Website Paper
Fields
Behavioral & Experimental, Innovation
Job market paper
Better at the Task, Worse Off for Choosing It: Generative AI, Task Misallocation, and Learning
While the adoption of new technologies promises productivity gains, their effects on labor allocation depend on whether workers correctly anticipate their own returns to using them. We study this margin in a laboratory experiment with generative AI. Participants choose between a complementary task and a non-complementary one under randomized access to AI. AI access increases selection into the complementary task but also raises earnings-based misallocation: many participants who choose it improve on their own Baseline performance in that task, yet earn less than they had earned in the other task at Baseline. We focus on unexpected task misallocation: cases in which the participant expected the complementary task to earn at least as much as the alternative, but ultimately earned less. Its incidence rises from zero in Control to 19 percent under AI access. Two learning treatments—social information and repeated task selection—are associated with a substantially lower incidence of this outcome, consistent with a role for learning about AI-assisted returns. These findings highlight a friction in early technology adoption: a worker can become more productive in a task the technology assists and still earn less by choosing it.