Understanding Citizens’ Value Position of AI-Based Decision Making in Government

• In Event: Poster Session: Science, Technology, and Environmental Policy 2 & Comparative Education Reform
  In Poster Session: Poster Session 2: Science, Technology, and Environmental Policy

Abstract

As the deployment of automated decision systems (ADS) with artificial intelligence (AI) is increasingly gaining traction within the public sector, the question of how they impact public values has become a central issue. Yet, there is a limited understanding and lack of empirical studies on citizen perceptions concerning AI governance.
To fill this intellectual gap, this paper investigates the value preferences and attitudes of citizens towards automated decision-making applications in public service provision. We combine the public values embedded in the public administration literature and derive three value positions that are highly relevant to the AI concerns: traditional public administration (i.e., accountability, transparency, equity), new public management (i.e., effectiveness, efficiency), and new public service (i.e., service quality, privacy protection). Knowing how much citizens care about different public values is critical for e-government managers, but existing measurement approaches have significant limitations. Prior research often identifies an inventory of public values yet fails to assess their relative importance or relevance in specific contexts.
We employ a survey-experimental, choice-based approach that simultaneously tests the influence of seven AI system attributes in generating support for the usage of public sector algorithms. The conjoint experiment will be embedded in an online survey of U.S. citizens administered through CloudResearch, a survey platform known for high data quality. The sample includes quotas (e.g., race, political party, education) to achieve representativeness of the U.S. population and improve the external validity of the study. The survey will provide 7,200 evaluated profiles from 1,200 respondents.
Using prominent real-world ADS applications in healthcare eligibility and criminal justice, the preliminary experiment results reveal that it is most salient to the public that AI systems should be effective, transparent, and accountable. Moreover, we expect that citizens from different racial groups may react differently to the equity issue of potential algorithmic bias. Political ideology and personal relevance may also moderate citizens’ value priorities. These findings would provide implications on how to communicate with the public when policymakers propose AI initiatives. Knowing and integrating dominant value preferences into AI governance would also help e-government managers deliver public services that are better aligned with citizens’ needs and expectations, ultimately leading to higher levels of trust and satisfaction among the public.

Authors

• Guimin Zheng, Florida State University

• Yixin Liu, Northern Arizona University

• Heewon Lee, Florida State University