The Digital Future of European Public Administration: Introduction to the Symposium on Safeguarding the Right to Good Administration in the Age of AI

Public sector decision-making increasingly relies on artificial intelligence (AI). Crime prevention and criminal justice, border management, tax administration, and the granting of social benefits are some of the areas where the use of AI shapes, sometimes transforms, administrative practices. The incorporation of new technology carries the promise of improving accuracy, efficiency, and potentially bias. At the same time, it also raises significant challenges for the control of public power. Algorithmic decision-making can weaken mechanisms for accountability, transparency, and fairness, perpetuate discrimination and biases, or undermine the humanity of public administration more generally.

The Symposium is dedicated to exploring these challenges from the perspective of the right to good administration, guaranteed under Article 41 of the EU Charter of Fundamental Rights (‘the Charter’, CFR). The provision codifies the general principle of EU law developed in the case law of the Court of Justice of the EU (CJEU). Essentially, good administration reflects the view that the EU is a community based on the rule of law, in which individuals are not subjected to arbitrary administrative decision-making. To benefit from good administration, individuals enjoy certain essential entitlements designed to ensure just and fair administrative procedures, including the right to be heard, the right of access to one’s file, and the right to a reasoned decision.

Initially, the rights now protected under the umbrella right to good administration emerged out of the canon of defence rights related to an effective remedy (now guaranteed under Article 47 CFR). Prior to be formulated as a fundamental right in Article 41 CFR, the Court of Justice would refer to national concepts of ‘good’, ‘sound’, or ‘proper’ administration, implying an ‘obligation’, ‘duty’, ‘rule’, ‘requirement’, ‘reason’, or a ‘principle’. In more recent case law, good administration has been unequivocally accepted as a general principle of EU law, meaning its requirements are binding on both the EU institutions, bodies and agencies, as well as on the EU Member State administrations whenever EU law applies. Non-compliance with the requirements of good administration might therefore lead to a finding of unlawful public conduct as well as to the invalidation of specific administrative measures.

The Court’s case law defining the contours of the principle of good administration provides essential guidance to tackle modern challenges posed by AI-innovation of the European public sector. Irrespective of the distinct circumstances surrounding a particular use of an AI system, the principle of good administration imposes binding legal obligations pertaining to the soundness and accountability of public conduct. In so doing, the principle enhances legal protections for individuals who become subjects of administrative decisions on the one hand, and strengthens a constitutional narrative based on the rule of law and democracy on the other hand.

Taking the EU right to good administration as a starting point, the symposium gathers reflections about the existing and future relevance of EU public law in regulating the use of AI by public administration. It opens with two sets of horizontal perspectives. The first of these delves into the limits of the rights and obligations guaranteed under the umbrella principle of good administration when faced with the use of AI by public authorities. The second set discusses the interplay between the principle of good administration and the rules of EU secondary law, in particular the EU GDPR and the upcoming AI Act. The symposium then turns to the specific challenges of AI-assisted decision-making across different fields, such as tax administration or border management, as well as across different jurisdictions.

The contributions in this Symposium converge on two main threats to the EU standards of good administration posed by algorithmic decision-making: bias and opacity. Bias is intrinsically incompatible with fairness and equality before the law, two values that underpin and guide the action of EU administrations in their relationship with individuals. Opacity, in turn, is contrary to transparency, another pillar of the principle of good administration and an essential requirement imposed on administrative action in order to make judicial scrutiny possible. From this perspective, good administration has an important role to play as a break to the development of algorithmic decision-making procedures in the public sector. The issues raised in this Symposium will be addressed in an expert roundtable held in the context of the DigiCon III conference that takes place in Florence, Italy on 19-20 October.

The Principle of Good Administration: Reasoning Obligations, Transparency, and Beyond

Opening with a daring view, Albert Sanchez Graells suggests extending and broadening the good administration guarantees in the context of AI-assisted decision-making. The author calls for extension of the requirements to the phases of decision-making that are not yet directly relevant to the individual. Furthermore, he proposes broadening the good administration guarantees to a collective dimension, in order to account for the new risks arising in the AI-driven administrative context. To that end, the contribution also looks at ways to achieve this, whether through an expansive interpretation of Article 41 of the Charter or through a European legislative reform.

