Interpretable Neural System Dynamics: Combining Deep Learning with System Dynamics Modeling to Support Critical Applications

Deep learning models make accurate predictions but often function as "black boxes": The paper identifies a critical gap in AI development—while deep learning excels at complex pattern recognition and forecasting, it typically cannot explain why it makes specific decisions, which is problematic for high-stakes applications like autonomous transportation systems.

Combining AI with traditional system dynamics modeling improves both interpretability and reliability: By integrating deep learning with system dynamics approaches, the framework aims to preserve the transparency and causal reasoning of traditional models while gaining the scalability and predictive accuracy of modern AI, addressing a key tension in responsible AI development.

The framework targets real-world autonomous systems where explainability is essential: The research validates its approach through the EU-funded AutoMoTIF project on autonomous multimodal transportation, demonstrating that interpretable AI isn't just theoretical—it's being tested in safety-critical applications where understanding system behavior is crucial for public trust and regulatory compliance.

Three complementary interpretability techniques work together: The pipeline combines concept-based interpretability (understanding what features the model learns), mechanistic interpretability (understanding how the model works), and causal machine learning (understanding cause-and-effect relationships), suggesting that responsible AI may require multiple explanation approaches rather than a single solution.