The goal of this project is to use the Disciple approach to build intelligent knowledge-based agents that expand the capabilities, generality and usefulness of educational software, such as the Multimedia and Thinking Skills (MMTS) shell. A Disciple-based intelligent agent has been integrated with this (non knowledge-based) educational software package, enhancing its functionality and usability by adding learner assessment and some intelligent tutoring capabilities.
The Disciple Toolkit has been used to build agents that are taught by domain experts on the MMTS project team. These learning agents are taught to perform actions on behalf of these domain experts. In the Disciple approach, an expert teaches an agent in much the same way that a human expert would teach a human apprentice. For instance, a historian, curriculum developer or teacher, has taught an Assessment Agent for MMTS to generate tests on higher-order thinking skills in much the same way a teacher would teach a student, by providing the agent with specific examples of test questions and solutions, explanations of these solutions, and supervising the agent as it generates and solves similar test questions. The outcome of this effort is a trained agent that is capable of generating a variety of questions that test the studentŐs ability to apply higher order thinking skills in the performance of MMTS-based writing assignments. For instance, to test a student's ability to judge the relevance of historical documents, the student could be shown a document and be asked if it is relevant to a certain task (see the figure for an example of such a test). Based on the answers to these tests, each learnerŐs proficiency in a particular higher order thinking skill can be assessed. The Assessment Agent assists the teacher to individually test each student, and it can directly assist each student to test themselves while learning.
The main goal of the Assessment Agent is to act as a natural, indirect communication channel between the teacher and the student. During interactions with the domain experts, the agent learns general rules and concepts for test generation and explanation, through apprenticeship and multistrategy learning, synergistically combining various learning strategies such as explanation-based learning, learning by analogy and experimentation, and empirical inductive learning from examples. In this manner, the agent has been taught to perform tasks within the problem domain of a novice journalist. Once sufficiently trained, the agent uses its rules base, its semantic network of concepts and features, and other task-related knowledge to "reason" about writing assignments and the relevance and credibility of potentially useful multimedia historical sources. As a result of this reasoning process, the agent can deduce the correct answers to test questions and provide intelligent feedback and hints to the learner. Thus, this agent assists in tutoring the student about how to apply the particular higher order thinking skill to the writing task.
Last updated on 07/13/01