The Teacher's Knapsack: Generative AI and Teacher Productivity

Abstract

Problem definition: Generative AI are multipurpose tools, meaning they can be used for a variety of different types of tasks. A primary reason generative AI may have a particularly large impact on teacher productivity is because teacher work involves a large variety of tasks (i.e., activities, quizzes, projects, presentations). Yet, teacher work is also characterized by high amount of  task discretion. Teachers often choose which task they do, such as between creating a worksheet or doing a hands on activity to teach a concept. How does generative AI impact the productivity of teachers who both plan and do their work? Methodology/results: We conduct a study that integrates a theoretical model of teacher task selection---a time-minimizing knapsack---with qualitative evidence about teacher use of generative AI. Specifically, we conducted in-depth interviews (29), observations (360 minutes of recordings of teachers using ChatGPT for their work), and surveys (34) at multiple points in time from 24 US K12 teachers varying in grade level and subject taught. We find that teachers use generative AI at the task level, as support for completing their to-do list, but also in some cases at the planning level, as support for coming up with their to-do list. Within a knapsack framework, planning-level generative AI use helps teachers ``re-optimize" their set of selected tasks given the impact of generative AI on task speed and quality. However at the task level, teachers simply use generative AI for their original set of tasks, leaving potential productivity gains unrealized. The productivity gap between planning-level and task-level generative AI use could be quite large, particularly when AI makes some tasks much faster or much higher quality---or much slower or worse--- than they were before generative AI. Managerial implications: Our study resolves an ongoing debate as to whether generative AI is helpful for teacher ``lesson planning." Productivity gains are not fully realized when teachers simply use generative AI on tasks they have done in the past. For teacher work, productivity gains from generative AI will be driven more by the substitution of tasks rather than the acceleration of existing ones.

Keywords: generative AI, productivity, education operations