Generative AI, the Future of Work, and Policy Preferences

How to cite

Betai, Neha; Chen, Maggie. 2025. "Generative AI, the Future of Work, and Policy Preferences." AEA RCT Registry. AEARCTR-0017244. https://doi.org/10.1257/rct.17244-3.0

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

Neha Betai* , Maggie Chen

* Primary investigator

Registry ID
AEARCTR-0017244
DOI
https://doi.org/10.1257/rct.17244-3.0
First registered
Last updated
Status
On going
Trial dates
2025-12-02 – 2026-07-31
Countries
Private
JEL codes
O33, J24, F42
Registry record
http://www.socialscienceregistry.org/trials/17244

Abstract

This project examines how individuals in low- and middle-income countries perceive and respond to the rapid diffusion of generative artificial intelligence across economic, social, and governance domains. While generative AI offers substantial potential productivity gains, uncertainty about its implications for employment, privacy, fairness, and global power dynamics may change adoption patterns. The uncertainty and risks associated with its adoption also raises concerns about data governance, unequal global distribution of AI capabilities, and appropriate roles of governments in regulating or promoting AI.
To identify the mechanisms shaping adoption, beliefs, and governance preferences, we conduct a randomized survey experiment with participants across diverse demographic and occupational groups. Respondents are assigned to one of six informational treatments: (1) practical instruction on AI use and prompt engineering; (2) labor-market disruption risks; (3) labor-market opportunities; (4) algorithmic bias; (5) privacy concerns; and (6) geopolitical competition and global inequality in AI capacity. The experiment measures effects on willingness to reskill, willingness to use and invest in AI, peer contact and social effects, and willingness to supply data. It also captures policy preferences, including attitudes toward AI-related redistribution, desired government promotion and regulation of AI, the use of AI in allocation decisions, and preferences over domestic versus international AI development and governance. These outcomes allow us to assess not only behavioral adjustments but also how individuals think governments should respond to the opportunities and risks of AI.
The study aims to provide causal evidence on the belief channels shaping AI adoption, skill adaptation, and support for national and global AI governance. The results inform policy interventions aimed at expanding access to AI’s benefits while addressing emerging socioeconomic, ethical, and geopolitical concerns.

Intervention

The intervention consists of a randomized provision of short informational modules designed to influence participants’ perceptions of generative artificial intelligence. Each participant is randomly assigned to one of six treatment conditions or a control condition. All interventions are delivered within an online survey with no deception being used.
Each informational module highlights a distinct dimension of AI’s potential impact:
1. AI Skills and Usage Tutorial
2. Labor-Market Disruption
3. Labor-Market Opportunities
4. Algorithmic Bias
5. Privacy Risks
6. Geopolitics and Global Inequality in AI Capacity
7. Control Group

After receiving the assigned module, respondents complete a set of questions measuring willingness to use or invest in AI tools, willingness to reskill, peer contact and social effects, willingness to supply data, trust in AI technologies, preferences for domestic AI regulation and promotion, preferences for AI’s role in public decision-making, preferences for national vs. international AI governance.

The intervention is minimal-risk and informational in nature.

Experimental design

The study conducts an online randomized survey experiment to examine how different types of information about artificial intelligence can influence individuals’ beliefs, adoption intentions, and policy preferences. Participants are recruited from two populations, general online workers (e.g., MTurk) and skilled freelancers on a major freelancing platform, with an emphasis on respondents residing in low- and middle-income countries.
After providing baseline demographic information, each participant is randomly assigned to one of six informational treatment conditions that highlight the impact of AI or a control group. Following exposure to the assigned module, all participants complete a standardized set of outcome questions on AI-related adoption intentions, willingness to reskill, attitudes toward data sharing, and preferences for domestic and international AI governance.
Randomization is conducted at the individual level within the survey platform. Total planned sample size is approximately 3,500-4000 respondents.

Design details

Randomization method
Randomization is conducted automatically by the survey platform using a computer-generated random number assignment. Each participant is assigned to one of the treatment arms or the control group with equal probability. The process is fully automated and free of any influence from the researchers.
Randomization unit
Randomization is conducted is at the individual level.
Clusters
Approximately 3,000-4000 individuals (no higher-level clusters; randomization at the individual level).
Observations
Approximately 3,000-4000 individuals.
Arms
Even distribution of participants across all treatment arms and control groups
Primary outcomes
The primary outcomes capture individuals’ intended behavioral responses to AI and their preferences regarding AI-related regulation and governance. Outcomes span six broad domains: (i) AI adoption and investment, (ii) reskilling and labor-market adjustment, (iii) social interaction and peer substitution, (iv) Data-sharing behavior, (v) trust and fairness of AI based decision making, and (vi) preferences for AI governance and regulation. All outcome variables are measured after treatment exposure.

Cited by

No citation data available for this registration yet.

Keywords