Human behavior modeling within an integrative framework

There exists a diverse variety of human behavior models, spanning multiple disciplines, each addressing a particular aspect of the whole problem of creating agents that behave realistically. This research focuses on the benefits and challenges of attempting to assemble such models into a unified system. It will be shown that, despite the inevitable complexity involved, the integration of human behavior models into a common framework can provide opportunities to leverage the work and assumptions of existing models to simplify the introduction of new ones. This is exemplified in the development of a novel, non-statistical solution to the problem of predicting future agent behavior from past observations that is particularly well suited to environments with large state spaces. By modeling the behavior of other agents in terms of the motivations underlying their actions, this method allows for the rapid inference of models that robustly predict behavior across novel states of the environment. This method adjusts quickly to new information, and is sensitive to the difference between observations that invalidate its current estimates and those that directly contradict hard constraints implied by past observations.