The Teacher's KnapsackGenerative AI and Teacher Productivity

Much of the existing evidence on generative AI productivity studies settings in which tasks are assigned. We study K12 teachers, who choose which instructional tasks to prepare. We model teacher preparation as a time-minimizing knapsack in which the teacher chooses the least-time bundle of instructional tasks that meets a required learning value. We ground the model in a case study of 24 teachers over the 2023--2024 school year, including 360 minutes of observations, 29 interviews, and 34 surveys. Regarding the impact of generative AI, the evidence motivates a distinction between teachers' task selection and task execution. Teachers who use generative AI to execute familiar tasks in their legacy bundle of tasks can save time or increase the learning value of those tasks. Yet, moving away from legacy tasks can generate additional time savings when AI makes alternative tasks faster, more valuable for learning, or newly feasible. However, there is a cost to switching away from the legacy set of tasks, and for some teachers this may be too high, particularly when generative AI substantially improves the legacy tasks themselves. Environmental changes may also make selecting a new bundle of tasks necessary, even for teachers who do not personally use AI. Our case study evidence provides insight into these mechanisms: teachers commonly used AI for historically familiar tasks, whereas teachers reporting productivity gains increased their use of AI for jumpstarting and iterating on what tasks to do in the first place. The central implication is that AI productivity in discretionary work depends jointly on task acceleration, substitution, and the cost of evaluating alternatives.

From SSRN, via Crossref