ChatGPT and teams set their own norms of behavior

A recent study reveals that ChatGPT and other large AI-based language models have the ability to self-organize and do not need human intervention. When in groups, these language models spontaneously create behavioral norms and social conventions, just as happens in human society, Italian media reported today.

This unique study, published in the journal Science Advances, was led by Italian scientist Andrea Baronchelli, who works at City University of London, St George's, according to the Ansa agency.

"The mechanism is the same as that underlying most complex human behavior, such as rules of etiquette, dress codes, and moral norms," ​​he said.

The researchers conducted an adapted experiment that is also used to study social conventions in humans, forming two sets of language models like ChatGPT.

Each group was asked to choose a name from several options provided – in this experiment, a letter of the alphabet or a random string of characters.

"Linguistic models, the so-called LLMs (large linguistic models), received positive points whenever they successfully coordinated the use of the same name, while in case of failure they received negative points," Baronchelli explained.

According to him, this mechanism serves to motivate actors to agree through local interaction – a scheme similar to those previously used in laboratory experiments with humans, who were offered small monetary rewards for successfully solving tasks. /Telegraph/