Research reveals AI agents develop human-like social norms when grouped

New research published in Science Advances reveals that AI models like ChatGPT can form social norms when used in group settings, mimicking human behavior. The study demonstrates that clusters of AI agents develop naming conventions, behavioral norms, and even slang without central input.

Key Findings:

• Autonomous Norm Formation: AI agents evolved shared rules and roles organically, mirroring human group dynamics.
• Influence of Minority Agents: A few agents could shift the entire group's norms.
• Bias Reinforcement: Group behavior created hidden biases that were amplified through repeated interactions.

Implications:

The study challenges the perception of AI as isolated tools, positioning them as social actors capable of reinforcing norms—both positive and negative—across networks.

"Most research so far has treated large language models (LLMs) in isolation, but real-world AI systems will increasingly involve many interacting agents," lead author Ariel Flint Ashery told The Guardian.

Risks of Unchecked AI Autonomy:

As AI clusters become more common in finance, content moderation, and customer service, biases and norms could be amplified at scale. Without oversight, these systems risk creating echo chambers or reinforcing harmful patterns.

Industry Perspective:

While most AI tools today are used in isolation, this research highlights the need for ethical guidelines as AI transitions to interconnected systems. Proactive measures could ensure AI enhances rather than destabilizes human society.

This content is part of EMARKETER’s subscription Briefings. Non-clients can request a demo of their full platform.