VIKING: Vertrauenswürdige Künstliche Intelligenz für polizeiliche Anwendungen
Motivation
In complex police investigation procedures, such as those involving organized crime or terrorism, large amounts of digital data regularly accumulate. Artificial Intelligence (AI) based solutions can support investigators in a timely evaluation of this data. However, police practice also shows that the use of AI comes with various risks for the success of the investigation. For example, unbalanced training datasets can lead to investigative errors, or cyberattacks on AI systems can mislead investigations in the wrong direction. Lastly, the traceability of investigation results is a crucial point for their admissibility in court.
Goals and Procedure
The goal of VIKING is to research and implement solutions for measuring and optimizing the accuracy, traceability, and robustness of trustworthy AI in police applications. VIKING explores an interdisciplinary approach that combines technical methods with legal-ethical requirements and contributions to the standardization and regulation of testing protocols. The solution to be developed will be exemplified by various AI systems in the field of image, video, text, and audio data analysis, implemented, and evaluated together with users. Subsequently, the possibilities and limits of transferring the technical approaches to other fields of application will be investigated.
Innovation and Perspective
VIKING provides for the first time a catalogue of requirements with concrete, operationalizable, ethical, and legal guidelines for the development and use of AI methods in police applications. Furthermore, the results from VIKING can promote the future use of AI methods by the police, thus contributing to the strengthening of law and security in Europe.
Selected Results
UKON contributes to VIKING as the technical and research coordinator, developing a holistic framework for trustworthy artificial intelligence for police applications and advising other work packages in using XAI techniques.
Fundings
Federal Ministry of Education and Research (BMBF, Germany) in the project VIKING (project number 13N16242).