Foundation for effective machine learning in criminal justice: Filling the data gap

A recent doctoral thesis from the University of Sussex explores the challenges of using machine learning in criminal justice systems. The paper highlights critical issues such as overcrowding in jails, highly disparate outcomes, and strained resources. While algorithms have been identified as possible tools to tackle these problems, early studies reveal that they may also lead to further disparity. The data used in machine learning systems may be influenced by past discriminatory behavior and societal complexities, limiting the potential of these tools. The paper seeks to address these issues by linking multiple datasets to demonstrate racial disparity in US drug enforcement, creating a repository of criminal justice datasheets, and developing an interactive weak supervision framework to fill gaps in current datasets. These efforts can serve as a foundation for effective machine learning in the criminal justice system, as researchers evaluate the effectiveness of these tools and help to eliminate bias. By recognizing the limitations of current datasets and tools, fair algorithms can be refined and implemented, helping to create a more equitable system.