A Robot Dynamically Asking Questions in University Classes

Abstract

We developed a robot that dynamically asks questions after listening to a lecture in university group classes to enhance students' learning. Based on interview results with ten professors, we designed our robot to let professors choose when a robot asks questions, ask various kinds of questions by letting professors choose question types, and ask questions dynamically without sharing them with professors in advance. The prepared question types include such as clarification questions, which are fact-based questions about a lecture, and thought-provoking questions, which activate students' critical thinking. We expanded them into eight sub-categories of questions that the most common types that are likely to emerge in classrooms. Leveraging large language models (LLMs), based on lecture transcripts, we developed a robot system that generates questions in the eight sub-categories and yields 0.93 classification accuracy toward human coding results. We also conducted a case study with our developed robot in four university lectures. In these lectures that involved our robot, the professors often asked reasoning questions and criticism questions, and the students showed engagement with the robot-professor dialogues. Interviews with 20 students revealed that the robot's questions contributed to the classes by helping the students reflect on the lectures and gain new perspectives. The professors also recognized some benefits of the robot, perceiving its presence as a question facilitator, a mood maker, a communication tool as well as a motivator to prepare more diligently for their own lectures.