Interfaces for Mixed-Initiative Planners
Position paper for the IUI Workshop '2000 Workshop on Using Plans in Intelligent User Interfaces
I have a computational background in machine learning, planning
and case-based reasoning, and I also have a strong background in cognitive
science and psychological
methodology. As further explained in my thesis
summary, my dissertation research investigated goal-directed learning
by creating computational models of introspective multistrategy learning.
I used these specifications to model the learning performed by humans during
complex reasoning tasks, and I applied the model to the task of selecting
a learning algorithm in machine learning contexts. The system I developed
is a hybrid model, that uses case-based methods to retrieve Meta-Explanation
Patterns (Meta-XPs) that support blame assignment after reasoning failure
and that assist in the generation of a set of learning goals. Given such
learning goals (representing desired changes in the systems background
knowledge or BK), the second part of the hybrid model uses a nonlinear
planner to construct a partially ordered sequence of calls to various learning
algorithms.
The foremost problem I am examining at this time is the role of the user in a mixed-initiative setting where human decisions are integrated with automated planning decisions. As further explained in my current research summary, I take three approaches toward this problem. First, I have developed an extensive graphical user-interface to the Prodigy planning and learning architecture that enables the human planner to examine the decision processes, to inspect the full data structures supporting planning decisions, and to visualize the plan as it unfolds. Secondly, I have relaxed the restrictiveness placed upon goal expressions by the classical AI planning paradigm. The user can express goals as actions as well as states, can represent goals along a continuum of specificity, and can interleave top-level goals and constraining subgoals in the input. Thirdly, I have developed a novel approach to replanning in the face of dynamically acquired information at planning time. Instead of replanning only by the adaptation of a current plan, I allow both the user and the planning system to shift planning objectives in a space of goal transformations. Taken together, this research provides novel means for automating the production of plans, but with human guidance.
As outlined in my plan for future research and Section 10.1 of my dissertation, I intend to use the concepts and technology developed in my thesis research to further formalize the task of learning-strategy selection. Furthermore, as a new research goal, I intend to study compare-and-contrast algorithms for category learning and theory revision. (See Section 10.2 - Learning Bias and Category Membership: An extension of my thesis for details) Given my experience, I have two additional research goals I wish to advance. From an intelligent systems approach, I am most interested in developing integrated multistrategy systems that combine planning algorithms with situation assessment algorithms to automate goal generation and to monitor plan execution. I have acquired experience in the former class of algorithms at CMU; whereas, I gained experience in the latter during my dissertation work. From an applied perspective, I am interested in continuing my examination of mixed-initiative systems. In particular, I plan to explore methods for understanding the explanation failures of students that learn diagnosis skills in intelligent tutoring environments. The results of such analyses would be to automatically assist students to learn from their mistakes. (See also Section 10.3 - Understanding student explanation failures: An application of my thesis)
In addition to research interests, I have equally strong interests in teaching both graduate and undergraduate computer science and cognitive science courses. My teaching assistantships, my contact with students as laboratory assistant, and my experience in creating and presenting tutorials for artificial-intelligence programming techniques have all been personally rewarding. I am committed to teaching and have found that it adds a richer dimension to my research. Not only does teaching encourage me to articulate my experiences and ideas for a more general audience, but because my work focuses on learning in both humans and machines, a very personal link exists between my teaching duties and research.
