Data-driven police work
The lab studies data-driven skills for investigation and intelligence and techniques to improve the efficiency of the learning process.
Many important police facts are stored as unstructured free text. This data is difficult for a computer to analyze. Machine Learning models can help to provide insights into the data, for example by categorizing the data or extracting information. Many of these models often require labeled training examples, i.e. of the texts and the desired outcome. Labeling is an expensive process if done manually. In my research project I investigate how techniques such as Active Learning can be used to efficiently create models that work on (police) texts and to quickly find all relevant information for a search query.
To what extent can the knowledge in a machine learning model be formalized into a structured knowledge representation for the purpose of transparent and explainable decision-making? By combining machine learning with a structured knowledge representation, the respective advantages of both methods can be utilized. With machine learning models, knowledge can be captured from data, and by formalizing this knowledge into a structured knowledge representation, it can be presented in an explainable and robust way. This research will focus on formalizing knowledge in the form of Computational Argumentation using Graph Neural Networks and Large Language Models.
A dialogue system, or conversational agent, is an agent that communicates with another agent. Dialogue has been studied in the domains of human-computer interaction and formal argumentation. In my dissertation research, I explore the possibilities of combining these two approaches into hybrid argumentative-conversational agents in the legal domain. By combining the strengths of machine learning based conversational agents and formal argumentation, I aim to develop agents that can accurately and efficiently respond to natural language input by asking relevant questions and drawing explainable conclusions. In my function as AI Scientist at the Dutch National Police, I am currently implementing a dialogue system for the automatic intake of complaints of online trade fraud.
Publications:
Odekerken, D., Lehtonen, T., Borg, A., Wallner, J.P. & Järvisalo, M. (2023). Argumentative Reasoning in ASPIC+ under Incomplete Information (pdf, 244kb). Proceedings of the 20th International Conference on Principles of Knowledge Representation and Reasoning (KR'23). Supplement.
Odekerken, D., Borg, A. & Berthold, M. (2023). Accessible Algorithms for Applied Argumentation (pdf, 5.1mb). First International Workshop on Argumentation and Applications (App&Arg'23). Source code Visual Interface Documentation
Odekerken, D., Bex, F., & Prakken, H. (2023). Justification, Stability and Relevance for Case-based Reasoning with Incomplete Focus Cases (pdf, 632kb). Proceedings of the 19th International Conference on Artificial Intelligence and Law (ICAIL'23). Implementation and Demo Extended version with full proofs
Odekerken, D. (2022). Justification, Stability and Relevance for Transparent and Efficient Human-in-the-Loop Decision Support (pdf, 379kb). In Online Handbook of Argumentation for AI, Vol. 3.
D. Odekerken, F. Bex, A. Borg & B. Testerink (2022) Approximating Stability for Applied Argument-based Inquiry (pdf, 1.22 mb). Intelligent Systems with Applications.
D. Craandijk & F. Bex (2022) EGNN: A Deep Reinforcement Learning Architecture for Enforcement Heuristics (pdf, 107 kb). Proceedings of the ninth International Conference on Computational Models of Argument (COMMA’22).
A. Borg & D. Odekerken (2022) PyArg for Solving and Explaining Argumentation in Python: Demonstration (pdf, 240 kb). Proceedings of the ninth International Conference on Computational Models of Argument (COMMA’22). GitHub Visualization
I am interested in the intersection of AI and domain knowledge in practice at the police. How are decisions made at the police and how can AI (argumentation or ML) assist in this? To bring those two worlds together, both must become more transparent to one another. That means formal descriptions of police practice and XAI solutions for AI applications.
Publications:
J. Peters, F. Bex & H. Prakken (2023) Model- and data-agnostic justifications with A Fortiori Case-Based Argumentation (pdf, 839 kb). Proceedings of the 19th International Conference on Artificial Intelligence and Law (ICAIL'23).
J. Peters, F. Bex & H. Prakken (2022) Justifications Derived from Inconsistent Case Bases Using Authoritativeness. Proceedings of the 1st International Workshop on Argumentation for eXplainable AI (ArgXAI, co-located with COMMA ’22). PDF
J. Peters & F. Bex (2020) Towards a Story Scheme Ontology of Terrorist MOs. Proceedings of the 2020 IEEE International Conference on Intelligence and Security Informatics (ISI 2020). PDF
Among others, the National Police Lab AI produces new and innovative initiatives for the Dutch National Police. For my PhD I will conduct research to develop a method that can guide these new initiatives through further development phases, with the ultimate goal to deploy them as reliable and available applications that provide added value for the Police. The main focus in this research will most likely be an initiative that uses speech-to-text and other artificial intelligence techniques in order to reduce the administration burden on employees within the organization.