Researchers at the University of Szeged (SZTE) are examining the potential use of artificial intelligence in criminal cases, with the aim of developing a decision-support system that could make judicial sentencing practices more transparent, consistent and fair, the university’s communications directorate announced.
According to the statement, the research seeks to create a model capable of predicting judicial decisions in specific types of criminal cases. While judicial prediction and litigation analytics are among the fastest-growing areas of the international legaltech sector, most existing solutions focus on civil or commercial law and lack dedicated criminal law modules.
The research project is led by Krisztina Karsai, head of the Institute of Criminal Sciences at the Faculty of Law and Political Sciences. As an initial step, the team is focusing on cases of human smuggling.
Karsai explained that this type of offence was chosen partly because relatively few factors need to be considered in order to reach a well-founded decision, and also because such cases tend to be highly similar and display recurring characteristics. The model is based on the analysis of 541 human smuggling cases heard by the Szeged District Court. Researchers are currently working on the theoretical framework of the algorithm, which is expected to be completed this year.
Once the algorithm is finalized, it will review the processed cases and propose a sentence for each one. These recommendations will then be compared with the penalties actually imposed by the courts. In a later phase, the team plans to expand the database by analyzing cases from additional district courts and further refining the model.
Based on the results, researchers will be able to identify general criteria to determine which types of criminal cases are suitable for algorithm-based judicial support. In addition to human smuggling, certain property crimes and some traffic-related offences may also fall into this category.
Krisztina Karsai stressed that the algorithm, which is expected to reach a level suitable for practical use by early 2027, is intended solely to support judicial decision-making. If approved, it could eventually be used in courts, but it will not replace judges.
She emphasized that there is no realistic prospect in the foreseeable future of algorithms deciding prison sentences on their own, noting that current legislation would not allow this in any case.
The system developed by the researchers, which has already received SZTE’s innovation award, is also planned to include a dynamic platform allowing members of the public to query aggregated sentencing data. This could help not only legal professionals but also non-experts better understand how the criminal justice system operates, strengthening transparency and public trust.
Related articles:
