In robotics activities, students are theorized to benefit from "learning-by-doing" activities where they set their own goals, but in practice, these activities have failed to produce the expected effects on STEM outcomes. To improve learning from robotics, this project will leverage teachable agent technologies, where students learn about a domain by teaching a computer agent. The agent interacts with students to expose misconceptions and encourage them to persist in the face of failure. By integrating the structure of teachable agents with the exploratory and engaging features of learning from robotics, the project will enhance the benefits of both approaches. The investigators will implement a robot that students can teach about the concepts they are learning in their middle school mathematics class. They will engage in design exercises with teachers and students to identify features of the robot that might be particularly important for improving student motivation and learning. Finally, they will conduct two studies, one in the laboratory and one in the classroom, to explore how students react to the robot in a realistic setting.
Intellectual merit: The project will improve understanding of what features students respond to in a teachable robot for mathematics, and of the potential educational benefits such a robot might have. The project will make a technological contribution by inventing methods for integrating teachable agents and robotic learning environments.
Broader impacts: The project will impact middle school students from underrepresented groups who will help design and test the teachable robot, exposing them to new technologies and engaging aspects of STEM careers. The project will also lead to a better understanding of the impact of teachable robots in the classroom and pave the way for these technologies to be adopted more widely in education.
Teachable agents foster student learning by employing the "learning by teaching" paradigm. Since social factors influence learning in this paradigm, understanding social behaviors a teachable agent should embody is an important first step for designing such an agent. This project studied the impact of causal attributions made by a teachable agent. Students were recruited to engage with a teachable social robot prototype. Subsequent interviews and discussions showed that with responses to problems solved/answered incorrectly : (1) students prefer we attributions that share the blame between them and the robotic agent, (2) you attributions might be motivating for students who see them as a playful challenge, and (3) I attributions could motivate students to engage further in teaching the robotic agent. With problems solved/answered correctly: (1) students perceive all of the robotic agent’s attributions positively, (2) you attributions might be particularly effective for students who need to boost confidence in their teaching abilities, and (3) I attributions in response to student effort might encourage students to take on effort attributions for their own problem-solving. In carrying out the above studies, we were also able to study the design and deployment of non-WIMP systems in a school setting, as the teachable robot agent and its environment, when regarded as an interface for engaging with an educational program, are a non-WIMP system. We created a list of five design recommendations based on our observations: 1) target multiple learning objectives, 2) emphasize the collaborative affordances, 3) optimize for training, 4) innovate the use of known system components, and 5) design for the adaptation and rotation of students. In the process of analyzing our data, we also developed a tool for integrating log and video data for exploratory analysis and model generation that 1) facilitates the annotation of log data with information from video data, and 2) automatically generates models of student problem-solving processes that include both video and log data. We demonstrated the tool with analysis of the teachable robot system for geometry and see this type of analysis as a starting point for quantitative analysis of the relationship between embodiment and problem-solving success in our system.
