Profilazione e discriminazione. Dal GDPR alla proposta di Regolamento sull'Intelligenza Artificiale

The thread linking the practice of algorithmic profiling to the concept of discrimination is twofold. On the one hand, in fact, profiling itself is discriminatory in a “broad” or “literal-ethimological” meaning, insofar as it intrinsically consists in the creation of categories and therefore is deliberately aimed at the identification of differentiations between individuals; on the other hand, profiling can be discriminatory in a “strict” or “juridical” meaning, insofar as that identification of differences finds its reason in the recurrence of legally protected features. Having noted that, while in the latter case the problem originates in the discriminatory composition of the dataset, in the former one the problematic issue lies in the (un)knowability of the algorithmic processes that lead to the construction of the profile, the limits of the GDPR discipline will be highlighted pointing out its inability to offer satisfactory solutions. While not exempting itself from some critical points that will be raised in the conclusions, it is instead in the most recent proposal for an AI Regulation that useful tools can be found to address the identified problematic issues.