Abstract:
This dissertation presents an approach to the abstraction of the reasoning of a
knowledge-based agent that facilitates human-agent collaboration in complex problem
solving and decision-making and the development of systems for tutoring expert problem
solving to non-experts.
Effective human-agent collaboration requires an ability of the user to easily
understand the complex reasoning generated by the agent. The methods presented in this
dissertation allow the partition of a complex reasoning tree into meaningful and
manageable sub-trees, the abstraction of individual sub-trees, and the automatic
generation of an abstract tree that plays the role of a table of contents for the display,
understanding and navigation of the concrete tree.
Abstraction of reasoning is also very important for teaching complex problem-solving
to non-experts. This dissertation presents a set of integrated methods that allow the
abstraction of complex reasoning trees to define abstract problem solving strategies for
tutoring, the rapid development of lesson scripts for teaching these strategies to nonexperts,
and the automatic generation of domain-specific lessons. These methods are
augmented with ones for learning and context-sensitive generation of omission,
modification, and construction test questions, to assess a student’s problem solving
knowledge.
The developed methods have been implemented as an extension of the Disciple
learning agent shell and have led to the development of the concept of learning and
tutoring agent shell. This is a general tool for building a new type of intelligent assistants
that can learn complex problem solving expertise directly from human experts, support
human experts in problem solving and decision making, and teach their problem solving
expertise to non-experts. The developed learning and tutoring shell has been used to build
a prototype tutoring system in the intelligence analysis domain which has been used and
evaluated in courses at the US Army War College and George Mason University.
