VIKING investigates how artificial intelligence methods can be used responsibly in police investigations. Complex investigations may involve large volumes of digital data. AI-based methods can support experts in analysing these data, but their results must remain reliable, understandable and legally reviewable.

The project does not aim to develop autonomous surveillance or decision-making systems. Instead, VIKING explores technical and methodological approaches for assessing and improving the accuracy, robustness, transparency and explainability of AI systems. The project also considers potential biases in training data, the fairness of the methods used and risks arising from targeted attacks on AI systems.

VIKING combines technical research with legal and ethical requirements. The developed approaches are examined through exemplary applications for the analysis of image, video, text and audio data and evaluated together with practitioners.

Research Questions

• How can AI-based analysis methods be designed so that their results remain transparent and understandable?
• How can the accuracy and robustness of AI systems be measured and improved systematically?
• How can biases in data and models be identified and reduced?
• Which legal and ethical requirements must be considered when developing and using AI methods for police applications?
• How can verifiable standards and testing procedures for trustworthy AI be developed?

Selected Results

The University of Konstanz contributed to VIKING as a technical and scientific coordinator. The research group focused on Visual Analytics and on methods for transparent and explainable artificial intelligence.

The results include:

• a comprehensive framework for trustworthy AI in police applications,
• methods for examining transparency, explainability and fairness,
• contributions to test catalogues and evaluation procedures,
• contributions to DIN SPEC 91517:2025-05 for trustworthy AI methods in police applications,
• scientific publications on Explainable AI, time-series analysis, Visual Analytics and the interactive analysis of complex models.