AI Can Be More Persuasive Than Humans In Debates, Study Shows

A new study published in Nature Human Behaviour shows that AI, particularly GPT-4, can be as persuasive as humans in debates—and even outperform them when armed with basic demographic information about their opponent.

In the study, participants were randomly assigned to argue either for or against sociopolitical issues like school uniforms, fossil fuel bans, or the societal benefits of AI.

Before each debate, they filled out surveys with demographic details such as age, gender, education, and political affiliation. Their opinions were measured before and after each exchange.

When no background data was shared, GPT-4 and human debaters were equally persuasive. But when given demographic information, GPT-4 outperformed humans 64% of the time.

Interestingly, AI was more effective on topics that participants didn’t have strong prior opinions about.

Researchers noted that AI tended to use more structured and analytical arguments than humans. However, this alone didn’t fully explain its persuasive edge.

The findings raise questions about how language models could be used to shape public opinion, particularly in areas like politics or targeted marketing.