Abstract for: Exploring the Role of Explainable AI Across the System Dynamics Modelling Process

SD is widely used to model systems, yet interpreting large volumes of simulation output becomes difficult as models grow in complexity, particularly for non-specialist stakeholders. This encouraged interest in AI tools for post-simulation analysis. However, many AI approaches lack transparency and conflict with SD’s emphasis on causal reasoning. This study examines the potential role of XAI across SD modelling stages and proposes a classification of its appropriate and inappropriate applications. This study adopts a stage-based conceptual comparative analysis informed by the literature. Given the limited work directly integrating XAI and SD, the analysis examines the two literatures in parallel: the SD literature is used to identify the main interpretive requirements across the modelling process, while the XAI literature is used to identify the explanatory functions and limitations of explainability-oriented techniques. We evaluate potential XAI roles across the five-stage SD modelling. In the early stages of problem articulation and dynamic hypothesis formulation, its role appears limited to supporting tasks such as highlighting candidate patterns for expert review, rather than shaping system boundaries. During model formulation, its use appears more defensible in calibration-related support than in structural validation. The most plausible role for XAI is at the testing stage, where it may assist with analysis of simulation outputs, scenario comparison, sensitivity analysis, and uncertainty exploration. The findings suggest that XAI should be viewed as a supplementary analytical aid within SD rather than a mechanism for generating or validating causal structure. Its value lies primarily in supporting the interpretation of model-generated outputs while preserving the central role of structural reasoning. By proposing a stage-based classification, the study provides a conceptual foundation for future empirical research examining how XAI may support SD practice in applied modelling contexts.