CoChat

The Administrative State and Artificial Intelligence: Toward an Internal Law of Administrative Algorithms

Open access
Haim, Amit · 2023 · SSRN Electronic Journal

The DOI matches its Crossref record; 28 of the 28 references it cites were found in a registry or an index.

Abstract

The administrative state is gradually embracing artificial intelligence (AI) algorithms. The advent of the so-called automated state has raised concerns over accountability, transparency, and fairness and engendered proposals for legal regulation. Yet the notion that algorithms are not merely technical instruments but encode social behavior embedded in a bureaucratic context has largely been missing from the discourse. This article identifies algorithmic as sociotechnical systems embedded in an organizational context, which can function as bureaucratic governance instruments. It argues that external legal institutions, whether legislative endeavors or judicial review, lack the capacity, insight, and perspective to achieve meaningful accountability in reviewing the administrative use of AI algorithms. The article suggests moving beyond seeing algorithms as a distinct phenomenon to regulate, to a more holistic view of the bureaucratic challenge, which entails confronting inconsistencies, inaccuracies, and administrative culture, and taking account of institutional design. By doing so, it follows in the footsteps of scholars of internal administrative law, comprising of the complex set of rules, guidelines, and procedures of the bureaucracy, and highlighting internal governance as means of improving outcomes and ensuring accountability. It also discusses the points of contention and outlines the doctrinal questions of administrative law that are likely to occupy judges and lawyers when dealing with internal governance with AI in the government. Concurrently, this article uncovers the internal law of administrative algorithms, which is emerging from a set of informal documents developed in the Federal government. The article reviews this corpus and distills the main tenets of responsible and trustworthy AI that will guide how administrative agencies design and implement their AI systems in the foreseeable future, and imbue general principles with actionable goals.

