This doctoral dissertation examines the policies and rationales behind the implementation of predictive policing from a multidisciplinary perspective, focusing on Italy as the primary case study. Predictive policing, the use of algorithmic predictions to forecast police targets such as future crime spots or potential perpetrators, has attracted growing scholarly and regulatory scrutiny in recent years. However, Italy remains an understudied country among those showing evidence of predictive policing use, and no attempts have previously been made to engage developers or public users in systematic studies aimed at addressing the limitations of available public information.

The dissertation comprises this synopsis and three substudies or research articles, which collectively cover the main thematic areas of the research. These areas include: the theoretical background and justifications of predictive policing, both as surface-level explanations and deeper criminological rationales; the model of governmentality emerging from predictive policing initiatives, including the altered interactions between private and public actors; and a field-based analysis aimed at reconstructing the characteristics of two proprietary algorithms used in Italy, leading to a case-based assessment of existing legal gaps and regulatory shortcomings.

One specific contribution to the field, particularly in the first article, is a rebuttal, on new theoretical grounds, of the traditional dichotomy between place-based and person-based approaches to crime prediction, challenging the idea of starkly higher invasiveness of the latter. In the second substudy, I examine shifts in the relationship between private and public powers, highlighting the risks to democratic accountability and outlining the concept of ‘soft privatisation’ for this specific segment of the law-enforcement chain. The third substudy analyses the responsiveness and transparency of public and private actors regarding the use and characteristics of two Italian predictive tools, advancing the field-based reconstruction of software workings, and identifying legal gaps in light of real-world implementations.

The aim of this synopsis is to extend beyond the individual articles by deepening theoretical and legal reflections outside the scope of the papers, while interweaving the different articles as interconnected pieces of a synergistic ensemble. Thus, the theory-based critique of the assumptions behind different techniques of crime prediction guides the presentation of the case study; likewise, by keeping in mind the same theoretical misconceptions – often combined with biomorphic conceptions of AI – the legal argument is built to highlight how regulatory countermeasures fall short of the goals they claim to pursue.