The Impact of AI on Team Cooperation

How to cite

Long, Xianling; Huang, Kaixing; Pi, Lulu. 2026. "The Impact of AI on Team Cooperation." AEA RCT Registry. AEARCTR-0018641. https://doi.org/10.1257/rct.18641-1.0

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

Xianling Long* , Kaixing Huang , Lulu Pi

* Primary investigator

Registry ID
AEARCTR-0018641
DOI
https://doi.org/10.1257/rct.18641-1.0
First registered
Last updated
Status
Completed
Trial dates
2025-05-20 – 2025-10-30
Countries
China ()
Registry record
http://www.socialscienceregistry.org/trials/18641

Abstract

As artificial intelligence increasingly enters work, decision-making, and collaborative settings, an important question arises: does the introduction of AI promote or weaken human team cooperation? Existing research has focused more on the effects of AI on individual decision-making, productivity, and judgment accuracy. However, there is still a lack of direct evidence on how AI changes information exchange, trust formation, responsibility allocation, and collaborative performance among team members.

This study proposes to use the card game Guandan, played in two-person teams against another team, as the experimental setting to examine how the introduction of AI into the team collaboration process affects the quality of team cooperation, communication patterns, perceived trust, and task performance. Guandan is chosen as the research setting mainly because the task has a clear common objective, strong team-coordination features, quantifiable cooperation outcomes, and high operational feasibility.

Intervention

The intervention varies the presence of AI in a two-person team-based Guandan game. Treatment conditions include: whether AI is actually introduced as a teammate; whether a human teammate is replaced by another human; whether teammate test-round performance is disclosed.

Experimental design

Participants are randomly assigned to experimental conditions before the experiment begins. The randomization is organized around two dimensions. First, the study varies whether a participant’s original teammate is replaced in the formal task, and whether the replacement is another human participant or an AI robot. Second, the study varies whether participants are informed of their teammate’s performance in the test round, including the test-round performance of an AI teammate when applicable.

Design details

Randomization method
Randomization will be conducted by computer before the experiment begins.
Randomization unit
Individual.
Clusters
400 individuals
Observations
400 individual-level observations. In terms of gameplay, participants are organized into four-person groups, implying approximately 100 four-person experimental groups. Since each participant completes one test round and two formal rounds, the planned design also generates approximately 1,200 participant-round observations.
Arms
approximately equal allocation across 6 arms, around 67 individuals per treatment arm.
Primary outcomes
Main outcomes include team cooperation quality, communication patterns, perceived trust, responsibility attribution, strategic choices, task performance, collaboration efficiency, and win/loss outcomes. Additional recorded outcomes include game records and round-level results.

Cited by

No citation data available for this registration yet.

Keywords