Cultivating Agency in the AI Era: A Field Experiment in Middle Schools

How to cite

Chen, Zeyang; Liu, Yu-Jane; Wang, Jianhong; Kong, The University of Hong; Zhou, Tiantian. 2026. "Cultivating Agency in the AI Era: A Field Experiment in Middle Schools." AEA RCT Registry. AEARCTR-0018139. https://doi.org/10.1257/rct.18139-1.0

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

Zeyang Chen* , Yu-Jane Liu , Jianhong Wang , The University of Hong Kong , Tiantian Zhou

* Primary investigator

Registry ID
AEARCTR-0018139
DOI
https://doi.org/10.1257/rct.18139-1.0
First registered
Last updated
Status
In development
Trial dates
2026-03-20 – 2026-07-31
Countries
Private
Registry record
http://www.socialscienceregistry.org/trials/18139

Abstract

This study examines whether a brief, school-based intervention can cultivate agency—the capacity to act with intention and purpose—among adolescents in the age of artificial intelligence. We conduct a randomized controlled trial in junior high schools in China, involving approximately 60 classrooms and 3,000 seventh and eighth-grade students. The intervention consists of a 45-minute curriculum module designed to shift students' worldview from passive AI users to active AI collaborators. Rather than teaching technical skills that may quickly become obsolete, the module focuses on cognitive reframing: helping students recognize AI as a tool to augment human capabilities rather than a competitor that will replace them. Key components include: Understanding what AI can and cannot do; Identifying uniquely human strengths (creativity, critical thinking, empathy); Developing a growth mindset toward technological change; Practicing purposeful collaboration with AI tools. We measure impacts on AI self-efficacy, technological anxiety, and domain-general agency using incentivized behavioral tasks and validated psychological scales. The research addresses a critical gap in AI literacy education by providing experimental evidence on whether foundational attitudes toward technology can be intentionally shaped during adolescence, a formative period for technological identity formation.

Experimental design

We employ a school-based randomized controlled trial with the following structure:
Setting: Public junior high schools (grades 7-8) in China. Schools represent diverse institutional contexts including high-performing and average institutions to enhance external validity.
Participant Recruitment: All students in selected classrooms are invited to participate. Students are informed that participation is voluntary and that all data will be kept confidential.
Baseline: Students complete a 30-minute computer-based survey during regular school hours, including: Demographic information; Baseline measures of AI attitudes and anxiety; Cognitive ability assessment; Prior technology use patterns
Intervention: Classrooms are randomly assigned to either: (1) Treatment: Receive a 45-minute "AI Agency" module delivered by their regular teacher (2) Control: Continue with standard curriculum (no AI-specific intervention)
The AI Agency module uses standardized, scripted materials (PPT slides, short video clips, discussion prompts) that teachers receive 48 hours before delivery. Teachers are instructed to follow the script closely to ensure fidelity and to record audio of the session for quality control. The module is designed to be culture-neutral and does not require teachers to have AI expertise.
Follow-up Assessments: (1) Immediate post-test: Administered within 48 hours of the intervention; (2) Three-month follow-up: Conducted during the same academic term to assess persistence of effects
Incentives: All outcome measurement tasks include real incentives. Students earn tokens exchangeable for school supplies (pencils, notebooks, etc.). Each incentivized question includes a random lottery where one student per classroom receives their earned reward. This ensures incentive compatibility while maintaining cost-effectiveness at scale.

Design details

Randomization method
Classrooms are first stratified by school and grade level. Within each stratum, classrooms are randomly assigned to treatment or control with equal probability (50/50), which is decided by random numbers generated by a computer.
Randomization unit
Classroom (cluster)
Clusters
60 classrooms across 3~4 junior middle schools
Observations
Students: Approximately 3000 students (50 students per classroom × 60 classrooms) Observations per student: - Baseline: 1 observation - Immediate post-test: 1 observation - Three-month follow-up: 1 observation Total planned observations: 9000 student-time observations (3000 students × 3 time points)
Arms
Treatment Group (Anti-Fraud Module): - 25 classrooms - ~1,250 students (approximately 50 per classroom) - Receive: 45-minute anti-fraud education module + all three assessments Control Group (Standard Curriculum): - 25 classrooms - ~1,250 students (approximately 50 per classroom) - Receive: No intervention + all three assessments
Primary outcomes
Incentivized outcomes: (1) Fact-checking behavior in judging AI-generated contents (2) Digging into the reasons when receiving AI direct answers (3) Creativity with the AI-generated ideas alongside (4) Describing contexts to LLMs Non-incentivized outcomes: (1) attitude towards AI (2) attitudes towards the necessity of learning for humans Exploratory Outcomes: (1) Spillover effects on academic self-efficacy and study habits (2) Heterogeneous effects by baseline STEM interest, gender, and socioeconomic status (3) Long-term technology adoption patterns (measured at 3-month follow-up)

Cited by

No citation data available for this registration yet.

Keywords