AI will change the task. Who keeps the authority?

On 1 September, I spoke to Nordic public-sector union representatives at Hotel KG in Klaksvík, at a conference hosted by Starvsfelagið. My subject was the Faroese national AI strategy: how a small society can use AI systematically, safely and to create value.

I began with a worker who still has responsibility, but is losing authority.

Imagine an AI system preparing a recommendation in a public case. An employee must approve it. Their name remains attached to the outcome. But they cannot see all the sources, correct the underlying information or change the quality criteria. Because the system is supposed to make work faster, they now have less time for each case.

The institution can say that a human remains involved. Yet the conditions for exercising professional judgment have deteriorated.

That possibility deserves a central place in the discussion about AI and work. Alongside questions about jobs and productivity, we need to ask who will have the power to shape the work that remains.

What a small society needs to keep

The Faroese AI strategy connects three ambitions: strengthening people, the Faroese language and the economy. Those ambitions depend on shared foundations such as governance, data, infrastructure and practical capability. The implementation arrangements are still being developed.

For me, a small society’s opportunity is to connect these pieces closely. We can start with a public need, test a limited service, involve the people who use and operate it, and retain what we learn. We can evaluate quality, access and workload alongside speed.

This requires clear decisions about authority. If an institution introduces AI into municipal casework, experienced caseworkers should help define the process and its exceptions. If a system brings information together for a customs officer, uncertainty must remain visible and the officer must be able to correct a mistaken identification. These are design questions to resolve before automation becomes routine.

Five responsibilities around the task

In the presentation, I used five responsibilities, adapted from Michael Houck’s five human roles in an AI organisation, to make this concrete for public services.

Maintain the human relationship. People need a practical route to a competent person when the standard process does not fit. Someone must be able to listen, explain and repair trust after an error. An automated response may help, but the institution remains responsible for the encounter.

Maintain the context. Rules, exceptions, professional terminology, previous decisions and corrections all need care. Someone must decide which sources are authoritative and notice when they become outdated. In a small-language community, this includes maintaining Faroese knowledge and language resources. The work needs time, recognition and an owner.

Set the quality bar. Qualified professionals should help define what counts as a good result. A measure of speed cannot capture whether advice is appropriate, an exception has been understood or a resident can act on the answer. Employees need influence over what is measured and how those measures affect the workflow.

Stop, challenge and correct. A reviewer needs the ability to reject a recommendation, pause an action, escalate a difficult case and record a correction. Those powers must be usable under normal working conditions. A button labelled “override” achieves little if exercising it is discouraged or there is no time to investigate.

Steward the public interest. Institutions must retain the capability to make choices about public money, citizen rights, data, suppliers and long-term dependence. That includes deciding when a system should change or stop, and preserving credible options for moving to another tool.

These responsibilities are choices about how we organise public work. They do not depend on a prediction that a particular task will remain beyond AI’s capabilities forever.

Participation before the decisions are fixed

My proposal to unions and public employers was to agree the conditions for responsible use before deployment.

Workers and their representatives need transparency about where AI is used, what it influences and who is accountable. They should participate while the workflow and procurement requirements can still change. Training needs protected time, and maintaining the knowledge behind the system must count as work.

The same applies to intervention. Staff need clear authority to pause, challenge and escalate, with support when they report recurring problems. Citizens need a route to a person when an automated interaction cannot resolve their situation.

An important question is where any saved time goes. Does it create room for complex cases, better explanations and professional development? Or does every gain become a higher throughput target, leaving less room for judgment? That choice will shape whether AI strengthens the service and the profession.

Institutions must keep their memory

This is also why organisational memory matters to my work with Usable.

Models and tools will change. Institutions need to retain their trusted sources, decisions, corrections and lessons across those changes. The next team should be able to understand why a workflow was designed as it was, what failed and which limits still apply.

Keeping that knowledge under institutional control can support continuity and learning. Software cannot settle who gets a voice in decisions, who has authority to intervene or what constitutes legitimate public service. Those responsibilities still need people and agreed working arrangements.

A practical question for the next AI project

Choose one proposed AI-supported workflow and identify who maintains its knowledge, defines acceptable quality, helps the person whose case does not fit and can stop the process when something goes wrong. Then check whether those people have the information, competence, time and authority to fulfil those responsibilities.

If they do not, the design is unfinished.

AI will change tasks. We should use it where it improves public service and removes frustrating work. We should also insist that the people who remain accountable can understand, challenge and change what happens.

Do not negotiate only over whether AI takes the task. Negotiate over who keeps the authority.

Based on my presentation, “A Small Society’s AI Strategy: Create Value, Keep the Authority,” to Starvsfelagið’s Nordic conference on 1 September 2026. For related background, see why the Faroese AI strategy needs a living memory.