Understanding the Role of Metacognitive Control in AI-Augmented Medical Diagnostic Decision-Making

• Applicant: Professorin Dr. Ekaterina Jussupow
• Subject Area: Operations Management and Computer Science for Business Administration ; Data Management, Data-Intensive Systems, Computer Science Methods in Business Informatics
• Term: since 2025
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 568056994

As AI applications are introduced into various domains, radiological decision-making remains among the first work areas in which AI systems are introduced to support human decision-makers. In this context, it is necessary that humans are enabled to correctly assess the provided AI advice, avoiding fallacies of over- or underlying on the AI. This project aims to understand the role of metacognitive control strategies in fostering adaptive reliance on AI-based medical decision-support technologies. From a theoretical perspective, this project will investigate the role of different metacognitive control (i.e., decisions by individuals to influence their thinking process while monitoring their own performance) in AI-augmented medical decision-making, whether different strategies result in better performance and how metacognitive control strategies change over time and under different conditions of AI support. This work will significantly contribute to the understanding of adaptive reliance by taking a metacognitive control perspective, thereby addressing the current gap in our understanding of why individuals use AI and how they allocate cognitive effort in evaluating it as they have to interact over a longer period of time. Thus, we also contribute to the literature on meta-reasoning by understanding how metacognition and cognition interplay in complex decision-making processes. The resulting paper will be submitted to MIS Quarterly or Organizational Science (VHB A+). From a methodological perspective, this project will showcase how to holistically analyze verbalized decision-making protocols, by bridging qualitative inductive work, using LLMs as second coders, and using process mining methods to identify, visualize, and make sense of patterns in large think-aloud data sets. Thereby, it contributes to the emerging methodological perspective of computational intensive theory construction and quantitative grounded theory as well as to user behavior mining. The objective of this project would be the submission of a methodological paper in an A-outlet, such as the JAIS. From a practical perspective, a more in-depth understanding of verbalized thinking patterns allows future work to develop automatic training tools to help medical novices in their decision-making as it will enable comparing real-time thinking through verbalization with “effective” reasoning pathways monitoring which trajectories novices are currently at, potentially correcting them for. Furthermore, in-depth knowledge about the role of metacognitive control allows for the design of AI applications that foster this skill and support decision-making in high-stakes decisions.

• DFG Programme: Research Grants
• International Connection: Canada, United Kingdom
• Cooperation Partners: Professorin Susanne Gaube; Professor Vern Glaser