Reflections on the KVL Tutoring Framework: Past, Present, and Future

One of the great pleasures of working in a rapidly changing field is watching it grow and mature. With an auspicious beginning in the late 1960s, the field of AI in Education has evolved from a gleam in its founders' eyes to a well established discipline with an active scientific community that has produced an imposing literature. Now with more than three decades of experience under its belt, it is a good time for us to pause, take stock, and reflect on what works and what does not. We are most fortunate that Kurt VanLehn has done precisely this. He is particularly well suited to the task. Having led many of the key AI in Education projects spanning much of the history of the field, he has an excellent vantage point from which to observe the evolution of intelligent tutoring systems (ITSs) first-hand. In "The Behavior of Tutoring Systems," VanLehn has moved beyond literature reviews and architectural descriptions to present a unified procedural view of ITSs. Generalizing over the myriad implementation details of the most successful projects, he has distilled more than three decades of research into a generic tutoring procedure defined by two nested loops that iterate over tasks and steps. The two loops also serve as an organizing device for systematically exploring the core design decisions faced by tutoring system developers as they devise task selection, feedback, hinting, assessment, and review functionalities. To ground the discussion, examples are drawn from six landmark tutoring systems: Steve, the Algebra Cognitive Tutor, Andes, Sherlock, AutoTutor, and SQL-Tutor. The KVL Framework concisely articulates the collective wisdom of the AI in Education literature as it now stands. Considering the breadth of the research it summarizes, its appearance invites speculation about how it would have been perceived when the field was founded, how it can serve as a guide for scientists and practitioners in the present, and how it will be seen by future researchers. For purposes of discussion, we shall take the year of the framework's creation (2005) as the "bisecting year" between the founding of the field 35 years ago (the year in which Scholar was introduced) and the year 2040, 35 years from now. We first take an historical perspective and consider how the founders of the field might have reacted to this synthesis. We then take a contemporary view and argue that the framework should be employed as a blueprint for today's researchers and developers. We conclude with an exploration of how it may be seen by future scientists with another 35 years' experience designing, building, and deploying tutoring systems.