Multidimensional Learner Assessment to Support Intelligent, Adaptive Digital Tutors

One-on-one tutoring has long been the gold-standard for individualized, adaptive instruction. Intelligent Tutoring Systems are Artificial Intelligence systems that have as their goal realizing the benefits of one-on-one teaching using technology. In order to deliver on this promise, these tutors will need a multidimensional assessment capability that is the mark of the good human tutor. Such a capability is however still under development. We propose to develop and implement a technique for building individualized representations of trainee performance that can be used for a multi-faceted assessment of current performance and to forecast future training needs. Drawing upon research in diverse fields, we will in Phase I both develop an initial model of performance assessment and design techniques for automating this analysis. The model will integrate physiological measurements, psychometric evaluations, and simulations to support assessment and training. When situated within a specific training context, the model will (1) operationalize system performance and (2) define normative behaviors within that context. The assessment system will use a variety of ontological representation and probabilistic reasoning techniques to analyze performance and sensor data into an actionable student model. Phase I will also result in a draft model, Phase II system design and a proof-of-concept prototype.