Simulation finds construction workers' safety behavior worsens as they work more with AI

The researchers built an agent-based model grounded in stimulus-organism-response theory. Simulated workers moved in a bounded two-dimensional space, and the model tracked their cognitive, emotional and behavioral responses to AI systems and colleagues.

The authors say the results point to cumulative cognitive and emotional effects of working with AI. They propose the model as a basis for deciding where human-AI boundaries should sit and as a tool for long-term safety decisions on construction sites.

The study addresses a gap the authors describe: although protecting worker well-being in human-AI interaction is a shared goal, there has been no quantifiable way to capture its effects on workers. Their answer is a simulation framework that tracks how AI exposure shapes workers' thinking, feelings and conduct.

The experiments covered three scenarios. The first used repeated runs with balanced parameter settings, the second used batch runs with varied settings, and the third checked the model against a real smart construction site.

Across the scenarios, safety behavior worsened as interaction with AI grew deeper. The authors say this reflects cumulative cognitive and emotional effects, and they suggest managers could use the model to define limits on human-AI interaction.