Overview

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.2139/ssrn.4498470

Cites

28 of 28 references verified0 citation errors · 0 not checkable

Statistics

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Show all 28 references

Research works printed but not linked: 79

  1. Engstrom Et, under the Quick Disability Determination program, and using clustering for micro-specialization
  2. 2019, For example, assisting screening of hotline calls in Child Protection Services, ALLEGHENY METHODOLOGICAL REPORT
  3. Gail S Ennis, 2019, The Social Security Administration's Use of Insight Software to Identify Potential Anomalies in Hearing Decisions
  4. J Steven, 2022, Responding to Mass, Computer-generated and Malatributed Comments, 74 ADMIN, L. REV, 95
  5. See Generally, supra note 2; Coglianese and Ben Dor, supra note 2. 45 Such as prior art search at the USPTO. ENGSTROM ET AL, 48
  6. 2021, For example, for medical analysis in the Department of Veteran Affairs, See VA ARTIFICIAL INTELLIGENCE (AI) STRATEGY, US DEPARTMENT OF VETERANS AFFAIRS
  7. 2018, For a skeptical voice documenting some of these use-cases, see VIRGINIA EUBANKS, AUTOMATING INEQALITY: HOW HIGH-TECH TOOLS PROFILE, POLICE, AND PUNISH THE POOR
  8. 2020, In the Netherlands, a government was ousted due to a scandal that arose under similar circumstances, where an automated system wrongly accused dozens of thousands of families of fraud in child support benefits, Welfare surveillance system violates human rights, Dutch court rules, 22
  9. Sandra G Mayson, 2019, Bias In, Bias Out, 128 YALE LAW JOURNAL 2218
  10. Rashida Richardson, 2019, Bad Predictions: How Civil Rights Violations Impact Police Data, Predictive Policing Systems, and Justice, L. REV
  11. 1990, The field of science and technology studies (STS) has long advocated the notion that technology is not a foreign object but rather part of a wider social and political context, Technology is Society Made Durable
  12. Rebecca Crootof, 2023, Humans in the Loop, 76 VANDERBILT L. REV
  13. Christian Leibig, 2022, Combining the strengths of radiologists and AI for breast cancer screening: a retrospective analysis, 4 THE LANCET DIGITAL HEALTH e
  14. Avi Gesser, 2022, When Humans and Machines Disagree -The Myth of "AI Errors" and Unlocking the Promise of AI Through Optimal Decision Making -Debevoise Data Blog, 16
  15. Sherman Ho, supra note 16; Ho, supra note 119; HO, MARCUS, AND RAY, supra note 11. 118 Ho and Sherman, 16, 254
  16. Marcus Ho, AI can help agencies shift from a purely ex-post model of quality review and toward continuous assessment. Relatedly, Insight reflects the potential for AI to promote continuous learning for actors across the system
  17. 2002, See Levin, supra note 153
  18. Alexander Nabavi-Noori, 2022, Agency Control and Internally Binding Norms, 131 YALE L, J
  19. RECOMMENDATION 2017-5: AGENCY GUIDANCE THROUGH POLICY STATEMENTS
  20. For the importance of transparency with algorithms for the purposes of public legitimacy. See Amit Haim & Dvir Yogev, Perceived Algorithmic Legitimacy in the Administrative State
  21. Ari Waldman, 2022, Governing algorithmic decisions: The role of decision importance and governance on perceived legitimacy of algorithmic decisions, 9 BIG DATA & SOCIETY 20539517221100450
  22. Engstrom, Algorithmic Accountability, supra note 21, 846. 176 Engstrom and Haim
  23. Thomas W Merrill, 1954, See Magill, supra note 125 at 873-882, L. REV, 569
  24. For example, through good guidance practices
  25. 1980, SSA ALJ alleged invasion of statutory right to decisional independence upon receipt of "mandatory, unlawful instructions regarding every detail of their judicial role
  26. 1953, Federal Trial Examiners Conference, 345 US, 128
  27. U S C � � 554, 2016, 557. algorithm, that is calibration, false negative equity, and false positive equity. This means there cannot be a fully fair algorithms according to mathematically definable notions of fairness, Sendhil Mullainathan & Manish Raghavan, Inherent Trade-Offs in the Fair Determination of Risk Scores, 556
  28. See Ben Green, 2022, The flaws of policies requiring human oversight of government algorithms, 45 COMPUTER LAW & SECURITY REVIEW
  29. Lilian Edwards, 2022, Regulating AI in Europe: four problems and four solutions
  30. Engstrom and Ho started laying out the foundations of administrative law for algorithms. see Engstrom and Ho
  31. 2020, 216 Engstrom and Haim, supra note 8. See also EXAMINING THE BLACK BOX
  32. Andrew Tutt, 2017, An FDA for Algorithms, 69 ADMIN, L. REV, 83
  33. 2021, Itay Ravid & Amit Haim, Progressive Algorithms, 12 UC IRVINE L. REV, 527
  34. E T Dillon, 2018, ALGORITHMIC IMPACT ASSESSMENTS: A PRACTICAL FRAMEWORK FOR PUBLIC AGENCY ACCOUNTABILITY
  35. 2016, There is a blind spot in AI research, NATURE, 538
  36. Min Kyung, WeBuildAI: Participatory Framework for Algorithmic Governance, 3 PROC
  37. Justin B Matthew M Young, 2019, Artificial Discretion as a Tool of Governance: A Framework for Understanding the Impact of Artificial Intelligence on Public Administration, PERSPECTIVES ON PUBLIC MANAGEMENT AND GOVERNANCE gvz, AI & SOC, 29
  38. A L Reisman, supra note 63; JOSHI, supra note 4 at, 21
  39. Engstrom and Haim call this the "snapshot" and "bookend" problems. See Engstrom and Haim
  40. Robert A Wilson ;, 2017, supra note 12; MASHAW, supra note 12, Varieties of Bureaucratic Justice: Building on Mashaw's Typology
  41. Paul Verkuil, 2017, Meeting the Mashaw Test for Consistency in Administrative Decision-Making, in ADMINISTRATIVE LAW FROM THE INSIDE OUT 239
  42. Edelman, supra note 12; DiMaggio and Powell
