Experimental Evidence That AI-Managed Workers Tolerate Lower Pay Without Demotivation

TL;DR

This study addresses how AI-managed work affects worker responses by creating a high-fidelity, ecologically valid Minecraft-based workplace that tracks real-time behavior and ties wages to AI- or human-generated evaluations. It compares four management conditions, including two AI evaluators (rule-based AI and a decision-tree AI) and a human manager, plus a human supervisor informed by AI recommendations. Key findings show that a rule-based AI downgrades evaluations and wages by about 40% without reducing motivation, while a more aggressive AI can undermine perceived fairness and motivation; human-delivered AI recommendations can preserve fairness alignment. The work highlights a potential silent exploitation risk in algorithmic management and underscores the need for auditing AI training data and governance to prevent erosion of worker rights as such systems scale.

Abstract

Experimental evidence on worker responses to AI management remains mixed, partly due to limitations in experimental fidelity. We address these limitations with a customized workplace in the Minecraft platform, enabling high-resolution behavioral tracking of autonomous task execution, and ensuring that participants approach the task with well-formed expectations about their own competence. Workers (N = 382) completed repeated production tasks under either human, AI, or hybrid management. An AI manager trained on human-defined evaluation principles systematically assigned lower performance ratings and reduced wages by 40\%, without adverse effects on worker motivation and sense of fairness. These effects were driven by a muted emotional response to AI evaluation, compared to evaluation by a human. The very features that make AI appear impartial may also facilitate silent exploitation, by suppressing the social reactions that normally constrain extractive practices in human-managed work.

Figures (2)

• Figure 1: The experiment interface. An example of the interface for workers under AI management. Workers complete the iron pickaxe task for three repeated rounds, each lasting 10 minutes.
• Figure 2: Summary of the main results (worker N = 382). Both self-assessment and manager evaluation were implemented at three levels: beginner ($0), intermediate ($0.2), and advanced ($0.5). Each level corresponds to a different bonus, as shown in the parentheses. Fairness scores aggregate procedural and distributive fairness; segmenting the two dimensions yielded similar results.