Human-Centred Digital Governance Starts With the Right to Question a Decision

Imagine a resident trying to correct an error in an online public-service record. The portal rejects the request. Its chatbot repeats the same answer. There is no telephone number, no explanation and no person authorised to change the outcome. The service is digital, but the resident has lost a practical way to be heard.

This illustrative scenario captures the question at the centre of human-centred digital governance: can people understand, use and challenge the systems that affect them? Faster processing is valuable. It is insufficient if the cost of a mistake falls on someone who cannot obtain help.

Start before the software is bought

Human-centred governance covers the rules, institutions and decisions around digital services, data and AI. It begins with deciding whether a proposed system solves a real public problem. It continues through procurement, design, operation, complaints and retirement. Adding a friendly interface at the end cannot repair an unaccountable decision process.

UNESCO's AI ethics recommendation connects human dignity with participation, privacy, fairness and accountability. Its Caribbean AI Policy Roadmap offers a regional framework for policy development and collaboration. These are reference points for policy work; they should not be described as a single law automatically applying across Caribbean jurisdictions.

Make participation capable of changing the design

A consultation should happen while there are still meaningful choices to make. Invite the people who use the service, frontline staff, disability organisations and groups likely to be missed by an online survey. Explain the proposal in ordinary language, show concrete examples and publish what changed in response.

For a Caribbean service, questions about language, travel and connectivity belong in that discussion. Can a person complete the task without a smartphone? Can someone who uses a screen reader navigate the form? Does an alternative channel work when electricity or connectivity is interrupted? The answers will differ by territory and service; a regional template should leave room for those differences.

Participation also requires boundaries. People should not need to disclose private circumstances publicly to explain why a service excludes them. Offer confidential routes and collect only what the consultation needs.

Give a decision an accountable owner

For each consequential service, identify who owns the outcome and who can correct an error. Explain whether AI assists staff, recommends an action or determines a result. Publish a useful description of the criteria and limitations without exposing private records or creating security risks.

An appeal route should say how to request review, what evidence is needed, when to expect a response and who can reconsider the result. A meaningful review must allow correction. Sending a complaint back through the same automated process is not enough.

These are practical design recommendations. The precise legal duties on explanation, review, data protection and administrative decisions depend on the jurisdiction and service, and require appropriate local advice.

Govern the infrastructure beneath the interface

The UN Universal DPI Safeguards Framework, released in September 2024 under the stewardship of UNDP and the Secretary-General's Envoy on Technology, offers guidance for safer and more inclusive digital public infrastructure. Its scope extends beyond an individual AI application to the foundations through which digital services operate.

For a purchasing team, that means asking where information travels, who can access it, how long it is retained and how a person can correct it. Require a workable exit plan: exportable records, documented interfaces and clear arrangements when a supplier changes. Test these capabilities before dependence makes them expensive to recover.

Keep service information separate from optional marketing consent. Do not collect additional personal data merely because a platform makes it easy. If a proposed use requires sensitive information, establish a legitimate purpose and the applicable safeguards before collection begins.

Measure the experience of people who struggle

A dashboard that reports only average completion time can hide exclusion. Examine failed applications, abandoned journeys, repeat visits, correction delays and unresolved complaints. Where it is lawful and appropriate to investigate differences across groups, protect privacy and avoid conclusions that small samples cannot support.

Ask staff and service users to walk through difficult cases together. An organisation should be able to demonstrate that a wrong record can be corrected and that an interrupted application can be resumed. Publish understandable results and the improvements still needed.

What this means for CAIA

Human-centred digital governance belongs alongside technical capability in CAIA's public mission. The practical work is to convene people affected by technology, help institutions examine procurement and oversight, and develop accessible guidance grounded in Caribbean conditions. It does not give the Association regulatory powers or replace national institutions.

A useful next step is to choose one service and publish its accountability map: purpose, responsible office, data flows, non-digital access, review route and measures of success. Pair that with the technical testing described in our AI safety and alignment article. People should be able to benefit from a digital service and retain the power to question it.