Safer cities with AI and algorithms

Urban safety policy is a sensitive topic that often lends itself to political interpretation, attracts criticism, and is vulnerable to ideological debate. Yet designing effective strategies to address urban crime requires decisions grounded in evidence. The institutions responsible for public safety should be able to rely on robust data, but such information is not always available or sufficiently complete and up to date. Today, however, new technological tools—from satellite imagery to artificial intelligence algorithms—are providing powerful support for understanding increasingly complex urban dynamics and informing public safety policies. A study led by the University of Trento, with contributions from Fondazione Bruno Kessler (FBK), and published in Nature, examines how advances in big data, machine learning, and deep learning are transforming academic research on urban crime and reshaping the study of urban safety.

“Through our bibliometric analysis, we show that the use of computational methods and new forms of data collection to study crime in urban settings has grown substantially in recent years.
These methodological innovations, we believe, will contribute to the design of more targeted and equitable public safety policies,” explains Gian Maria Campedelli, criminologist, assistant professor in the Department of Sociology and Social Research at the University of Trento and researcher at Fondazione Bruno Kessler, as well as first author of the paper.

The study examines the evolution of urban criminology, which has traditionally relied on sociological theories, such as social disorganization, and environmental criminology, which focuses on the convergence in space of a motivated offender, a suitable target, and the absence of effective guardianship. Today, these theoretical approaches are increasingly integrated with computational methods that complement traditional data sources, such as crime statistics collected by law enforcement, and enable researchers to test these theories with greater precision. The new sources of information include social media posts, which help map mobile populations and real-time perceptions of safety; anonymized GPS data from smartphones and wearable devices, which reveal patterns of human mobility across urban environments; transportation data, including taxi flows and public transit networks, to identify factors associated with crime risk; and satellite imagery analyzed using deep learning algorithms to assess nighttime lighting, infrastructure development, and land use in relation to crime hotspots. 

These advances do not mean that algorithms can understand crime better than traditional investigative methods. Rather, they complement them. “We believe this evolution will significantly strengthen both research and the development of public policies that better respond to citizens’ needs. At the same time, it is essential to address the associated risks and establish appropriate governance. Algorithms can enrich traditional investigative approaches by improving our understanding of crime, but they cannot and should not replace them, ” says Campedelli.

As Campedelli explains, advanced computational methods extend well beyond artificial intelligence. They also include network science, which can analyze the relationships among neighborhoods, and agent-based modeling, which simulates criminal behavior under different scenarios.

These new digital data sources provide unprecedented spatial and temporal resolution—often down to specific hours and precise geographic coordinates—overcoming many of the limitations of traditional datasets, which are generally slower to collect, more expensive, and more highly aggregated.  This technological leap offers an unprecedented lens through which to understand and address complex social phenomena.

At the same time, these approaches operate within well-defined legal and ethical frameworks that establish safeguards for privacy and responsible data use. Addressing concerns about algorithmic bias, Campedelli notes:”In the past, predictive policing software was rightly criticized for algorithmic bias against minority groups and for the risk of reinforcing existing ethnic and social inequalities. Our findings highlight the importance of open science, not only to protect researchers but also to safeguard citizens. The data used today often contain inherent biases because they are typically collected for commercial rather than research purposes and therefore may not accurately represent the communities being studied.”

“Artificial Intelligence can be an important tool to help design policies based on data and evidence, but for this reason it must not limit itself to guessing where the next crime will take place, but must help understand the root causes of crime,” emphasizes Bruno Lepri, senior researcher at Fondazione Bruno Kessler, where he directs the Mobile and Social Computing (MobS) Lab research unit, among the authors of the study.

The authors therefore argue that AI should be used not merely as a “black box” for prediction, but as a scientific tool for validating causal relationships, advancing research, and strengthening the evidence base for public policy.

The study also identifies Italy as lagging behind other Western democracies in terms of data transparency. “Our analysis shows that Italy trails many other Western countries in the availability of both traditional and emerging data sources,  as well as in sustained collaboration between academia and public institutions. This significantly limits informed debate on urban safety, hinders the development of effective long-term policies, and makes it difficult to evaluate the impact of existing interventions,”Campedelli says.

The authors hope that greater transparency and stronger collaboration among universities, public institutions, and law enforcement agencies will emerge in the future,  together with platforms that foster multidisciplinary scientific cooperation.  Reflecting this approach, the research team includes Italian and U.S. scholars from institutions including Princeton University, the University of Pennsylvania, and Northeastern University, representing expertise in sociology, computer science, criminology, and economics.  Ariadna Albors Zumel, a PhD student in Sociology and Social Research at the University of Trento, also contributed to the study.

The article “Computational approaches and the future of urban crime research” is published in Nature and is available at this link: https://www.nature.com/articles/s41586-026-10622-4