Optimal Training System

Successful training in complex environments is normally accomplished through the interaction of a trainee and a skilled expert, but due resource constraints, experts' use in training can be problematic. Developing expert models for training is one solution, but constructing expert models is costly and time-consuming, and they tend to be difficult to validate, test, and debug. Alternatively, using an optimal model of task performance may be more efficient since optimal models are simpler to validate, test, and debug. Using a simulated task environment (STE) permits the necessary close model-trainee interaction. The simplifying assumptions of STEs often enable construction of optimal performance models, allowing them to perform the same task as the trainee using the same interface while closely observing and guiding trainee performance. We propose to develop the concept of using a normatively correct model of task performance as the core engine of an automated tutor with an initial application to a national missile defense (NMD) task STE. The NMD STE is a complex task requiring skilled operators to allocate assets under time constraints to minimize expected losses, yet is amenable to construction of an optimal model, and therefore has potential for exploring a normative modeling-based tutoring approach.