Opacity and discriminatory potential of AI- and GenAI-powered public revenue collection systems: food for thought for the Italian case

The use of artificial and generative intelligence (AI and GenAI) is increasingly widespread and is rapidly revolutionizing a wide range of areas of social and economic life, from finance to healthcare, from logistics to security. The use of applications, pc programs, technologies and systems that use machine learning algorithms and neural networks to autonomously perform tasks that would traditionally require human intelligence promises to open a new and fascinating chapter in many sectors. More recently, for example, AI and GenAI systems have shown that they can also be used very profitably in the field of education, where they are used as a powerful teaching and pedagogical tool to personalize the educational experience and thus improve the effectiveness of teaching, and in the world of public administration, where they seem to be the most suitable response to the growing need to increase efficiency and effectiveness in the management and delivery of public services. In addition, in this area, we can note that the possible applications of AI and GenAI are innumerable – from the simplification of procedures to the optimization of available resources, through the implementation of predictive systems for urban planning and emergency management – and the horizon that opens up reveals a broad panorama of options and possibilities, which until now have remained largely unexplored. In fact, it is not only a matter of technological innovation, but of a change of epochal dimensions that requires a responsible approach to development and the formulation of ethical and legal principles, to govern a phenomenon that, in addition to countless opportunities, undeniably presents many risks. These latter, however, are often underestimated or even ignored, while they should be taken into consideration and studied more carefully, because around the aforementioned phenomenon there is also developing a rapid change in individual and social behavior, in the world of labour, in the economic system and, consequently, in the way of conceiving and guaranteeing people's social rights (welfare). This paper intends to examine the issue of risks related to the use of artificial intelligence by the Financial Administration in one of the public functions that will be most affected by technological evolution in the coming years, that is the function of tax revenue collection. This latter is crucial because the financing of goods and services of general utility (such as healthcare, education, the construction of infrastructures, etc.) and, therefore, the provision of services useful for satisfying the aforementioned social rights depends on it. The effectiveness and efficiency of the tax collection system are also factors of essential or, better, strategic relevance, because they allow the financing of public investments, which in turn are fundamental for economic growth and social progress. The enhancement of the collection system through AI systems is therefore undoubtedly desirable, because this innovation would bring a certain benefit to the community with the help, in some cases, of unsophisticated and easily controllable tools. It is good to consider, however, that AI, depending on the way it is developed and used, can potentially generate and/or aggravate distortions, often determined by the prejudices and cognitive biases of those who develop and train the algorithms, and thus produce different forms of automated discrimination. In this work, therefore, we will try to analyze the type of distortions that can affect AI- powered taxation systems and the consequences that can arise from them. In particular, in the first part of the work, we will analyze the critical issues that inevitably relate to the use of automated procedures in the tax field , as well as the regulatory apparatus that has been prepared at EU level to protect the fundamental rights of individuals and to guarantee the fairness and impartiality of administrative action. In the second part, instead, we will try to verify what has been done by the Italian legislator to reduce the risk that the opacity and discriminatory potential of algorithms produces an increase in economic inequality among taxpayers. In the third and final part of the work, we will finally identify some principles that should and/or could be introduced into the Italian tax system to obtain the right balance between the public interest in collecting an amount of resources that satisfies the general interests of the community, while simultaneously reducing collection and accomplishment costs to a minimum, and the rights of taxpayers not to be discriminated against on the basis of personal characteristics (such as race, sex, sexual orientation, religion or disability), consumption habits and/or their lifestyle.