Behavior prediction using LLMs

Anthony M Harrison, Laura M. Hiatt, J. Gregory Trafton

Abstract: Both cognitive architectures and neural networks have been used to understand and predict behavior. Here we present work on the prediction of human performance on a computer task using a transformer-based neural network trained on data generated by a cognitive architecture. The resulting system is able to make fine-grained predictions of the human's next actions, given a window of prior actions.