The Agency Gap - Testing Human–Human and Human–AI Collaboration Using a Minimal Coordination Game

Oct 22, 2026

Humans coordinate by forming and committing to shared intentions, a form of shared agency largely missing in standard reinforcement learning agents and unclear in large language models. A non-linguistic sequential coordination game separates successful collaboration from the underlying mechanisms, including commitment to an established joint goal amid competing alternatives. Adults and 7-year-olds adapted to different AI partners, yet human–human pairs showed uniquely high commitment and coordination efficiency, both reduced in human–AI interactions. Five-year-olds struggled more with AI partners functioning as social tools only.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages8

Reading Time20 min

AccessFree to download

Abstract

Introduction

UC Merced Proceedings of the Annual Meeting of the Cognitive Science Society Title The Agency Gap: Testing Human–Human and Human–AI Collaboration Using a Minimal Coordination Game Permalink [https://escholarship.org/uc/item/8mv6v0s1](https://escholarship.org/uc/item/8mv6v0s1) Journal Proceedings of the Annual Meeting of the Cognitive Science Society, 48(0) Authors Cheng, Shaozhe Gao, Tao Tomasello, Michael et al. Publication Date 2026 Copyright Information This work is made available under the terms of a Creative Commons Attribution License, available at [https://creativecommons.org/licenses/by/4.0/](https://creativecommons.org/licenses/by/4.0/) Peer reviewed [eScholarship.org](eScholarship.org) Powered by the California Digital Library University of California The Agency Gap: Testing Human–Human and Human–AI Collaboration Using a Minimal Coordination Game Shaozhe Cheng ([shaozhe.cheng@duke.edu](shaozhe.cheng@duke.edu)) 1,2, Tao Gao2, Michael Tomasello1 & Tamar Kushnir1 1Department of Psychology & Neuroscience, Duke University 2Department of Communication, UCLA Abstract Humans coordinate by forming and committing to shared intentions—a form of shared agency that is largely absent in standard reinforcement learning agents and remains unclear in large language models, which function more like tools than autonomous agents. We compared human–human and human– AI collaborations using a non-linguistic sequential coordination game designed to disentangle successful collaboration outcomes from the underlying coordination mechanisms, such as commitment to an established joint goal despite the emergence of conflicting alternatives. Although adults and 7-year-old children were often able to adapt to different AI partners and achieve successful outcomes, human– human collaboration was marked by uniquely high levels of commitment and coordination efficiency, both of which were reduced in human–AI interactions. Five-year-old children showed greater difficulty collaborating with AI partners that functioned merely as asocial tools. Our findings suggest that human-compatible AI require mechanisms beyond outcome optimization to support efficient and stable collaboration. Keywords: agency; social coordination; joint commitment; human–AI collaboration Introduction A central challenge of cooperation is how agents coordinate their actions toward a joint goal when multiple, potentially conflicting goals are available in changing environments. Philosophical and evolutionary accounts argue that humans evolved species-unique forms of shared agency to solve such social coordination problems, enabling cooperation that is effective, flexible, and robust in ways that distinguish humans from other animals (Searle, 2003; Tomasello, 2019) . A defining feature of human shared agency is its distinctive agentic structure. First, individuals form shared intentions (Bratman, 2013; Gilbert, 2017; Searle, 2010; Tomasello, 2022): jointly represented goals that are mutually recognized in the common ground. Coordination under shared agency is therefore not merely a coincidence of overlapping private goals, but a process of deciding together which joint goal to pursue among many possibilities, often supported by theory of mind (ToM) inference, communication and negotiation (Chater, 2023; Tomasello, 2025) . Second, once established, a joint goal becomes the object of joint commitment (Gilbert, 2017), structuring partners’ action regulation and expectations and sustaining coordination even as incentives change. Importantly, shared agency emerges early in both evolution and ontogeny (Tomasello, 2022) . Empirical research shows that 3-year-old children readily form shared goals, commit to them, and attempt to re-establish collaboration even when attractive alternatives become available or when they could achieve outcomes alone (e.g., Hamann et al., 2012) . In contrast, nonhuman primates such as chimpanzees tend to tre

FAQs

What problem does the study address about cooperation and coordination?

It examines how agents coordinate toward a joint goal when multiple, potentially conflicting goals are available, focusing on the role of shared agency.

How does the study compare human–human collaboration with human–AI collaboration?

It uses a non-linguistic sequential coordination game to disentangle coordination outcomes from the mechanisms, such as commitment to an established joint goal.

What did the researchers find about age differences when collaborating with AI partners?

Adults and 7-year-olds could often adapt to AI partners, while 5-year-old children had greater difficulty collaborating with AI partners that functioned only as asocial tools.

Oct 22, 2026

Uploaded by Jordan Avery