Pricing Intelligence: Task-Based Learning and Labor Displacement in the AI Economy

Abstract

What determines the speed of labor displacement through AI? I study this question in a microeconomic model in which AI providers sell access to users who must each complete a set of tasks using AI or labor. Users differ in the share of complex tasks they face. Using AI on complex tasks improves AI's ability to perform them. When AI initially has a comparative advantage in easy tasks, users with many complex tasks have low willingness to pay for AI. This creates a tension between profit maximization and complex-task learning. I characterize when this tension gives rise to a convex technological takeoff or a learning trap. Symmetric competition can slow adoption relative to monopoly when weak, but accelerates it when sufficiently intense by shifting usage toward more complex tasks. Asymmetric competition can further accelerate adoption by inducing endogenous specialization among providers.