Dynamic planning models to support curriculum planning and multiple tutoring protocols in intelligent tutoring systems

We are building an Intelligent Tutoring System called CIRCSIM-Tutor designed to help medical students learn to solve problems in cardiovascular physiology. This dissertation describes two planning models intended to improve the adaptive dynamic planning capabilities of CIRCSIM-Tutor. These models have been implemented using Reva Freedman's new Atlas Planning Engine.; The first new planning model is a Curriculum Planning Model. This work endows our system with the ability to build a different individualized curriculum for each student. To make curriculum planning more effective, we also developed a new student assessment method for the curriculum planning model. This assessment is used in determining which set of problems should be presented to the student at each point. The new curriculum planning rules have been shown to be complete and consistent.; The other model extends our Discourse Planning Model to provide capabilities for Multiple Tutoring Protocols. A tutoring protocol defines the overall communication between the tutor and the student. The implementation of multiple tutoring protocols allows us to compare the effect of different protocols in teaching causal reasoning. We have used machine learning methods to analyze a set of human tutoring transcripts to discover how and when human tutors switch protocols. Although they originally planned to use a protocol that collects a whole series of predictions from students before tutoring them, our human tutors switched protocols when students showed serious difficulties in solving a problem, in order to give the students more immediate feedback.; We are especially interested in this issue of immediate feedback, so we carried out an experiment to find out how students feel about immediate feedback, in which we analyzed the students' performance using CIRCSIM-Tutor and their responses to a questionnaire. Results showed that the weaker students are more likely to prefer immediate feedback.