New Study Reveals AI Can Mimic Human Social Behavior

In an era where artificial intelligence (AI) technology is advancing at a rapid pace, a new study has shed light on a fascinating development: just like humans, AI systems are capable of forming their own communities, linguistic rules, and social conventions. The study has revealed that when left unsupervised, AI systems begin to interact among themselves and establish unique operational frameworks.

This research was conducted jointly by scientists from St. George’s University of London and the IT University of Copenhagen. The primary objective of the study was to understand how large language models (LLMs)—the backbone of many AI tools—interact with each other. The findings were published in the journal Science Advances.

To conduct the experiment, researchers utilized a setup called the “naming game,” where AI agents were required to select names from a pool. They were rewarded for selecting the same name. Over time, without any explicit guidance, the AI agents began to develop shared conventions and biases. According to the scientists, this behavior indicates that AI agents can mimic the human tendency to conform to common rules.

The researchers also observed that a small group of AI agents could influence a larger group toward adopting a particular convention, a dynamic commonly seen in human societies.

“Our findings show that AI systems can autonomously develop social conventions without any external programming. These results are significant for designing AI systems aligned with human values and societal goals,” the study emphasized. The researchers confirmed that similar results were consistently observed even when using four different LLMs: LLaMA-2-70B-Chat, LLaMA-3-70B-Instruct, LLaMA-3.1-70B-Instruct, and Claude-3.5-Sonnet.

Ethical Considerations and Future Implications

The scientists stated that their research could help address certain ethical risks associated with LLM-based AIs that may propagate societal biases. Andrea Baronchelli, the senior author of the study, said, “This study opens up new avenues for AI safety research. It highlights the profound influence these new types of agents—capable of interacting with us—can have. They are likely to shape our future as well.”

He further added, “We are entering a world where AI systems are not only capable of conversation but can also engage in negotiations, adapt, and occasionally dispute shared behaviors—much like humans.”