Intelligent Tutoring Systems (ITS) promise improved quality of education at reduced cost: they could provide the benefits of one-on-one tutoring without the personnel cost. However, this potential has not yet been realized. We propose a new paradigm for intelligent tutoring systems, that addresses a fundamental flaw in traditional approaches. The outcome of any teaching intervention is uncertain, and the state of the student's knowledge is uncertain. Our approach explicitly models this uncertainty, providing a more accurate model of the student and enabling the tutoring system to learn fran experience. We will apply this paradigm to medical statistics education. Medical practitioners need a good grasp of statistical concepts to read and critique medical literature, plan and carry out projects involving data collection, understand and practice scientific method in research, and make decisions in the face of uncertainty. However, there is widespread concern that statistics is not being taught effectively to health care professionals. Biostatistics is difficult to teach in a traditional classroom setting and is widely disliked. An ITS, providing one-on-one tutoring, could transform biostatistics education.
