Choice and Learning under Uncertainty: A Case Study in Baseball Batting

This paper describes the modeling of human performance in a real-world, embodied, stochastic task: baseball batting. Experimental results were gathered in a virtual reality setup and a Markov model of performance, especially errors, was developed. The focus of this paper is on a model of the task developed in the ACT-R cognitive architecture, most specifically of the critical subtask of generating an expectation for the next pitch. The model required no parameter tuning and provides an a priori account of the results based on the architectural constraints of declarative memory. The Markov and ACT-R models are briefly compared. The broader relevance of the task is discussed and possible applications are suggested.