Why the public sector needs better rules, not more rules, for AI

This article is written by Gianluca Sgueo, École d’Affaires Publiques, Sciences Po

It has been commissioned and published as part of the Government AI Campus — an initiative by Apolitical to prepare public servants and policymakers to lead in the age of AI.

Many were shocked to read it: a study of 14,000 workers in 14 countries published in November 2023 revealed that over half of those workers use generative AI at work without official permission, and 40% are reckless enough to use banned AI tools. It’s as if the study said: stopping the advancement of automation is pointless because the first to want it are ordinary workers, not companies.

Does the same apply in the public sector?

Only partially, and probably not for the reasons many would think.

On the one hand, automation, powered by Artificial Intelligence (AI) and machine learning, is increasingly adopted in various sectors of public administration, from healthcare and transportation to urban planning and environmental management. Since 2022, Singapore has employed AI-driven virtual assistants to streamline its public services. These digital aides not only guide citizens through governmental processes but also help in making internal administrative decisions more efficient. Another example: over the last decade, Estonia has automated a significant portion of its public services, including e-Residency, e-Tax and digital voting, making public administration highly efficient and transparent.

"The underlying message is clear: even a highly automated public administration relies on significant human capital."

The truth is, however, that the process of automation in public services has been much slower, fragmented and sporadic compared to other areas of the economy. There are two important reasons for this:

First and foremost, the deployment of frontier technologies in public administrations is usually slower than in the market due to resistance to innovation, lack of funding, budget cuts and skills mismatch. In 2019, the OECD Observatory of Public Sector Innovation (OPSI) published a paper welcoming the use of Artificial Intelligence in the public sector, but it only described 7 case studies (most of which were just guidelines or directives). Objectively, the situation does not seem to have changed radically since then.

Second, public administrations are less incentivised to support automation. Despite being a major purchaser of goods, services and labour, the average public sector organisation does not react to changes in the environment by radically changing its approach and structure. Public sector organisations have usually responded to recurring demands and public pressure for further efficiency and cost-reduction by adopting more traditional approaches like staff restructuring, setting up new agencies or high-skill recruitment policies, but only for limited periods of time. Take the example of the present European Commission. Given its ambition to define a mature and flexible framework of rules as regards technologies (data, digital markets and services, and, of course, AI), since 2022, the Commission has embarked on a recruitment campaign in search of expertise essential for managing highly technical dossiers – skills that it was largely lacking, as these were traditionally unattractive to the public sector. These include, for example, data science experts, digital competition lawyers and algorithmic auditors. The Commission has primarily resorted to fixed-term recruitment of contract or temporary agents. This highly specialised workforce employed by the Commission will, at the end of their six-year (generally non-renewable) contracts, move on to new jobs. We can expect a problem of continuity of administrative action to arise soon.

In addition, automating public services creates new challenges to manage. Who is liable in the event of an algorithm error? There is a precedent for this: Mark Rutte’s government fell in 2023 following a scandal created by an algorithmic bias. But the same may not necessarily happen in other situations. A further problem: How does an automated public decision guarantee inclusiveness and transparency? Most of these questions remain unanswered or have been answered partially and inadequately.

So, should we be worried or not about public-service automation, and what should we expect from automating public services?

Many believe that all problematic aspects related to AI and automation (including the automation of public services) can be solved through regulation: “new rules”, where they do not exist, or “better rules”, where they do. According to the Global AI Legislation Tracker, legislative efforts are being made worldwide, including attempts to create comprehensive legislation for specific use cases or voluntary guidelines and standards. Stanford University’s 2023 AI Index shows that, globally, 37 AI-related bills were passed into law in 2022. But legislation in itself does not solve any of the issues that slow down automation in the public sector, least of all those making it unsafe from the point of view of rights.

First, the legal instrument is incompatible with technological systems that are evolving ever more rapidly, generating new risks associated with their use. The negotiation of the European AI Act, for example, stalled just when it seemed to be on track. The reason? National interests in emerging technology and generative AI escaped the rules up to that point.

Second, a fully automated public administration, despite its efficiency, would not be so great. Among the four scenarios produced by the simulation “The Future of Government”, run by the European Commission between 2018 and 2019, one imagined that AI-driven public services would be offered in an individualised and predictive way to citizens, but at the expense of their freedoms.

Not rules then, but better rules. That can mean many things of course. Here are two really essential ones: One is rethinking the design of digital interactions, abandoning forever the paradigm of simplicity that technology has accustomed us to. Another is constantly training public workers, while recruiting the most specialised skills on a temporary basis. The underlying message is clear: even a highly automated public administration relies on significant human capital.

The state, after the pandemic, is powerfully back in the market. Automation turns the scenario upside down: how much market do we want within the state?

👋 You can write an article like this one! Share your thoughts with a community of public servants. Learn more

(Image Credit: Unsplash)