AA.VV
What is the predictive accuracy and effectiveness of algorithms designed for crime prevention purposes? Are they sufficiently transparent? Are there any inherent biases? This report presents an analysis of three algorithmic tools used by the Spanish administration for predictive analytics in the police and prison sector (VioGén, RisCanvi and VeriPol, the latter of which employs artificial intelligence). The aim is to evaluate the functioning of these three algorithmic systems, particularly in terms of their adherence to fundamental rights. The study employs a mixed-methods approach, integrating quantitative and qualitative techniques. It has been carried out by professors at the Universitat de València and the Universitat Politècnica de València, experts in a range of legal fields (criminal, administrative and constitutional law), as well as in statistics and mathematics. The principal findings of the report are as follows: 1) It analyses the predictive accuracy of the three systems and their effectiveness in achieving their intended goal (crime reduction); 2) assesses their degree of transparency for the general public and for the people directly affected by their use; 3) identifies and discusses possible biases; 4) analyses whether they comply with the current European and Spanish legal framework, with particular attention to the protection of personal data and the regulations governing the use of automated decision-making systems. The study discusses whether the costs incurred in terms of fundamental rights exceed what is necessary and/or proportionate to the actual or likely benefits in terms of crime prevention. Furthermore, it underscores the importance of enhanced transparency to facilitate the necessary public debate on the instruments employed by public authorities, such as the three systems under analysis, which may result in limitations on fundamental rights.