Exploring ‘what can we (not) learn from current EU administrative law’ about compliance with the obligations of good administration, Filipe Brito Bastos proposes to foresee the types of problems raised by AI-assisted decision-making as not entirely new to European administrative law. Reflecting on the many similarities with other types of administrative decision-making, the author suggests looking into the CJEU’s case law on internal guidelines as a useful starting point for determining how the duty to give reasons must be complied with in AI-assisted decision-making.

Building on the many documented challenges accompanying the ADM, Marco Almada takes us into the specific issue of algorithmic uncertainty. By recalling its origins in techno-scientific uncertainty, which surrounds the inner workings of AI systems, the author explains how different types of algorithmic uncertainty affect the information that is eventually to be relied upon by the administration. The author then reflects on the possible ways (and their limits) to address the uncertainty in ADM.

Good Administration and Secondary Legislation: The AI Act and the GDPR

Mitisha Gaur zooms in onto the EU’s very-much awaited piece of legislation that will transcend all of the EU policy areas: the Artificial Intelligence Act. Exploring this emerging High-risk AI governance ecosystem – “Government HRAI” – the author shows how the emerging legal framework accounts for many of the horizontal challenges associated with the use of AI, inter alia, in the context of public administration.

Jacopo Dirutigliano provides a discussion on the relationship between the right to a reasoned decision and the right to explanation in the EU’s pending AI Act. The author offers two key perspectives in this respect: first, exploring the requirement to provide explanation under the duty to give reason in EU administrative law, and, contrastingly, exploring the requirement to provide explanation under the so-called “right to explanation” (RTE), introduced by the proposed AI Act, and its impact in both public and private law sectors.

Marco Fontana’s contribution delves into the nature of the information produced by automated decision-making (‘ADM’), reflecting on whether and how the secondary rules governing ADM, namely stemming from the EU’s General Data Protection Regulation (‘GDPR’) apply to administrative ADM. In this regard, the author reflects on the key principles of EU administrative law in light of the emerging rules under the EU’s Artificial Intelligence Act (‘AI Act’).

Good Administration Across Different Fields: Law Enforcement and Tax Administration

Zooming into the existing case-law of the CJEU concerning law enforcement use of AI for the purposes of ensuring border security, Alexandra Karaiskou explores how general principles of EU law pertaining to procedural fairness, namely the duty to give reasons and the principle of equality of arms during judicial review can be re-interpreted in the digital era. In this respect, the contribution revisits recent case law, including the Ligue des droits humains judgment to map the jurisprudential standards for transparent and accountable use of AI in individual public sector decision-making.

Arthur Bianco and Katerina Pantazatou delve into challenges posed to compliance with the right to good administration arising from the increasing reliance on AI in the context of cross-border taxpayers’ tax assessments in the EU. Arguing that in the absence of a harmonised legal framework in direct taxation, and in light of mounting practices ranging from tax planning, tax avoidance, and even tax evasion, lead European tax administrations to explore AI uses, which have the potential of directly conflicting with the guarantees of Article 41 of the Charter.

Good Administration Across Different Jurisdictions: Examples from France and the UK

The symposium closes with two examples of new developments in the use of AI in French and UK public administrations. Alexandre Stepanov takes us to France for exploration of how the French legislator’s conception of a dual nature of the obligation to give reasons for administrative decisions.

Alexandra J Sinclair offers a look into the UK’s Department for Work and Pensions use of different machine learning models to identify potential fraud by applicants for Universal Credit, the UK’s living support payment for those out of work or on low incomes. The author argues in this respect that a blanket refusal to disclose any of the risk model indicators or decisional criteria used by these systems does not adequately balance the risks of procedural unfairness for citizens against the risks of gaming and manipulation of these fraud detection systems.