Episode 5: Criminal Justice Technology: The flaws, fallibility, and future of a quantitatively informed justice system

This episode of the University of Oxford’s Centre for Criminology’s podcast focuses on criminal justice technology: the flaws, fallibility, and future of a quantitatively informed justice system. MSc students AnneMichaela MacDonald Brodaric and Henry Rhyu speak to Savas Hadjipavlou who provides an industry perspective on themes such as accountability and bias in decision-making which are prominent in criminological discussions of legal technology. draws from his experience as the director of Justice Epistime, a data analytics company which within the criminal justice sphere. He also works with Cest Advisory, which notably provides ‘Poliscope’, a forecasting product that utilizes detailed police data to produce analytic and predictive results for police use. His background in in physics and mathematics led to him working on data analytics for social, economic and procedural issues in government and industry. Savas also became interested in criminal justice policy, and the how to engineer the system to better deliver services.Savas claimed “nobody has the perfect answer” but that with a proper use of quantitative techniques, applied in a measured way, strategy can shape and influence both the framing of law and the way in which services are delivered.

Discussion on how data presents one of the most pressing challenges faced by criminal justice services due to the immensity of the data available, and the fact that much of that data is inward looking or incoherent. Savas claims that people tend to seek comfort in working with an average, but that this number does not tell you very much. This is what Environs addresses. It is designed to bring together population data, ,such as gender, age, income data, economic data, school data and more ,integrate it, and put it on a map for visualization and exploratory purposes. A strong motivating factor for creating Environs was the recognition that criminal justice problems are multifactorial and multidisciplinary. The ambition of this project is to change, improve and understand population dynamics; a goal which has great utility at all levels of the public sector.

Savas also spoke about his work on modeling the impact of COVID-19 on the justice system. What stood out to Savas is that the change of behavior during the Covid pandemic meant that some crime types reduced, while other crime types increased, like drugs. Offenses that were not directly linked to interactions between people largely stayed the same. This opens up pathways for inquiries surrounding how much of this is durable change. There were also practical issues, such as hypothetically coping with half the police force falling ill, but in the end when the asymptomatic nature of the disease became clear, some of the scenarios that were foreseen did not materialize. The question of what it will mean for our future planning and resources is still as pertinent as ever.

The discussion concluded with a brief overview of machine learning, algorithmic bias, and risk assessments. Savas categorically stated that the risk assessment process is imperfect.

This is partly due to the fact that assessments generally consider a specific point in time within a specific context and therefore are not widely applicable or generalisable. If you couple this with questioning whether there is a quantum change, it lends itself to the conclusion - probably not. Savas distinguishes between the dangers of using prediction for large scale population trends and for predicting the behavior of a specific individual-nevertheless, both can benefit from informed human and machine insights.