In the optimized modeling framework (bottom), a scientist trains both an animal (‘system’, orange, right) and an artificial neural network model (ANN) on the same or an analogous task (yellow). The scientist then generates data from both the system and the optimized model (green, right and left, respectively). If the model data and the system data are quantitatively similar, then the scientist attempts to discover the mechanism (cyan) in the optimized model. That mechanism provides a hypothesis of how the optimized model works. Credit: David Sussillo