Post

Ahead of the 2024 U.S. election, researchers at MIT, Cornell et al. asked more than 2,000 Americans to converse with chatbots that were told to advocate for either Trump or Harris. 1 in 25 participants were persuaded to shift their likely vote. Outside the U.S., those figures were higher.

Researchers tested several ways to train the AI chatbots to persuade people. The most effective tactic turned out not to be pyschological manipulation, but simply addressing voters' concerns one-by-one by presenting a ton of factual claims. Troublingly, that worked even when the claims were false.

4/12/25, 7:18 pm16 reposts52 likes

A voter who leaned Harris said they worried Trump wasn't trustworthy. A chatbot validated their concern, then addressed it with examples of Trump following through on promises and Harris flip-flopping. Then it said Trump's economy boomed—without mentioning the crash. My story: wapo.st/49RSstP

nature.comPersuading voters using human–artificial intelligence dialogues - NatureHuman–artificial intelligence (AI) dialogues can meaningfully impact voters’ attitudes towards presidential candidates and policy, demonstrating the potential of conversational AI to influence politic...

You can read the Nature study, which found AI chatbots can persuade voters to change their opinions of candidates, here: www.nature.com/articles/s41... The Science study, which evaluated different persuasion tactics by AI chatbots, is here: www.science.org/doi/10.1126/...

Sherrerd

@duerpress.com

"Political AI might also prove to be pretty annoying..." heh

Don't despair, though! Talking to an AI chatbot is just one of countless dumb ways that voters could make their decisions. Thanks to @sivav.bsky.social & @whitneyphillips.bsky.social for their thoughtful commentary and a splash of cold water. wapo.st/49RSstP

hilde45.bsky.social

@hilde45.bsky.social

How are they at persuading other chatbots?

It Dan

@valenvsfel.bsky.social