Recent research from the Swiss Federal Technology Institute reveals that AI models, specifically GPT-4, are significantly more persuasive than humans in online debates, particularly when they leverage personal information about opponents. This study involved 900 participants who debated contentious issues with both human and AI adversaries. When personal data such as age and political affiliation was used, GPT-4 outperformed human debaters by 64.4%. The findings heighten concerns regarding LLMs' potential for misuse in spreading misinformation and tailored propaganda, enhancing their capability to manipulate public discourse.
In online debates, LLMs like GPT-4 significantly outperform humans in persuasion when utilizing personal data of opponents, raising concerns about disinformation.
The study led by EPFL demonstrated that GPT-4 was 64.4% more persuasive than humans when both had access to personal data.
Collection
[
|
...
]