(iPoster) AI, Accountability, and Blame Avoidance in Public Service

• In Event: Poster Sessions: Public Administration & Political Organizations and Parties
  In Poster Session: Poster Session: Public Administration

Abstract

Artificial intelligence (AI) is increasingly used in government to support screening, eligibility review, and case adjudication. Existing discussions often frame AI adoption as a technical capacity issue or as an ethics-and-compliance challenge addressed through “human-in-the-loop” oversight. This paper advances a different argument: public officials may prefer AI not only because it increases efficiency, but because it enables blame avoidance. AI can provide procedural cover, shift responsibility, and reduce personal exposure when policy outcomes become politically contested.
We test this claim using a conjoint experiment with Taiwanese public officials. Respondents are asked to imagine that their agency has launched a targeted allowance/subsidy program, and that they serve as case officers responsible for processing applications and verifying eligibility. The policy is politically salient and time-sensitive, as agency leadership frames it as a key “governance highlight” ahead of an election and demands rapid, visible results. The agency has implemented an AI-assisted review system.
The conjoint randomizes core features that should shape blame avoidance incentives. These include 1)AI involvement (data organization only; recommendations requiring human confirmation; abnormal-case screening lists; and full automation generating approval/rejection outcomes), 2) potential harms (privacy leakage; algorithmic discrimination; and opacity), 3) expected benefits (speed; consistency/error reduction; reduced administrative and psychological burden), accountability assignment (case officer; system designer; shared responsibility), 4) external visibility (no disclosure; basic disclosure; proactive disclosure with appeal and human review), and 5) the strength of the agency’s adoption push (discretionary; soft encouragement; incentives; mandates).
We hypothesize that blame avoidance will be most pronounced when higher AI involvement reduces individual liability, especially under shared or designer-centered accountability, and when organizational pressure prioritizes rapid delivery. Conversely, high perceived harms and high external visibility should reduce support for automation by increasing the likelihood that errors trigger public controversy rather than shielding officials from blame. We aim to further the understanding of bureaucratic accountability and delegation by showing when AI is treated as administrative capacity versus when it becomes a strategic tool for managing political and organizational risk.

Authors

• Li-Yin Liu, University of Dayton

• Unregistered Participant