Automated Socratic Tutors for High-Level Command Skills

摘要

Under stress, human decision-makers revert to their best-practiced habits. This includes military commanders who may fail to act effectively under pressure for lack of sufficient practice. The US Army Research Institute (ARI) developed a training methodology emphasizing repeated exposure to small challenging vignettes enabling drill on command decision-making. Rather than role-play to a simulated conclusion, mentors focus on analyses and dialogues that explore reasoning and rationale. Adoption of this methodology in courses at Forts Leavenworth and Knox is helping to validate this approach to honing command skills. However, intense practice with human mentors is problematic as there are generally too few expert mentors available. This paper describes an ongoing Small Business Innovative Research (SBIR) project to develop automated intelligent tutors to fill the role of expert mentors. This project is developing a novel capability to understand, critique, and discuss proposed courses of action in a Socratic mode, guiding the student as an expert would. The approach emphasizes multi-modal interaction (e.g., language and graphics), models of expert human tutors, and development of authoring tools to reduce training system costs. The paper presents data from analyses of expert mentor and student dialogues during 'tactical decision games.' It then describes how this data is used to develop and assess the project's intelligent tutor. Additional preliminary data from early informal formative evaluation of the Phase I prototype system, and initial student feedback on some Phase II refinements is also reported.