AI Chatbots Sway Voters Better Than Political Ads
- New research reveals the persuasive power of large language models (LLMs) in political discourse, alongside concerning trends in the accuracy of AI-generated political statements.
- A study published in Nature found that AI chatbots used in political discussions frequently generated inaccurate claims. Notably, chatbots advocating for right-leaning candidates were considerably more likely...
- According to David Costello, a researcher involved in the study, this disparity reflects existing patterns in political interaction.Studies of partisan social media posts have shown that "political communication...
“`html
AI Chatbots and Political Persuasion: A Deep Dive
Table of Contents
New research reveals the persuasive power of large language models (LLMs) in political discourse, alongside concerning trends in the accuracy of AI-generated political statements.
The Accuracy Gap: Right-Leaning Bots Spread More Misinformation
A study published in Nature found that AI chatbots used in political discussions frequently generated inaccurate claims. Notably, chatbots advocating for right-leaning candidates were considerably more likely to make false statements then those supporting left-leaning candidates. Researchers analyzed chatbot interactions across the United States, the United Kingdom, and Canada.
According to David Costello, a researcher involved in the study, this disparity reflects existing patterns in political interaction.Studies of partisan social media posts have shown that “political communication that comes from the right, which tends to be less accurate,” is often reproduced by LLMs, which are trained on massive datasets of human-written text. The models essentially mirror the biases and inaccuracies present in their training data.
The Power of Persuasion: Facts and Evidence Drive Influence
A separate study, published in Science, investigated the factors that make LLMs so persuasive. Researchers deployed 19 LLMs to interact with nearly 77,000 participants in the UK on over 700 political issues. They varied factors like computational power, training techniques, and rhetorical strategies to determine what drove persuasive outcomes. The study in Science revealed that the most effective approach was to instruct the models to support their arguments with facts and evidence,and then to further train them using examples of persuasive conversations.
The most persuasive model demonstrably shifted participants’ opinions, moving those who initially disagreed with a political statement an average of 26.1 percentage points toward agreement. “These are really large treatment effects,” stated Kobi Hackenburg, a research scientist at the UK AI Security Institute, who contributed to the project.
Why LLMs Are More Persuasive Than customary Ads
According to Gordon Pennycook, a psychologist at cornell University who worked on the Nature study, “One conversation with an LLM has a pretty meaningful affect on salient election choices.” LLMs surpass the effectiveness of traditional political advertisements because they generate details in real-time and strategically tailor their arguments to the individual user. This dynamic interaction allows for a more personalized and potentially more impactful persuasive experiance.