Agent-Based Modelling and Machine Learning
Kristina Bogner, Matthias Müller, Johannes Dahlke, Bernd Ebersberger, Thomas Berger · 2024
Complexity economics and transitions studies face the challenging task of accounting for complexity to explain the dynamics of socio-economic systems against the backdrop of grand societal challenges. A promising method to represent and better understand complex systems through simulation is the practice of agent-based modelling (ABM). However, researchers using ABM face the trade-off between capturing the complexity and the ease of designing and interpreting respective models. Beyond highlighting the relevance of using machine learning (ML) as an enabling mechanism for ABM to address this conflict, we provide concrete examples of how employing machine learning techniques with the objectives to increase the accuracy, understanding, and validity of models, can help to capture or process complexity. While we argue in favor of realizing the potentials of the green and digital twin transition in scientific practice, the resulting size and opaqueness of models require researchers to design, interpret, and communicate ML-driven ABM responsibly.