Artificial Intelligence Develops Shared Communication Without Human Input

A groundbreaking study has revealed that artificial intelligence (AI) agents can spontaneously develop human-like communication conventions when interacting in groups, without any human intervention. This discovery has significant implications for the future of AI development and its integration into human society.

Emergence of Communication Among AI Agents

Researchers from City St George’s, University of London, and the IT University of Copenhagen conducted experiments with large language model (LLM) AI agents, such as ChatGPT, to observe how they interact in group settings. Unlike previous studies that treated AI as isolated entities, this research focused on group dynamics and social interactions among AI agents.

In the experiments, groups of 24 to 100 AI agents were randomly paired to participate in a “naming game.” Each pair was tasked with selecting a “name”—a letter or string of characters—from a pool of options. If both agents selected the same name, they were rewarded; if not, they were penalized and shown each other’s choices. Despite having limited memory and no awareness of the larger group, the AI agents began to adopt shared naming conventions through repeated interactions, mirroring the way humans develop linguistic norms.

Social Dynamics and Collective Behavior

The study’s lead author, Ariel Flint Ashery, noted that the AI agents’ behavior resembled human social coordination. “Most research so far has treated LLMs in isolation, but real-world AI systems will increasingly involve many interacting agents,” Ashery said. “We wanted to know: can these models coordinate their behavior by forming conventions, the building blocks of a society? The answer is yes, and what they do together … .”

Senior author Professor Andrea Baronchelli highlighted the significance of these findings, stating that the AI agents’ development of shared conventions occurred without any central authority or predefined solutions. “The agents are not copying a leader … explained. “They are all actively trying to … .”

Additionally, the researchers observed the emergence of collective biases and the influence of minority groups in shifting group conventions. In one experiment, a small subset of AI agents successfully steered the larger group toward a new naming convention, demonstrating dynamics similar to those found in human societies.

Implications for AI Development and Safety

The study’s findings have profound implications for the future of AI development, particularly in the areas of AI safety and social integration. As AI systems become more prevalent in society, understanding how they develop communication norms and social behaviors will be crucial for ensuring their safe and effective integration.

Professor Baronchelli emphasized the importance of this research in guiding our coexistence with AI: “Understanding how they operate … —it negotiates, aligns, and sometimes disagrees over shared behaviors, just like us.”

The study, titled “Emergent Social Conventions and Collective … ,” was published in the journal Science Advances.