How to prepare the Pa for the responsible use of artificial intelligence

These are useful skills, but they are insufficient if they remain detached from the day-to-day work of public authorities.

It is one thing to know how to use a generative AI assistant. It is quite another to decide whether a text produced by AI can be sent to a member of the public, whether an automatic summary of a file is reliable, whether a ranking generated by an algorithm is explainable, and whether a recommendation produced by a digital system meets criteria of fairness, transparency and proportionality.

The most important training questions then become very specific: at what stage of the administrative process does AI come into play? Who monitors the output? What criteria are used to assess an error? When is it necessary to pause and request a human review? What needs to be documented? Who is responsible for the final decision?

It is in this area that training must take a leap forward. It must help managers and civil servants to understand the organisational contexts in which AI is being introduced: processes, roles, decision-making criteria, responsibilities and coordination methods. Without this work, innovation risks remaining merely a technological layer built upon outdated practices.

In the press release, the National School of Public Administration is listed amongst the training bodies called upon to support this transition. This is an important point. The transformation will also involve programmes that enable people to tackle real-life scenarios: drafting a call for tenders, assessing an application, handling a complaint, organising a waiting list, making a service more accessible, or explaining a decision to citizens and stakeholders.

We need an institutional training ground: a place where we can practise asking the right questions, verifying the answers, recognising the limitations of the tools, and distinguishing between what can be automated and what requires human judgement.

The issue becomes even more sensitive with the agent-based AI.

Unlike more traditional systems, which respond to a request, agent-based AI systems can act more autonomously to achieve a goal. They can choose intermediate steps, initiate actions, coordinate information and propose operational decisions.

It is a powerful prospect, but one that calls for a clearer definition of responsibilities. The more autonomy technology gains in its operations, the more important it becomes to establish where human judgement comes into play, who is in control, who assesses the situation, and who makes the final decision.

Let’s consider the public health sector. AI systems can support the prioritisation of interventions, helping to manage waiting lists or assess the urgency of certain treatments. They can cross-reference clinical data, waiting times, the availability of facilities, the likelihood of a condition worsening, and prioritisation criteria.

This can help to make processes faster and more consistent. But when a decision affects people’s lives, the logic of optimisation is not enough.

A public sector manager must be able to understand the criteria on which the system bases its recommendations. They must be able to assess the effects of these recommendations, recognise any distortions, put the outcome into context and take responsibility for the final decision.

The same applies in many other areas: the allocation of grants, tax audits, the assessment of administrative applications, staff management, active labour market policies, and access to social services. Whenever AI helps to set priorities, classify cases or suggest decisions, there must be a human capacity for practical, documented and recognisable oversight.

Paradoxically, the faster the machine accelerates, the greater the value of the human ability to slow down when necessary: to check, interpret, ask questions and make decisions.

Collaboration between humans and machines therefore requires not only technical expertise, but also organisational, ethical and decision-making skills. The stakes are high, as this concerns services, rights and choices that have a direct impact on citizens’ lives.

There is also another aspect that we risk underestimating: trust.

Will citizens trust a public administration that also uses algorithmic systems to guide its decisions, services and priorities? Or will they see this transformation as a loss of control, transparency and accountability?

A great deal will depend on how the public administration explains the use of AI. It is not enough simply to state that the final decision remains with a human. It is necessary to demonstrate how this responsibility is exercised.

For example: a member of the public must be able to understand whether AI has been used to speed up a response, to analyse documents, to flag anomalies, to suggest a priority or to support an assessment. They must be able to know what checks are in place, what data has been taken into account, what scope for review exists, and who to contact in the event of an error.

Communication thus becomes an essential part of digital transformation. It serves to make more complex decisions easier to understand, to set out the safeguards put in place, and to prevent technology from being perceived as an opaque or uncontrollable force.

These days, everyone’s talking about soft skills. Far less often, however, is it made clear what they actually mean in practice.

In the public sector, these skills are reflected in very practical behaviours: knowing how to discuss a recommendation generated by a digital system; knowing how to request a review when something does not add up; knowing how to explain a decision to a member of the public; knowing how to coordinate different departments when AI changes a process; knowing how to recognise when efficiency comes into conflict with fairness, inclusion or the protection of rights.

They are not merely about the ability to collaborate, communicate or manage relationships effectively. They become crucial when they enable us to interpret ambiguous situations, negotiate meanings, take responsibility, build trust, and balance efficiency with rights, and innovation with legitimacy.

For this reason, within the public sector, they are regarded as genuine institutional hard skills.

The challenge in the coming years will be to translate these skills into training practices, assessment criteria and organisational behaviour. It will not be enough simply to say that critical thinking or sound judgement are needed. We will need to develop these skills through real-life scenarios, embed them in processes, and recognise them in the selection and development of public sector managers.

The future will belong to those administrations capable of adopting technology whilst exercising better judgement, responsibility and the ability to interpret the complexity that technology brings with it.

* Director of the Academy of Strategic Communication

** Associate Professor of Business Organisation at the University of Milan-Bicocca and Contact Person for the Human Resources Development and Innovation Division of the SNA