"A study published today in Nature found that participants' preferences in real-world elections swung by up to 15 percentage points after conversing with a chatbot. In a related paper published in Science, researchers showed that these chatbots' effectiveness stems from their ability to synthesize a lot of information in a conversational way. The findings showcase the persuasive power of chatbots, which are used by more than one hundred million users each day,"
"Both papers found that chatbots influence voter opinions not by using emotional appeals or storytelling, but by flooding the user with information. The more information the chatbots provided, the more persuasive they were - but they were also more likely to produce false statements, the authors found. This can make AI into "a very dangerous thing", says Lisa Argyle, a computational social scientist at Purdue University in West Lafayette, Indiana."
Participants' preferences in real-world elections swung by up to 15 percentage points after conversing with a chatbot. Chatbots are persuasive because they synthesize large amounts of information in a conversational way. Chatbots influence voter opinions not through emotional appeals or storytelling but by flooding users with information. Greater information provision increased persuasiveness while also raising the likelihood of false statements. Such effects can cause people to become more misinformed rather than better informed. Rapid mainstream adoption of chatbots since 2023 elevates concerns about manipulation of public opinion. Nearly 6,000 participants from Canada, Poland, and the United States provided preference ratings on a 0-to-100 scale.
Read at Nature
Unable to calculate read time
Collection
[
|
...
]