[Keynote 4] How Best to Harness AI’s Great Potential to Improve Education?

AI has tremendous potential to improve education. For example, a substantial amount of scientific evidence shows that AI-based tutoring systems can help students learn better than other forms of instruction. Also, some scientific evidence suggests that AI-based tutoring systems can help reduce inequalities in the educational system, even if not all available evidence points in this direction. Yet, in the US, standardized test scores are stagnant, including in K-12 mathematics learning, where AI-based tutoring systems are often used. Nor is there evidence that existing inequalities within the educational system are shrinking. These results are beginning to lead to calls to outlaw the use of computers in classrooms, in the US and elsewhere. How might we reconcile these seemingly opposing views from research and educational practice? More importantly, what might researchers do to help improve educational outcomes? We propose that it is important to focus on creating favorable circumstances for the use of AI-based tutoring systems. To this end, it is productive to view the smart classroom as a socio-technical ecosystem with many stakeholders: students, in the first place, and “facilitators” such as teachers, peers, human tutors, and parents/caregivers. By carefully designing human-AI interactions to support these stakeholders, we stand a good chance to harness AI’s great potential to improve education and reduce educational inequalities. We illustrate this vision with several example projects and promising empirical results from our lab. In these projects, students use AI-based tutoring software and facilitators are helped by a variety of novel AI-based tools, including as a mixed-reality analytics-based awareness tool for teachers, support for goal setting for students, and small-dosage AI-supported remote human tutoring.