Keeping judges in the loop: a human–machine collaboration strategy against the blind spots of AI in criminal justice

While seeping into every aspect of our lives, intelligent systems carry serious concerns, many of which stem from the need to secure control over machines ensuring they turn into an enhancement and not into a threat to human societies. Issues are not lacking. Whether one considers recruiting platforms perpetuating historical patterns of discrimination or AI-driven systems unfairly evaluating the creditworthiness of natural persons, it clearly appears vital to firmly place human beings with their rights and needs at the center of intelligent technologies. The paper tackles this kind of issue by focusing on the use of artificial intelligence in criminal justice, where predictive analytics and AI-driven decision systems have proven capable not only of enhancing the fight against crime but also of bringing the risk of opacity, aberrations, and injustices. We present a human–machine collaboration strategy providing judges with the advantages of AI and computational heuristics while offering control and understanding of the role played by machines. Drawing from a research that led to the development of an experimental platform supporting judges and public prosecutors dealing with organized crime, the strategy revolves around two components: (i) an online learning model designed to support judges in the evaluation of criminal dangerousness of individuals and groups capable of learning from users’ feedback; (ii) a human–computer interaction component exploiting visual metaphors to ease judges’ interaction with data, AI and other heuristics. The main contribution of the work is a novel and viable declension of the Human-centered AI paradigm in justice administration.