AI Accountability in Public Institutions

Investigating how governments design, document, and govern AI systems, with a focus on transparency, human discretion, accountability, and contestability in public-sector decision-making.

Research Summary

Governments around the world are increasingly adopting artificial intelligence to support decisions in immigration, healthcare, scientific research, law enforcement, public benefits, and administrative services. This research program examines how public institutions govern these systems through AI registers, inventories, impact assessments, procurement practices, and organizational processes. By studying the infrastructures surrounding AI adoption rather than AI models alone, I investigate how transparency, accountability, and human discretion are enacted—and where important gaps remain.

Why this research?

As artificial intelligence becomes embedded within governments, universities, and other public-serving institutions, questions of accountability extend well beyond algorithmic performance. AI systems are implemented through bureaucratic procedures, documentation practices, procurement processes, organizational routines, and professional judgment, making their impacts inseparable from the institutions in which they operate.

This research program investigates AI governance as a sociotechnical and institutional challenge. Rather than treating governance as a set of technical safeguards or policy requirements, I examine how accountability is enacted through administrative practice: how organizations document AI systems, translate governance principles into operational processes, exercise human discretion, allocate responsibility, and respond to uncertainty.

Across studies of AI registers, algorithmic impact assessments, and AI-supported decision making in immigration systems, homelessness services, and higher education institutions, my work examines how governance infrastructures shape what becomes visible, what remains hidden, and how individuals experience AI-enabled decision making.

Across this project, I investigate questions such as:

  • How do public institutions implement and govern AI in practice?
  • How do documentation, disclosure, and impact assessment shape algorithmic accountability?
  • How do bureaucratic routines, procurement, and organizational capacity influence AI governance?
  • How is human judgment redistributed through human-AI collaboration?
  • How do people experience, interpret, and contest institutional AI systems?

Research Approach

This research combines qualitative, computational, ethnographic, and critical approaches to examine AI governance across public institutions and other organizational settings.

Methods include:

  • Large-scale analyses of AI registers and inventories
  • Computational audits of deployed institutional AI systems
  • Ethnographic and qualitative studies of frontline work
  • Mixed-method analyses of public discourse and collective sensemaking
  • Critical analyses of documentation, impact assessments, and governance artifacts
  • Sociotechnical theories of infrastructure, bureaucracy, classification, and accountability

Rather than evaluating AI systems in isolation, I investigate the institutional ecosystems through which they are designed, implemented, documented, governed, and experienced.

Research Outcomes and Contributions

This research contributes to several areas within Human-Computer Interaction and CSCW.

  • Documentation and Accountability in Institutional AI Governance: I study how governments and public-serving organizations adopt, implement, and govern AI systems through organizational routines, procurement practices, documentation, and administrative decision making. Rather than viewing governance as a purely technical problem, I examine how accountability emerges through institutions, professional judgment, bureaucratic practice, and human discretion and what becomes visible, remains obscured, or is contestable by respective accountasbility measures. (See FAccT 26, HCOMP 26)

  • Human-AI Collaboration in Administrative Decision Making: AI systems increasingly support rather than replace professional decision makers. I examine how human judgment, expertise, and discretion are redistributed across institutional workflows and how these collaborations influence accountability. Rather than treating human oversight as a procedural safeguard, this work conceptualizes it as an organizational practice shaped by institutional values, expertise, workload, and uncertainty. (See CHI 25, HCOMP 26, COMPASS 26)

  • Fairness Audits of and Lived Experiences around Institutional AI: Fairness is produced by a series of processes, e.g., algorithms, data collection, organizational priorities, post-processing decisions, and resource allocation. My work develops computational methods for auditing deployed institutional AI systems while situating inequalities within broader sociotechnical contexts. I also study how institutional accountability is experienced those affected by AI systems by examining how people interpret opaque decisions, engage in collective sensemaking, and navigate algorithmic governance in practice. (See CanadianAI 26, COMPASS 26)

Current Directions

This research program continues to expand in several directions:

  • Comparative analyses of AI registers across countries
  • Governance of generative AI in the public sector
  • Human-AI collaboration in administrative decision making
  • Documentation standards for trustworthy AI
  • Public participation and contestability in AI governance
  • Comparative studies of algorithmic accountability infrastructures