From Audit to Algorithm: Ethical Challenges of AI Inclusion in Public Tax Administration
Bibliographic record
Abstract
Globally, taxes are the unarguable lifeblood of a government body, instrumental in providing essential revenue to fund public welfare programs, building and maintaining critical infrastructure, and social welfare programs critical to societal stability and progress.[1] For decades, the public tax administration relied on human tax auditors to review returns, conduct interviews, and apply judgment within legal boundaries [2].The fast-paced adoption of artificial intelligence (AI) in public tax administration resulted in unprecedented efficiency in revenue collection, fraud detection, and compliance monitoring [3][4].A fast-changing department with some level of resistance to the change-from traditional human-led audits to algorithm-driven decision systems-it also raises profound ethical questions [5].This article examines the transition through three lenses: fairness and bias, transparency and accountability, and privacy versus surveillance.Drawing on a few global case studies from the Netherlands, Canada, and India [6][7][8], this paper argues that while AI can bring efficiencies via reducing administrative costs and close taxation gaps and targets, unanswered ethical risks threaten public trust and democratic legitimacy [9].Artificial intelligence (AI) promises transformative efficiency in tax administration, yet its deployment risks amplifying bias, eroding privacy, and undermining public trust if not guided by rigorous ethical safeguards.This paper proposes a policy framework rooted in human-centered values, fairness, transparency, robustness, and accountability-aligned with ISO/IEC 42001:2023 [11] and ISO/IEC 22989:2022 [12]to ensure AI serves taxpayers equitably while enhancing compliance and operational integrity.[11][12] Drawing on U.S. federal findings, international standards, and practical tools such as AI impact assessments, threat modeling (e.g., STRIDE), and the TRUST principles (Fairness, Accountability, Transparency, Privacy, Inclusivity), the framework outlines a lifecycle-based governance model tailored to tax contexts.[13][14] Key recommendations include mandatory bias audits, human-in-the-loop oversight for highstakes decisions, public model registries, and regulatory sandboxes for low-risk testing.[15][16] An implementation roadmap with phased milestones and measurable KPIs demonstrates feasibility, illustrated through global benchmarks from Sweden, Australia, and Brazil.[17][18][19] By embedding these principles, tax authorities can harness AI's potential to reduce administrative burdens, minimize disparate impacts, and foster societal trust in digital governance.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.095 | 0.127 |
| Meta-epidemiology (narrow) | 0.000 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.002 | 0.002 |
| Science and technology studies | 0.013 | 0.072 |
| Scholarly communication | 0.026 | 0.019 |
| Open science | 0.003 | 0.011 |
| Research integrity | 0.015 | 0.024 |
| Insufficient payload (model declined to judge) | 0.002 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".