Research: AI agent with explicit social norms coordinates with humans almost 4 times better

Researchers in July 2026 demonstrated through an example of pedestrian-driver interaction that if implicit social behavioral norms are explicitly formalized and embedded in an AI agent, its coordination with humans in dynamic situations improves by nearly four times compared to a baseline strategy.

Why AI struggles to coordinate with people in the moment

Humans constantly coordinate with each other through implicit, difficult-to-formalize social norms — common tacit expectations that participants bring to interactions. As AI agents, including large language models, become increasingly embedded in everyday life, they too participate in such dynamic interactions, but often coordinate with people inefficiently, inattentively, and "unnaturally."

The authors hypothesize that the reason lies in how such models are typically trained: their behavior is fitted to human demonstrations, but the norms that generate this behavior are not explicitly identified. In other words, the model copies the result but does not understand the rule behind it.

How the hypothesis was tested

To test their hypothesis, researchers chose pedestrian-driver interaction — a typical example of dynamic coordination where both sides constantly read each other's intentions without words — and built a simplified experimental platform that reproduced the key features of such interaction.

• 3,456 dynamic human-to-human interactions were collected through this platform
• Based on these data, three principles underlying social norms were identified: outcome predictability, value alignment, and advantage awareness
• These three principles were explicitly embedded into the AI agent's behavior
• The effect was validated in a closed-loop task — that is, in real dynamic interaction with live humans, not in a static test on historical data

How much coordination improved

In the closed-loop task with humans, the AI agent trained with social norms achieved nearly four times higher total scores than the baseline strategy without norms. Moreover, this agent outperformed human-to-human pairs in coordination quality by 43%.

What this means

The results suggest that formalizing implicit social norms into explicit, measurable principles — rather than simply copying behavior from demonstrations — could be a workable approach to making AI agents more natural partners for humans in dynamic interactions: from autonomous vehicles that must "negotiate" with pedestrians without words, to any other scenarios where robots and assistants need to coordinate with people in real time.

An important detail: the authors validated the effect not on historical records, but in a closed loop — meaning the agent actually interacted with live humans in real time, not merely predicting what a human would have done in a past recorded interaction. This distinguishes the result from many laboratory tests of human-machine interaction where models are evaluated by their match with already-recorded human behavior, rather than by the quality of real-time coordination.

Frequently asked questions

How many interactions were collected for the study?

The authors collected 3,456 dynamic pedestrian-driver interactions through a specially constructed experimental platform.

What three principles formed the basis of social norms?

Researchers identified outcome predictability, value alignment, and advantage awareness as key principles explaining human social coordination norms.

How much better did the AI agent coordinate after training on norms?

The agent trained with social norms achieved nearly four times higher scores than the baseline strategy and outperformed human-to-human coordination by 43%.