What AI can and cannot do for public services

Public institutions are asked to serve more people every year with the same staff and budget. AI tools promise relief, and some of that promise is real. But public services carry a weight that a private company does not: their decisions touch citizens’ rights, money, health and safety. So the useful question is where AI helps, and where it must never be the one deciding.

Where AI helps

AI tools do well on work that involves a lot of text and has a result a person can check. In a ministry or a municipal office that includes:

• Drafting letters, notices and reports for an officer to review and sign
• Summarising long documents, meeting minutes and public consultation responses
• Answering routine questions, such as opening hours, required documents or how to apply, with a clear route to a person when the question is not routine
• Organising and searching records so staff find a file in minutes rather than days
• Translating information between Somali and English so more people can understand it, with a person reviewing anything official

In every case a public servant stays responsible for the result. AI saves time on preparation; people check, decide and sign.

Where AI must not decide

Some decisions should never be handed to a machine: who qualifies for a benefit or a service, who gets a permit or a licence, who is fined or penalised, and anything touching health, safety or legal status. AI tools can be wrong in ways that are hard to see, and they repeat whatever unfairness was in the material they learned from. A citizen on the receiving end of a decision deserves an official who understands their situation and can explain the reasons.

AI can still support that work, by organising a case file or drafting a summary for the officer, but the decision and the reasons belong to an accountable person.

Accountability stays with the institution

When a public body uses AI, it is responsible for the outcome, full stop. That means knowing which tools are in use and for what, keeping a record of how AI contributed to important documents, and being able to explain a decision to a citizen, an auditor or a court. “The system said so” is never an acceptable answer.

Build the oversight in first

Oversight works when it is designed in from the start. Before any tool is used with the public:

• Name an official who is responsible for the tool and what it produces
• Test it on real cases, in Somali and in English, including the awkward ones
• Tell citizens when they are dealing with an AI tool, and how to reach a person
• Set up a simple way to report errors and complaints, and act on what comes in
• Review how the tool is being used at set intervals, and switch it off if it causes harm

Start in the back office

For most institutions the safest first step is internal: helping staff draft, summarise and search, with a trained person checking every output before it goes anywhere near the public. That builds skills and confidence, and shows where AI genuinely saves time, before any tool faces a citizen directly.

Maangarad offers AI training, policy support and carefully scoped pilots for public institutions, always with people in charge.