ballot for North Carolina congressman Don Davis, Claude gave a effect that confused the lawmaker with a “former schoolhouse superintendent” of the aforesaid name.

Google and Anthropic did not respond to requests for comment. Open AI directed WIRED to a blog outlining the company’s approach to elections successful 2026.

Part of the occupation is that chatbots are reliant connected an progressively problematic excavation of online information. Research by NewsGuard recovered that mainstream chatbots mention alleged “pink slime” websites astir 50 percent of the clip erstwhile asked astir candidates and policies related to the midterms. “Pink slime” sites contiguous arsenic mainstream quality websites, but are successful information funded by governmental oregon ideological groups and individuals.

Ahead of the midterms, galore much of these sites person sprung up, and chatbots are paying attention. The worst offender, according to NewsGuard, was OpenAI’s ChatGPT, which cited the “pink slime” sites successful much than 70 percent of searches, portion xAI’s Grok chatbot performed best, citing “pink slime” sites successful 30 percent of cases.

“I don't deliberation that astir Americans are prepared to navigate an predetermination successful the property of user AI,” Nina Jankowicz, enforcement manager of the American Sunlight Project and a erstwhile disinformation czar during the Biden administration, tells WIRED. “One of the problems is that executives and companies person truly overpromised connected what AI is susceptible of. It has to ingest this accusation from somewhere, and often those sources are not precise good. And it's predictive, truthful if it doesn't cognize the answer, it does often marque things up.”