Abstract for: Artificial Intelligence and Simulation

This presentation begins with a brief overview of Artificial Intelligence and how it intersects with simulation. From here, the presenters provide an overview of their work with meta models, in which a complex system dynamics or agent-based model is approximated with some form of neural network. The model is then run the model lots of times to establish the input/output relationships. This highly dynamic approach facilitates very fast policy analysis and calibration. This line of inquiry is demonstrated with the BodyLogical model. The talk closes with reflections on deep reinforcement learning and opportunities for System Dynamics practitioners to engage more fully in this burgeoning field.