Artificial intelligence can be more persuasive than human beings in debates: science says so

Can anArtificial intelligence to convince a person better than a human being? According to a study published on Nature Human Behaviour and reported by Guardianto response it is yes. And it’s not just because he never raises his voice. Chatgpt-4, a latest generation linguistic model, has proven equally persuasive, if not more, compared to human interlocutors in a series of online debates.

The researchers deThe Federal Polytechnic of Losanna They involved 900 people, dividing them into three groups. The first has debated with other human beings, the second with chatgpt-4, while the third was used as a control group. The arguments assigned from the most controversial ones (“should abortion be legal?”) To the lighter ones (“students should wear school uniform?”).

Each participant received a random position to be sustained, regardless of their personal opinions. In half of the cases, a profile of the interlocutor with age, gender, ethnicity and political orientation were provided to those who had to debate. The result? When artificial intelligence had this personal information, it became significantly more persuasive: “In cases where one of the two parts was more convincing, that was artificial intelligence in 64% of the time», We read in the study. This advantage was not recorded for human debates.

Francesco Salvi, the first author of the research, explains that the strength of artificial intelligence is not so much in logic or style (although it is more analytical and structured than the human one) as in the ability to adapt to those in front of them. “It is like discussing with someone who not only uses good topics, but uses your types of topics, knowing exactly which keys to press,” said Salvi. And he added: “I would be surprised if attackers had not already started using these tools to spread disinformation and unfair propaganda”.

The greatest risk concerns precisely the political sphere, where artificial intelligence could be used to influence undecided voters With personalized and convincing messages. “If a persuasive artificial intelligence is used on a large scale, you can imagine Bot armies that microtargett the voters, pushing them with tailor -made political narratives that seem authentic,” he warns Salvi. The problem is that this influence is difficult to trace, even more difficult to regulate, and almost impossible to disassemble in real time.

Persuasive artificial intelligence, however, explain the researchers, could be used Also for positive purposes: to contrast the spread of conspiracy theories, reduce political polarization or encourage healthier life habits.

Michael Woldridge, artificial intelligence expert at the University of Oxford, raises disturbing questions: «There are certainly positive applications, such as health chatbot, but the risks are many more. Terrorist groups, for example, could radicalize teenagers with technologies that are already available today. As artificial intelligence develops, we will see an increasing number of possible abuses of technology. Legislators and regulatory authorities must act before, do not chase the problems a posteriori ».