AI-P02 — AI for Government and Public Policy

Course Code: AI-P02  |  School: School of Artificial Intelligence  |  Cluster: Level 3 — Applied & Sectoral Practice

Level: Intermediate  |  Duration: 5 weeks · 18–22 learning hours  |  Language: English  |  Certificate: Professional Certificate (non-degree)  |  Format: Self-paced with AI support under human supervision

Overview

When a private firm deploys a flawed model, customers suffer. When a state does, citizens lose entitlements, are wrongly investigated, or cannot appeal. The asymmetry is the reason this course treats public sector AI as a question of administrative law and legitimacy before it is a question of efficiency.

The material covers three settings distinctly, because their risk profiles differ. Internal analytical use, where a policy team drafts, summarises and models, carries mainly quality risk. Citizen-facing service delivery carries access and equity risk. Decision support in eligibility, enforcement and risk scoring carries the possibility of unlawful administrative action, and is treated most stringently.

Learners produce an impact assessment for a real or proposed deployment in their own institution, including a lawfulness analysis, an equity analysis, a design for contestability, and a public transparency statement. The course insists that a citizen must be able to find out that an automated system was involved and challenge the outcome.

Learning outcomes

On completion, a successful learner will be able to:

1. Distinguish internal analytical use, service delivery and decision support, and apply proportionate scrutiny to each.
2. Analyse the lawfulness of a proposed use against administrative law duties and the applicable data protection instrument.
3. Assess distributional impact, identifying who bears the cost of error and whether that burden is already unequally held.
4. Design contestability: notification, explanation, human review and appeal that a citizen can actually use.
5. Specify procurement requirements and audit rights adequate for public accountability.
6. Write a transparency statement in language a non-specialist citizen can understand.

Who this course is for

Civil servants and public managers, policy analysts, agency legal and data protection officers, ombudsman and audit institution staff, and advisers to government.

Prerequisites

Experience of public administration or policy work. AI-F01 recommended, AI-G01 useful. Learners should have a real or credibly proposed deployment to analyse.

Syllabus

Module 1 — Three settings, three standards

Focus. Why the same technology warrants different scrutiny depending on whether it informs an analyst, serves a citizen, or determines an entitlement. Mapping current use in your own organisation.

Lessons. 1.1 Internal analysis, service delivery, decision support. 1.2 Mapping actual use, including unsanctioned use. 1.3 Setting proportionate scrutiny. 1.4 The discretion that is quietly removed.

Core reading. Karen Yeung & Martin Lodge (eds), Algorithmic Regulation (Oxford: Oxford University Press, 2019), introduction and the chapter by Michael Veale & Irina Brass on public sector machine learning. Mark Bovens & Stavros Zouridis, “From Street-Level to System-Level Bureaucracies”, Public Administration Review 62, no. 2 (2002): 174–184.

Deliverable. Use map for your organisation, with each use classified and scrutiny level assigned.

Module 2 — Lawfulness and administrative duty

Focus. Legal basis for processing, the duty to give reasons, the prohibition on fettering discretion, and the requirement that a decision-maker actually decide rather than ratify an output.

Lessons. 2.1 Legal basis and purpose limitation. 2.2 The duty to give reasons. 2.3 Fettering discretion by system design. 2.4 Automated decision-making provisions.

Core reading. Undang-Undang No. 27 Tahun 2022 tentang Pelindungan Data Pribadi (Indonesia), or the applicable data protection instrument. Regulation (EU) 2024/1689 (Artificial Intelligence Act), annex III on high-risk public sector uses, read comparatively. Danielle Keats Citron & Frank Pasquale, “The Scored Society: Due Process for Automated Predictions”, Washington Law Review 89, no. 1 (2014): 1–33.

Deliverable. Lawfulness analysis for one proposed use, with the legal basis and reason-giving route identified.

Module 3 — Who bears the error

Focus. Distributional analysis of false positives and false negatives, the documented pattern by which automated eligibility and risk systems concentrate harm on the least able to appeal, and how to measure it before deployment.

Lessons. 3.1 False positives, false negatives, and who absorbs each. 3.2 Documented harms in eligibility and enforcement systems. 3.3 Measuring distributional impact in advance. 3.4 Deciding not to deploy.