  43. For an early antecedent see R. Shep Melnick, Administrative Law and Bureaucratic Reality
  44. L Rev ; Christopher, 1992, Put differently, federal agencies regulate us in many meaningful, and sometimes frightening, ways that either evade judicial review entirely or are at least substantially insulated from such review, Nicholas R Parrillo, Federal Agency Guidance and the Power to Bind: An Empirical Study of Agencies and Industries, 36, 1624
  45. 2019, Article 22 of the GDPR restricts fully automated decisions on data subjects. For exploration of this and other aspects see Margot E Kaminski, The Right to Explanation, Explained, 34 BERKELEY TECH, L. J
  46. 2019, For example, the New York City special task force on automated decision-making spent arduous efforts in coming up with a definition that will include all relevant automated and algorithmic systems, but will preclude banal instances such as use of spreadsheets like Excel, New York City Automated Decision Systems Task Force Report, 26
  47. Green, 2018, ALGORITMIC DECISION SYSTEMS
  48. Id, at 1917-1927; Ho and Xiang, supra note 56
  49. Heckler V Chaney, 1985, Engstrom, supra note 2; Solow-Niederman, supra note 21, 470
  50. Kate Crawford, 2019, Crawford and Schultz, supra note 53. 279 Engstrom and Haim, supra note 8; Glaze et al., supra note 11
  51. 2020, Executive Order 13960: Promoting the Use of Trustworthy Artificial Intelligence in the Federal Government, 85 FED, 78
  52. Id, 283 Agency inventories of AI use cases
  53. Executive Order 13960, supra note 289
  54. Russel T Vought, 2020, Guidance for Regulation of Artificial Intelligence Applications, 12
  55. 2018, ARTIFICIAL INTELLIGENCE: EMERGING OPPORTUNITIES, CHALLENGES, AND IMPLICATIONS
  56. BLUEPRINT FOR AN AI BILL OF RIGHTS: MAKING AUTOMATED SYSTEMS WORK FOR THE AMERICAN PEOPLE
  57. 2023, The Department of Education has begun a process of developing policies focused on effective, safe, and fair AI applications for education, Department of Veteran Affairs VA ARTIFICIAL INTELLIGENCE (AI) STRATEGY, supra note 45. The Department of Energy DOE AI Risk Management Playbook (AIRMP), ENERGY.GOV
  58. Ai (tai) Trustworthy, 2021, The Playbook specifically incorporates principles from EO13960 and the OMB Memo
  59. Safe And Effective Systems 2. Algorithmic Discrimination Protections 3. Data Privacy
  60. Notice And Explanation 5. Human Alternatives, Consideration, And Fallback OMB Memo 1. Public Trust 2. Public Participation 3. Scientific Integrity and Information Quality, Risk Assessment and Management, 4
  61. Interagency Coordination GAO Framework 1. Governance: Promote accountability by establishing processes to manage, operate, and oversee implementation 292 Id, Fairness and Nondiscrimination 8. Disclosure and Transparency 9. Safety and Security 10
  62. A I Doe, Risk Management Playbook (AIRMP), supra note 193
  63. Id, 317 Engstrom and Haim, supra note 8
  64. 2023, See supra note 69 at 93. In fact, the consistency of algorithms is also problematized with more advanced methods, where algorithmic outputs are sampled from a large probability distribution and may differ from query to query, The Role of Temperature in Large Language Model Response Generation
  65. Leibig, supra note 85; Raisch and Krakowski, supra note 85
  66. Gao Report, supra note 23; VA ARTIFICIAL INTELLIGENCE (AI) STRATEGY, supra note 45; NAIAC REPORT, supra note
  67. Schultz Crawford, Based in part on Engstrom and Haim, 325
  68. Id, at 61. This is reminiscent of protracted battles between the Social Security Administration and the ALJ union over decisional independence
  69. Raso, 2020, supra note 30; Raso, supra note 104; Rik Peeters, The agency of algorithms: Understanding humanalgorithm interaction in administrative decision-making, IP, 25
  70. Ames, supra note 16 at 61-62. This is common in situations where subordinates evaluate their superiors
  71. Hausman, supra note 140; Hausman, supra note 15
  72. Methodological and Informational Transparency
  73. Id, 65; Bednar, supra note
  74. Engstrom, supra note 3; Engstrom and Haim, supra note 8
  75. Cary Coglianese, 2019, Public Availability of Agency Guidance Documents
  76. Christie See, 2022, A recent whitepaper from researchers at Stanford University found the level of transparency of federal agencies regarding AI to be partial and lacking, Implementation Challenges to Three Pillars of America's AI Strategy
  77. Metzger has argued similarly regarding agency decision-making more generally, 22, 1366
  78. 2023, Artificial Intelligence (AI) at National Science Foundation
  79. John C Jeffries, 2007, Structural Reform Revisited, 95 CAL. L. REV. 1387

Cited by

Metadata

Authors
Haim, A.
Affiliations
Max Planck Society (DE)
Corresponding author
Haim, Amit
Venue
SSRN Electronic Journal
Publisher
Elsevier BV
Published
2023
ISSN
1556-5068 (ISSN-L)
OpenAlex
W4385322751
Cited by
3
References
24
Citations by year
2026: 2 · 2025: 1
Type
posted content
Language
en
Open access
Locations
  • SSRN Electronic Journal (acceptedVersion)
  • eScholarship (California Digital Library) (submittedVersion)
  • eYLS (Yale Law School) (submittedVersion)
Topics
  • Social Sciences › Law › Administrative Law and Governance
  • Physical Sciences › Computer Science › Artificial Intelligence › Ethics and Social Impacts of AI
  • Social Sciences › Law › Digitalization, Law, and Regulation
Keywords
artificial intelligence in government, internal governance, bureaucratic governance, accountability, transparency, fairness, administrative agencies, institutional design, administrative culture

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.