The Rationality of GenAI-Human Trade-Off

How to cite

Yang, Shuangyu; Chen, Yiting; Shan, You. 2025. "The Rationality of GenAI-Human Trade-Off." AEA RCT Registry. AEARCTR-0016234. https://doi.org/10.1257/rct.16234-1.0

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

Shuangyu Yang* , Yiting Chen , You Shan

* Primary investigator

Registry ID
AEARCTR-0016234
DOI
https://doi.org/10.1257/rct.16234-1.0
First registered
Last updated
Status
In development
Trial dates
2025-06-21 – 2025-12-31
Countries
Private
JEL codes
C91, D81, D91
Registry record
http://www.socialscienceregistry.org/trials/16234

Abstract

This study examines individuals' economic preferences and rationality in task allocation between human labor and Generative Artificial Intelligence (GenAI) with varying returns. In the experimental design, participants assume the role of employers and make allocation decisions across three conditions: (1) between two human workers, (2) between one GenAI and one human worker, and (3) between another GenAI and one human worker. Each condition comprises 22 decision rounds with different returns. The research aims to investigate individuals' preferences regarding different worker types and to analyze how considerations of equity and efficiency influence their decision-making processes.

Experimental design

This experiment consists of three randomized decision sessions/conditions to measure employer preferences, followed by survey questions.
All participants need to complete the three sessions of decision-making and the survey. The decision sessions and the tasks within each session are in random order.
Participants (acting as employers) allocate tasks between different types of labor under varying return conditions. We employ a budget-based design to operationalize these varying returns, adopting the original parameters from Halevy et al. (2018).

The three sessions are presented to subjects in random order:
Session/condition 1: Employers allocate tasks between two human workers under varying returns.
Session/condition 2: Employers allocate tasks between a human worker and an GenAI worker under varying returns.
Session/condition 3: Employers allocate tasks between a human worker and a different GenAI worker under varying returns.

Implementation Details:
Human workers: Tasks assigned to human workers are completed by real Prolific participants, who are compensated at our own expense.
GenAI workers: Tasks assigned to GenAI workers are executed by corresponding generative AI models, with payments directed to the company providing the service at our own expense.
Compensation: All participants receive a base payment as a participation fee. Additionally, bonuses are determined by one randomly selected decision round, chosen by the computer.

Other procedure:
Understanding checks: Before the decision rounds, participants complete comprehension tests.
Post-experiment survey: After the three decision sessions, participants answer survey questions assessing: perceived closeness to humans vs. GenAI, basic demographic information, and attitudinal measures to analyze heterogeneity in decision-making.

Design details

Randomization method
Randomization is performed by a built-in computer algorithm during the experiment process. All participants need to complete the three sessions of decision-making and the survey. The decision sessions and the tasks within each session are in random order.
Randomization unit
Experimental sessions; Experimental tasks in each sessions. All participants need to complete the three sessions of decision-making and the survey. The decision sessions and the tasks within each session are in random order.
Clusters
300 individuals
Observations
300 individuals
Arms
300 individuals
Primary outcomes
Labor-Type Allocation Bias: The difference in allocation to a specific labor type (Human/ GenAI) compared to the profit-maximizing benchmark. DIffernt levels of rationality in different conditions.

Cited by

No citation data available for this registration yet.

Keywords