AI-P05 — AI for NGOs and Social Impact

Course Code: AI-P05  |  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

Non-governmental organisations are offered AI as a way to do more with less, which is exactly the pressure they are under and exactly why the offer needs examining. The people an NGO serves are usually those least able to refuse data collection, least able to contest an error, and least likely to be represented in the data a model was trained on. Efficiency gains accrue to the organisation; error costs fall on beneficiaries.

This course is built around that asymmetry. It works through the uses where the case is strong — internal drafting, translation, monitoring, reporting burden reduction, back-office administration — and the uses where the case must be made much more carefully, principally anything involving beneficiary data, targeting, eligibility or vulnerability assessment.

Learners produce a data responsibility assessment and an implementation plan for one real use in their own organisation, together with a do-no-harm analysis and a consent and communication approach appropriate to the population served. The course treats donor pressure to appear innovative as a risk factor to be managed rather than a reason to proceed.

Learning outcomes

On completion, a successful learner will be able to:

1. Distinguish internal-efficiency uses from beneficiary-affecting uses, and apply proportionately different scrutiny.
2. Conduct a do-no-harm analysis for a proposed use, identifying who bears error and who benefits.
3. Apply humanitarian data responsibility principles to design, including minimisation, purpose limitation and retention.
4. Design consent and communication that are meaningful given real power asymmetries and literacy conditions.
5. Assess vendor and donor arrangements for data sharing, secondary use and jurisdictional exposure.
6. Implement a defensible internal use with capability building and a stated review point.

Who this course is for

Programme and country directors, monitoring and evaluation staff, protection and data officers, humanitarian information managers, and foundation and donor staff assessing AI proposals.

Prerequisites

Experience in NGO, humanitarian or development practice. AI-F01 recommended. Learners must bring one real intended use; the assessed work is applied to it.

Syllabus

Module 1 — Two categories, two standards

Focus. Internal-efficiency use where the organisation bears its own error, and beneficiary-affecting use where it does not. Mapping current and proposed use against that division.

Lessons. 1.1 Who bears the error. 1.2 Mapping actual use. 1.3 Donor pressure as a risk factor. 1.4 Setting scrutiny by category.

Core reading. Kristin Bergtora Sandvik, Katja Lindskov Jacobsen & Sean Martin McDonald, “Do No Harm: A Taxonomy of the Challenges of Humanitarian Experimentation”, International Review of the Red Cross 99, no. 904 (2017): 319–344.

Deliverable. Use map with category assignment and a note on any donor-driven pressure identified.

Module 2 — Do no harm, applied to data

Focus. Adapting the do-no-harm discipline to data and model use. Identifying the population that bears error, the pathways by which harm reaches them, and the mitigations that are actually available in the field.

Lessons. 2.1 Harm pathways from data and model error. 2.2 Protection risks from data existence itself. 2.3 Mitigations available in low-resource settings. 2.4 The decision not to collect.

Core reading. Mirca Madianou, “Technocolonialism: Digital Innovation and Data Practices in the Humanitarian Response to Refugee Crises”, Social Media + Society 5, no. 3 (2019). Virginia Eubanks, Automating Inequality (New York: St. Martin’s Press, 2018), chapters 2–4.

Deliverable. Do-no-harm analysis with harm pathways and available mitigations.

Module 3 — Data responsibility in practice

Focus. Minimisation, purpose limitation, retention and deletion, secure handling in the field, and the specific problem of sensitive data about people who cannot safely be identified.

Lessons. 3.1 Minimisation and purpose limitation. 3.2 Retention, deletion and the record you should not keep. 3.3 Field security and device practice. 3.4 Sensitive categories and protection exposure.

Core reading. International Committee of the Red Cross, Handbook on Data Protection in Humanitarian Action, 2nd edition (Geneva: ICRC, 2020), chapters 2–4. United Nations OCHA Centre for Humanitarian Data, Data Responsibility Guidelines (2021).

Deliverable. Data responsibility assessment for the proposed use, with a data flow diagram.

Module 4 — Consent, voice and communication

Focus. Consent under conditions of dependency, communicating what a system does in the languages and literacies of the population served, and complaint and feedback mechanisms that people will actually use.

Lessons. 4.1 Consent under dependency. 4.2 Explaining a system without jargon or false reassurance. 4.3 Language, literacy and channel choice. 4.4 Complaint mechanisms that function.

Core reading. ICRC, Handbook on Data Protection in Humanitarian Action, 2nd edition (2020), chapter on consent. Sandvik, Jacobsen & McDonald, “Do No Harm”, IRRC 99, no. 904 (2017).

Deliverable. Consent and communication plan with sample materials in the working language of the population.

Module 5 — Implementation, vendors and review

Focus. Implementing the defensible use: capability, cost including maintenance, vendor and donor data-sharing terms, jurisdictional exposure, and a stated review point with the option to stop.

Lessons. 5.1 Capability and realistic cost. 5.2 Vendor terms, secondary use and data residency. 5.3 Donor reporting without overclaiming. 5.4 Review point and exit.

Core reading. Ben Ramalingam, Aid on the Edge of Chaos (Oxford: Oxford University Press, 2013), chapters 8–10. OECD, Recommendation of the Council on Artificial Intelligence, OECD/LEGAL/0449 (2019, amended 2024).

Deliverable. Final submission: implementation plan with vendor assessment, cost estimate and review point.

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. Asymmetry awareness: does the analysis identify who benefits and who bears error, without conflating the two?
2. Field realism: are mitigations available in the actual operating environment rather than in principle?
3. Meaningfulness of consent: does the approach account for dependency, language and literacy?
4. Willingness to decline: does the learner identify uses that should not proceed, and say so?

Reading list

Core. International Committee of the Red Cross, Handbook on Data Protection in Humanitarian Action, 2nd edition (Geneva: ICRC, 2020). United Nations OCHA Centre for Humanitarian Data, Data Responsibility Guidelines (2021).

Peer-reviewed. Kristin Bergtora Sandvik, Katja Lindskov Jacobsen & Sean Martin McDonald, “Do No Harm: A Taxonomy of the Challenges of Humanitarian Experimentation”, International Review of the Red Cross 99, no. 904 (2017). Mirca Madianou, “Technocolonialism”, Social Media + Society 5, no. 3 (2019).

Wider reading. Virginia Eubanks, Automating Inequality (New York: St. Martin’s Press, 2018). Ben Ramalingam, Aid on the Edge of Chaos (Oxford: Oxford University Press, 2013).

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. Beneficiary data must not be used in assessed work. Learners work with synthetic or fully anonymised material, and any submission containing identifiable beneficiary information is destroyed and returned unmarked.

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