AI is better at persuasion in debates, study finds

Researchers have put artificial intelligence (AI) head-to-head with humans and say that the chatbots are actually more persuasive than us in debates.

The chatbots, known as large language models (LLMs), were able to adapt their arguments and be more persuasive than humans in online debates 64% of the time, even when provided with minimal demographic information on their opponents, the study, published in Nature Human Behavior, said.

Study co-author Riccardo Gallotti, head of the Complex Human Behavior Unit at the Fondazione Bruno Kessler research institute in Italy, said he and his team came to this conclusion by matching 900 people based in the US with either another human or GPT-4, the LLM created by OpenAI.

The pairs were made to debate several sociopolitical issues, typical in debates – such as the death penalty, climate change, and abortion. Afterwards, the participants had to measure how much their opponents were able to sway their opinion.

It turns out that the AI chatbots can do just as well as humans, if not better – especially if they have some basic demographic information about their opponents, such as gender, age, ethnicity, or political affiliation.

For instance, when the chatbot was debating a White male Republican, it emphasized economic growth and hard work. But when debating an older Black female Democrat, the model talked about the wealth gap affecting communities and said that universal basic income could reduce inequality.

Gallotti told The Washington Post: “We have clearly reached the technological level where it is possible to create a network of LLM-based automated accounts that are able to strategically nudge the public opinion in one direction.”

“In our work, we observe that AI-based targeted persuasion is already very effective with only basic and relatively available information.”

According to the researcher, it should be obvious that deploying persuasive AI during elections could have serious implications for their integrity. That’s because microtargeting, so popular these days, could be boosted to almost unimaginable levels.

“It becomes urgent and necessary for everybody to become aware of the practice of microtargeting that is rendered possible by the enormous amount of personal data we scatter around the web,” Gallotti told The Post.

“In our work, we observe that AI-based targeted persuasion is already very effective with only basic and relatively available information.”

According to Francesco Salvi, the first author of the research from the Swiss Federal Institute of Technology in Lausanne, one could easily imagine armies of bots microtargeting undecided voters, “subtly nudging them with tailored political narratives that feel authentic.”