Building Educational Technology Quickly And Robustly With An Interactively Teachable AI

Intelligent tutoring systems (ITSs) aim to support student learning through comprehensive adaptive features, making for costly development times—about 200–300 hours of development time per hour of instruction. This proposal outlines plans to overcome several technical challenges toward building authoring tools whereby a non-programmer can build ITSs by interactively teaching simulated students. I propose both interaction design considerations and machine-learning innovations. These include a multi-modal natural language processing mechanism that mimics student learning from narrated tutorial instruction, and an active-learning mechanism that identifies training examples likely to eliminate inaccuracies in the simulated student’s induced production rules. I propose to evaluate these features over 3 user studies and evaluate the generality of this authoring method in a final open-ended authoring study. This work aims to democratize ITS authoring by opening new authoring opportunities to non-programmers by making authoring as time-efficient and natural as human-to-human tutoring. © 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.