Can we predict what someone will do next? This research presents a novel approach to modeling and predicting human behavior by representing it as a sequence of dynamic models, specifically Kalman filters, linked together by a Markov chain. This approach aims to capture the dynamic nature of human actions and provide accurate predictions over short time horizons. The core idea is that complex human behaviors can be broken down into a series of simpler, dynamic models that capture the underlying patterns. The Markov chain then governs the transitions between these models, allowing the system to adapt to changing circumstances. This framework is tested in an experiment involving automobile drivers, where the model achieves impressive accuracy in predicting subsequent actions based on initial preparatory movements. This modeling paradigm offers a powerful tool for understanding and anticipating human behavior in various contexts. The implications of this research extend to areas such as robotics, artificial intelligence, and human-computer interaction, where predicting human actions is crucial for seamless collaboration.
Published in Neural Computation, this article is suitable for the journal's audience due to its intersection of computational modeling and neuroscience. The use of dynamic models and Markov chains to represent human behavior aligns with the journal's focus on computational approaches to understanding brain function and behavior.