AI Assistance, Social Preferences, and Beliefs in Economic Games

How to cite

Jiang, Xin; Ye, Bohan. 2026. "AI Assistance, Social Preferences, and Beliefs in Economic Games." AEA RCT Registry. AEARCTR-0019198. https://doi.org/10.1257/rct.19198-1.0

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

Xin Jiang* , Bohan Ye

* Primary investigator

Registry ID
AEARCTR-0019198
DOI
https://doi.org/10.1257/rct.19198-1.0
First registered
Last updated
Status
In development
Trial dates
2026-09-21 – 2026-12-31
Countries
Private
Registry record
http://www.socialscienceregistry.org/trials/19198

Abstract

This study examines whether access to AI assistance changes human decisions, beliefs, and reasoning in canonical economic games. Participants complete dictator-game, trust-game, and public-goods decisions first without AI assistance and then with the option to consult an AI assistant. The main analysis compares matched within-subject decisions across the no-AI and AI-assist conditions, with secondary analyses of belief revision, heterogeneity by baseline beliefs and reasons, and the content of AI prompts.

Intervention

Participants in the AI-assist condition are allowed to consult an AI assistant before submitting decisions in the dictator game, trust game, and public-goods game. The interface asks participants to record the prompt they gave to the AI assistant and the AI response. Participants may choose not to use AI assistance. In the no-AI condition, participants complete the same economic-game decisions without AI assistance.

Experimental design

This study uses a two-arm sequence design. In both arms, participants first complete the same set of economic-game tasks without AI assistance. Participants are then assigned to one of two second-wave conditions.

In the control arm, participants complete the tasks again without AI assistance: no AI -> no AI. In the AI-assist arm, participants complete the tasks again with access to AI assistance: no AI -> AI.
The economic-game tasks include a dictator game, a trust game, and a public-goods game. Participants are matched across the two waves using a study identifier.

Design details

Randomization method
randomization done by a computer
Randomization unit
No clustered treatment assignment is planned. The unit of observation is the individual participant.
Clusters
0
Observations
We plan to recruit approximately 500-600 participants total
Arms
We plan to recruit 150-200 participants to the no AI -> no AI control sequence and 300-400 participants to the no AI -> AI treatment sequence. This unequal allocation reflects expected implementation capacity and the study’s interest in measuring heterogeneity and mechanisms within the AI-assist condition. The primary treatment effect will compare the Wave 1 to Wave 2 change in the AI-assist arm with the corresponding change in the control arm.
Primary outcomes
Dictator-game giving Amount given to Player B in the dictator game. Trust-game sending Amount sent by Player A to Player B in the trust game. Trust-game returning Mean return share in the trust game, calculated for each participant as the average of returned amount divided by received amount across all nonzero received amounts. Public-goods contribution Amount contributed to the group project.
Secondary outcomes
Secondary outcomes include expected average public-goods contribution by others, trust-game expectations when collected, changes in beliefs between conditions, AI prompt use, AI prompt categories, and AI response content.

Cited by

No citation data available for this registration yet.

Keywords