Article published in journal

When analyzing criminal cases, large amounts of data need to be processed, making manual analysis impractical. Artificial intelligence (AI)-driven information extraction systems can assist investigators in processing this data, leading to significant improvements in effectiveness and efficiency. However, the use of AI in criminal investigations also poses significant risks to individuals, so contestability must be integrated into systems and processes. To meet this challenge, contestability requirements need to be tailored to specific contexts.

In the article “Contestable AI for criminal intelligence analysis: improving decision-making through semantic modeling and human oversight”, Falk Maoro and Michaela Geierhos have analyzed and adapted the existing requirements for the analysis of criminal cases, focusing on the retrospective analysis of police reports. To this end, they introduced a novel information extraction pipeline based on three language modeling tasks, which they refer to as semantic modeling. Building on this concept, the authors evaluated the requirements for challengeability and integrated them into their system. As a proof of concept, they have developed an AI-driven information extraction system that incorporates defeasibility features and provides multiple functions for data analysis. The results highlight three key perspectives that are essential for the contestability of AI-driven investigations: Information provision, interactive controls and quality assurance. This work contributes to the development of transparent, accountable and adaptable AI systems for law enforcement applications.

The article appeared in the July issue of the journal “Frontiers in Artificial Intelligence” in the section “Machine Learning and Artificial Intelligence” (https://doi.org/10.3389/frai.2025.1602998).

^{Source: Frontiers in Artificial Intelligence}

< Zur Newsübersicht