Beliefs about the performance of new technologies strongly condition their usage. Among new technologies, artificial intelligence (AI)—and in particular LLMs—stand out because they can be used for extremely diverse sets of tasks. This versatility, however, complicates the problem of learning about AI capabilities, especially since the technological frontier created by AI is often "jagged": among seemingly similar tasks, AI can vastly out-perform humans on some, yet under-perform on others. Inaccurate perceptions of performance could lead to productivity losses, if AI is used for tasks on which it performs poorly.

In a controlled experiment, participants ( n=60) competed in a monotonous task with an autonomous robot for real monetary incentives. For each participant, we manipulated the robot's performance and the monetary incentive level across ten rounds. In each round, a participant's performance compared to the robot's would affect their odds in a lottery for the monetary prize. Standard economic theory predicts that people's effort will increase with prize value. Furthermore, recent work in behavioral economics predicts that there will also be a discouragement effect, with stronger robot performance discouraging human effort, and that this effect will increase with prize. We were not able to detect a meaningful effect of monetary prize, but we found a small discouragement effect, with human effort decreasing with increased robot performance, significant at the level. Using per-round subjective indicators, we also found a positive effect of robot performance on its perceived competence, a negative effect on the participants' liking of the robot, and a negative effect on the participants' own competence, all at . These findings shed light on how people may exert work effort and perceive robotic competitors in a human-robot workforce, and could have implications on labor supply decisions and the design of compensation schemes in the workplace.