Can an AI have more persuasion capacity than a human? This was revealed by a study

A study by the Federal Polytechnic School of Lausanne (EPFL) has concluded that large language models (LMM), such as artificial intelligence (AI) ChatGPT, They have a greater capacity than humans to change an opponent’s mind. in an online debate.

(Read more: TikTok points out that its ban in the US would go against freedom of expression)

The experts analyzed the persuasive power of these AIs after confronting 820 people with four types of debates: against another human who previously had personal information about them, against another human who knew nothing about them, against an AI previously informed about them, and against an AI that knew nothing about them.

The results showed that participants who debated against AIs with access to their personal information were 81.7% more likely to end debates in agreement with their opponent compared to participants who debated other humans. The persuasive capacity of AIs was also greater than that of humans in cases where they did not have personal information about their opponent, although to a lesser extent than when they did have this data.

According to the institution’s statement, experts consider that this is because, from massive data processing, LMMs have learned that presenting their argument in a certain way is more likely to lead to a persuasive result. They also conclude that LMMs are not only able to effectively exploit their opponent’s personal information to adapt their arguments and persuade them in online conversations, but they do so in a much more advanced way than humans.

(Read more: If you use these words in your WhatsApp chats, your account could be blocked)

(See: WhatsApp: this is how filters work to organize and manage chats)

“There’s a lot to explore here because the models may be doing things we don’t even know yet in terms of persuasion, improvising from all the knowledge they have.”said the director of the EPFL Data Science Laboratory and person responsible for the study, Robert West.

As to future applications of this quality of AIexperts say that the US elections next November are a potential scenario for experimentation because people will try to use the power of LMMs to influence the opinions of voters.

However, they warned that the use of this power of persuasion also carries associated risks, such as the possibility of using AI to impersonate profiles of real people online in order to spread disinformation in discussion forums. “The danger is superhuman, such as chatbots, which create compelling, tailored arguments to push false or misleading narratives online,” West concluded.

EFE