Socratic interaction is a mainstay of instruction in professions such as law, business, and education, and to a lesser extent, in the military. Socratic instruction comprises a set of techniques that help students acquire and practice professional-level reasoning and decision-making skills. Our observations of exemplary tactical instructors demonstrate these techniques in use with military commanders. Unfortunately, this is a difficult and expensive form of instruction. Accordingly, we are extending the general drive towards development of Intelligent Tutoring Systems (ITSs) that automate military instructional expertise, seeking to understand and address the specific requirements of Socratic tutoring.

Socratic tutoring is characterized by (a) the setting of a thought-provoking problem, (b) a student's attempt(s) to provide solutions to the problem, (c) the instructor's repeated exploration and challenging of the student's solutions, which (d) elicits incremental justification, elaboration, refinement, and revision of both the student's understanding of the situation under discussion and their proposed solution. The prototypical structure of Socratic tutoring sessions involves a series of tutor-generated questions and student-generated answers. Major issues, then, are how should an automated Socratic tutor control its participation in such a dialog, and how can the behavior of such a tutor be specified cost-effectively?

This paper describes results from a U.S. Army Research Institute (ARI) sponsored project that developed a prototype Socratic tutor for battlefield command reasoning skills. In this application, the problem situations were tactical decision games. The tutor's behavior was modeled after that of expert tactical instructors. We present examples of the tutor's behavior, characterize its general capabilities, explore the discussion control mechanisms it uses to produce this behavior, show how scenarios and dialog moves are scripted, and analyze the costs and benefits of our approach, including its relation to prior work and likely future directions.