Algorithmic Empowerment or Algorithmic Control? A Survey Experiment on AI in Energy Rebate Take-Up

• In Event: Governing Clean Technology Diffusion Across Society: From Firm Innovation to Household Adoption

Abstract

Energy efficiency rebates are commonly used to encourage citizens to adopt more energy-efficient products (Allcott 2016; Gillingham et al. 2018). Yet claiming these subsidies requires navigating an application process, and prior work shows that the burdens involved can discourage eligible citizens from applying (Madsen et al. 2022; Giannella et al. 2024; Graff 2025). As governments increasingly incorporate AI into public service delivery, they face a fundamental design choice between citizen-facing and admin-facing configurations (Vogl et al. 2020). Citizen-facing AI assists applicants and can empower them by reducing the perceived costs of navigating services (Alishani & Homburg 2025; Aoki 2020). Admin-facing AI supports administrative decision-making but may introduce opacity and raise concerns about fairness and control (Aoki, 2021; Orbán & Stefkovics, 2025; Alon-Barkat & Busuioc, 2023). We refer to these contrasting possibilities as algorithmic empowerment and algorithmic control. Yet we know little about whether and how such design choices translate into differences in administrative burden and willingness to apply for public programs. 

To examine this question, we conducted a preregistered survey experiment with 300 U.S. adults recruited through a quota-based online panel. The sample approximated key demographic characteristics of the U.S. population. We used a 2×2 factorial design embedded in a local energy-rebate application scenario. Respondents were randomly assigned to one of four conditions varying whether AI assisted applicants in navigating the application (citizen-facing AI: present or absent) and whether AI assisted administrators in reviewing eligibility (admin-facing AI: present or absent). Program rules and eligibility criteria were held constant across conditions. After reading the scenario, respondents reported their willingness to submit an application. We also measured administrative burden across three dimensions (learning, compliance, and psychological costs), perceived autonomy, and trust in government as potential mechanisms. 

Results reveal a clear contrast between citizen-facing and admin-facing AI. Citizen-facing AI significantly reduces psychological costs and administrative burden, whereas admin-facing AI significantly increases administrative burden among respondents with higher levels of perceived life control. Yet these differences in burden do not translate into significant average behavioral effects. Mediation analysis shows that citizen-facing AI increases willingness to apply through a statistically significant indirect effect via reduced administrative burden. However, neither citizen-facing nor admin-facing AI produces a significant average direct effect on willingness to apply. These results suggest that AI shapes behavior primarily by reshaping citizens’ burden experience rather than through direct motivational change. 

This study contributes to research on energy efficiency policy by showing that the effectiveness of rebate programs depends not only on financial incentives but also on how application processes are structured. It also contributes to research on AI in public service delivery by demonstrating that different AI configurations shape citizens’ experiences of administrative burden through distinct mechanisms. The findings provide evidence of algorithmic empowerment while raising important questions about algorithmic control, calling for future research to examine how specific AI configurations shape citizens’ experience across different policy areas.

Author

• YU XIAO, Florida State University
  Presenting Author