AI for Transforming Public Service

Part of the Idasara Academy AI Hub — publishing home: academy.idasara.org. Editions: සිංහල · English · தமிழ்.

The dictated flagship of the series, born from training over 10,000 Sri Lankan public officers. From the officer's desk to the national digital plan: what Gen AI actually is, the nine prompting methods, the Single Window redesign of citizen services, the seven risks and the $440,000 Deloitte lesson, and the five golden rules of responsible use. Tasks go to the machine — the post, the judgment, and the accountability stay with you.

Start with the basics: AI for Everyone — the Universal AI Core Kernel — the seven foundation modules every reader shares.

Chapters

1. AI for Transforming Public Service
2. AI for Transforming Public Service: Concepts, Applications, and Guidance for Public Officers
3. From Computing Fundamentals to Generative AI: The Fall of the 70-Year Language Barrier
4. Large Language Models (LLMs), GPTs, and the Architecture
5. The Natural Language Revolution, the Humanities Renaissance, and AI Literacy
6. Clear Articulation, Curiosity, and the Art of Prompting
7. Advanced Prompting Methods and the Divisional Secretariat "Promptathon" Model
8. Single Window Government Architecture, RAG, and Public Finance Auditing
9. Risks of Generative AI, Cybersecurity, and Responsible Use
10. Citizen Service Innovation, the Avukana Model, and Future Systems
11. Conclusion, the 5 Golden Rules, and the Path Forward

Frequently asked questions from public officers

Can a public officer safely use AI at work?

Yes — for the work, never for the secrets. Drafting, summarising, translating and checking are exactly what it is for; citizen files, personal data and unpublished decisions must never enter a public AI tool. This book was built from training over 10,000 officers on precisely that line, and its five golden rules keep you on the right side of it.

Can I put citizen data into ChatGPT?

No — treat this as absolute. A public chatbot is an outside system with a memory: a name, NIC number, file detail or medical fact pasted "just to draft a letter" is a data breach, whatever the intention. Draft with placeholders ("the applicant", "case A"), insert the real details only in your own official system, and you get the speed without the exposure.

How can AI improve citizen services without replacing officers?

By collapsing the running around, not the responsibility. The Single Window pattern in this book lets one counter — one conversation — take a request that today crosses five desks, while the officer supervises the machine that reads, checks and routes. The citizen queues less; the officer stamps more decisions and fewer photocopies.

Will AI make the decisions instead of the officer?

No — and it must never be allowed to. AI prepares: it gathers the file, checks the circular, drafts the recommendation; the officer decides, signs and answers for it. The book's refrain is exactly this: tasks go to the machine — the post, the judgment and the accountability stay with you. An "AI decided it" excuse does not exist in administration.

How can AI help with letters, reports and circulars?

Give it the four-layer brief this book teaches — role, background, task, constraints — and a routine letter becomes a two-minute job: first draft in your language, correct format, ready for your corrections and signature. Summaries of long files and plain-language versions of circulars are the same motion. The nine prompting methods cover every common desk task with worked Sri Lankan examples.

What is the Single Window concept?

One entry point for the citizen instead of a tour of departments: a request arrives once, and the system — AI reading, checking and routing under officer supervision — moves it across the desks the citizen never has to visit. The book treats it as the flagship redesign of citizen services, with the Divisional Secretariat as the worked example.

How do we stop AI mistakes from harming citizens?

Assume the machine will sometimes be confidently wrong, and build the human wall: every AI-drafted fact checked against the file, every recommendation reviewed before signature, nothing auto-sent to a citizen. The book's $440,000 Deloitte lesson shows what skipping verification costs — and its three-layer verification pattern is the discipline that prevents your office becoming the next example.

Do officers need IT training to use AI?

No — the interface is the language you already write minutes in: Sinhala, Tamil or English. If you can brief a junior officer clearly, you can brief an AI; the skill gap is precision of instruction, not technology. The foundation chapters take a non-technical officer from zero to competent prompting in a few evenings.

How do I write a good prompt for administrative work?

Structure it like a file minute: who the AI should be (role), what the situation is (background), exactly what you need (task), and the rules it must respect (constraints — format, length, language, what to leave out). Then iterate: correct its draft the way you would a junior's. The advanced-prompting chapter turns this into the nine methods, from chain-of-thought to the Promptathon drill.

Where should a department start?

Small, visible and supervised: pick one routine burden — a standard letter, a monthly report, a summary queue — run the book's Promptathon model with a handful of officers, and measure the hours returned. Prove it in one unit, write the one-page usage rule, then scale. Transformation that starts with a pilot survives; transformation that starts with a circular usually doesn't.