Balancing Automation and Discretion: How Decision Stakes and Human-AI Collaboration Affect Citizen Perceptions in Public Administration

Paper Title

Balancing Automation and Discretion: How Decision Stakes and Human-AI Collaboration Affect Citizen Perceptions in Public Administration

Publication Info

• Topic area: Citizen perceptions of AI in public administration, focusing on fairness and adoption.
• Keywords: AI in public administration, human-AI collaboration, decision stakes, fairness perceptions, adoption intent, discretionary decisions, hybrid decision-making, procedural fairness, citizen trust, Intelligent Self-Service Kiosk.

Background and Problem

• Problem / challenge: Existing research has studied decision stakes and human-AI decision-making configurations in isolation, leaving a gap in understanding their combined effect on citizens’ perceptions of fairness and willingness to adopt AI in public services.
• Significance: Public administration decisions often carry serious consequences, requiring fairness, empathy, and discretion. Citizens cannot opt out of public services, making trust and legitimacy in AI systems critical.
• Motivation and related work: Prior studies show mixed citizen attitudes toward AI in public services, with preferences for human involvement in high-stakes decisions. However, the specific interaction between stakes and decision-making configurations remains underexplored. This paper addresses this gap by examining how these factors shape perceptions of fairness and adoption.

Solution

• Proposed approach: A mixed-method Wizard-of-Oz study using an Intelligent Self-Service Kiosk (ISSK) to simulate AI-driven public service decisions under varying stakes and decision-making configurations.
• Novelty:
  1. Systematic investigation of the interaction between decision stakes and decision-making configurations.
  2. Mixed-methods approach combining quantitative measures and qualitative thematic analysis.
  3. Insights into citizens’ nuanced reasoning about fairness, empathy, and human involvement in AI-mediated decisions.
• Procedure and key techniques:
  • Two scenarios representing low-stakes (ID renewal) and high-stakes (social housing) decisions were developed in collaboration with civil servants.
  • Three decision-making configurations were tested: fully autonomous AI, AI with human supervision, and AI advising a human.
  • Participants (n=43) interacted with the ISSK, rated fairness and adoption, and participated in semi-structured interviews.
  • Quantitative data were analyzed using repeated measures ANOVA, while qualitative data were analyzed through thematic analysis.

Results

• Concrete findings:
  • Quantitative analysis found no significant effects of stakes or decision-making configuration on fairness or adoption.
  • Qualitative analysis revealed that citizens valued human involvement, especially in high-stakes decisions, and emphasized the need for meaningful, interactive dialogue.
  • Citizens perceived human oversight as symbolic when it lacked visibility or tangible engagement.
• Advantage over baselines:
  • The study highlights the limitations of purely quantitative metrics in capturing citizens’ nuanced perceptions, advocating for mixed-method approaches.
  • Provides actionable insights for designing AI systems that align with democratic values and fairness expectations.
• Experiments / evaluation:
  • Scenarios: ID renewal (low-stakes) and social housing application (high-stakes).
  • Configurations: Fully autonomous AI, supervisory hybrid, and advisory hybrid.
  • Metrics: Procedural, informational, and distributive fairness; adoption intent.
  • Sample: 43 participants with diverse demographics and high AI literacy.
• Limitations and future work:
  • Limited ecological validity due to fictional scenarios and lack of real consequences.
  • Shortened civil servant involvement may have influenced perceptions of oversight.
  • Future work should explore interactive hybrid models, richer dialogue mechanisms, and the role of interpersonal fairness.

Summary

This study investigates how decision stakes and human-AI decision-making configurations shape citizens’ perceptions of fairness and adoption in public administration. While quantitative results showed no significant effects, qualitative findings revealed that citizens value meaningful human involvement, particularly in high-stakes decisions, and emphasize the need for interactive dialogue and empathetic engagement. The study highlights the importance of designing AI systems that balance efficiency with fairness, empathy, and trust. Policymakers and designers should adopt incremental, context-sensitive AI deployment strategies that uphold democratic values and ensure visible, meaningful human oversight.