Artificial intelligence is currently used in in the public sector in the European Union. In particular, it is employed by public authorities in their enforcement activities. Such expression must be interpreted in the broad sense employed by the OECD, that defined enforcement as: «all activities of state structures (or structures delegated by the state) aimed at promoting compliance and reaching regulations’ outcomes [including] information, guidance and prevention; data collection and analysis; inspections; enforcement actions in the narrower sense, i.e. warnings, improvement notices, fines, prosecutions etc» (OECD, 2014).
During the lecture, four representative cases of public authorities currently using AI in their enforcement activities will be examined: police-related activities (predictive policing and facial recognition), revenue and customs agencies control, food security inspections, and supervisory activities in financial markets (SupTech). Such activities are all characterized by deep intrusiveness and by a huge degree of non-compliance by regulated subjects. Therefore, the use of new technologies might contribute to make such activities more effective.
However, this comes at a price. The literature on AI used in administrative procedure has shown how algorithms are often biased, discriminatory, opaque, and may have a negative impact on citizens and enterprises. Moreover, due to the black box nature of such systems, the right to participate in the decision-making process of the citizens might be at stake, not to mention the right to privacy and data protection.
When it comes to AI-driven enforcement activities, things get a bit more complicated. In addition to the aforementioned criticalities, there might a dilemma between the effectiveness of such systems and the protection of fundamental rights but also an impairment to the relationship between citizens and public authorities, exacerbating non-compliance and impairing the right to a good administration. Algorithmic darkness seems to be especially hard to enlighten, since enforcement activities often need to be carried out in secret in order to be truly effective: national legislatures often preclude access and transparency when it comes to such activities. In fact, the act through which the AI-driven administration chooses who and when to control might not be challenged in Courts.
In the light of that, this lecture is aimed at discussing possible remedies to these technology-related externalities, starting from the EU Proposal Artificial Intelligence Act . In particular, it will be argued that some of the measures (such as risk management systems, record keeping, human oversight, etc.) provided for the so-called high risk systems (including e.g. some forms of predictive policing or facial recognition, but not the use of AI systems by revenue agencies) could be also extended and adapted to enforcement activities. Finally, it will be discussed how similar measures might become new remedies in the administrative system.
Maria Bianca Armiento is a visiting fellow at the Law Department of the European University Institute. She holds a degree in Law from Lumsa University (Rome), where she is a post-doctoral fellow in Administrative Law and adjunct lecturer in Public Economic Law. She obtained her PhD in Public and Economic Law at the University of Pisa, defending a dissertation on the use of artificial intelligence in the administrative decision-making process. Her research interests include the relationship between administrative law and technology, better regulation, public utilities, and antitrust law.
Please register at the link below and specify whether you will participate in-person or online. Please note that you must show a valid green pass to access the EUI campus.