Core reading. Virginia Eubanks, Automating Inequality: How High-Tech Tools Profile, Police, and Punish the Poor (New York: St. Martin’s Press, 2018). Ziad Obermeyer et al., “Dissecting Racial Bias in an Algorithm Used to Manage the Health of Populations”, Science 366, no. 6464 (2019).

Deliverable. Distributional impact assessment with the error burden identified by group.

Module 4 — Contestability by design

Focus. Notification that the citizen will actually receive, explanation they can use, human review with genuine authority to overturn, and an appeal route with published timescales.

Lessons. 4.1 Notification that reaches the person. 4.2 Explanation that supports a challenge. 4.3 Human review with real authority. 4.4 Appeal, timescales and remedy.

Core reading. Citron & Pasquale, “The Scored Society”, Washington Law Review 89 (2014). Helen Margetts & Cosmina Dorobantu, “Rethink Government with AI”, Nature 568 (2019): 163–165.

Deliverable. Contestability design with notification wording, explanation template and appeal procedure.

Module 5 — Procurement, transparency and public account

Focus. Contract requirements adequate for public accountability, publishing enough for meaningful scrutiny without compromising security, and answering to an audit institution or legislature.

Lessons. 5.1 Contract terms: documentation, audit rights, data residency, exit. 5.2 Publication registers and what to disclose. 5.3 Writing for citizens, not for lawyers. 5.4 Answering the audit institution.

Core reading. OECD, “Hello, World: Artificial Intelligence and its Use in the Public Sector”, OECD Working Papers on Public Governance No. 36 (Paris: OECD Publishing, 2019). OECD, Recommendation of the Council on Artificial Intelligence, OECD/LEGAL/0449 (2019, amended 2024).

Deliverable. Final submission: impact assessment, procurement annex and public transparency statement in plain language.

Assessment

Pass mark 70 per cent. All assessed components must be attempted. Every mark in this course is issued by a human assessor; no assessment outcome is generated automatically.

Rubric criteria

Each assessed artefact is marked against four criteria at four levels (distinction, pass with merit, pass, fail).

1. Legal accuracy: is the analysis grounded in the actual applicable instrument rather than in general principle?
2. Distributional seriousness: does the assessment identify who bears error, with evidence rather than assurance?
3. Usability of contestability: could an ordinary citizen, without a lawyer, learn of and challenge the decision?
4. Public intelligibility: is the transparency statement genuinely readable by a non-specialist?

Reading list

Core. Karen Yeung & Martin Lodge (eds), Algorithmic Regulation (Oxford: Oxford University Press, 2019). Virginia Eubanks, Automating Inequality (New York: St. Martin’s Press, 2018).

Peer-reviewed. Danielle Keats Citron & Frank Pasquale, “The Scored Society: Due Process for Automated Predictions”, Washington Law Review 89, no. 1 (2014). Mark Bovens & Stavros Zouridis, “From Street-Level to System-Level Bureaucracies”, Public Administration Review 62, no. 2 (2002). Helen Margetts & Cosmina Dorobantu, “Rethink Government with AI”, Nature 568 (2019). Ziad Obermeyer et al., “Dissecting Racial Bias…”, Science 366 (2019).

Policy and legal instruments. OECD, “Hello, World: Artificial Intelligence and its Use in the Public Sector”, OECD Working Papers on Public Governance No. 36 (2019). OECD, Recommendation of the Council on Artificial Intelligence (2019, amended 2024). Regulation (EU) 2024/1689. Undang-Undang No. 27 Tahun 2022 tentang Pelindungan Data Pribadi (Indonesia).

All items are published works identifiable by author, title and publisher. Learners obtain them through an institutional library or the publisher. The Academy does not distribute copyrighted texts.

Academic integrity and use of AI

Generative tools may be used in producing assessed work under three conditions. Use must be disclosed in a short statement appended to each submission, naming the tool and the task it performed. Any factual or technical claim originating from a generative tool must be verified against a citable source before it enters assessed work, and the verification must be evidenced. The analytical judgement in each artefact must be the learner’s own and must be defensible in a short follow-up. Where the honest conclusion of your impact assessment is that the deployment should not proceed, that conclusion is a pass. Assessments that reach a favourable finding by narrowing scope will not be credited.

Instructor: pending owner confirmation. Pricing: pending owner approval. Reference list verified against publisher records; any later addition is marked for verification before publication.