加载中...
This field-experiment paper shows that making workers’ AI reliance observable to an HR evaluator causally reduces their adoption of AI recommendations and lowers task performance: observed switching to AI falls ~30.5% → 26.2% and final accuracy declines ~79.1% → 76.4%, implying image concerns meaningfully block human–AI complementarities; workers report they fear reliance signals low confidence in their judgment, and simple informational reassurance does not eliminate the effect .
| Outcome | Control (mean) | Treatment effect (Public pooled) |
|---|---|---|
| AI recommendation reliance (all rounds) | 0.305 | -0.043*** (SE 0.014) |
| AI reliance (conditional on disagreement) | 0.640 | -0.080*** (SE 0.022) |
| Initial accuracy (before AI) | 0.553 | +0.008 (ns) |
| Final accuracy (after AI) | 0.791 | -0.027*** (SE 0.008) |
| Initial response time (sec) | 21.3 | +2.10* (SE 1.10) |
| Notes: Standard errors clustered at worker level; significance: * p<0.10, ** p<0.05, *** p<0.01. These estimates show lower AI reliance and lower final accuracy under visibility . |
元数据与概览(引言与报告概览)
逐节深度解读
图表与图形深度解读(逐图解析,与原文图像对应展示)
估值分析(无)
风险因素评估(由报告识别或隐含)
批判性视角与细微差别(审慎观点)
结论性综合(关键发现与政策含义)
补充材料与重要附图(选录)
最後的评估性总结(客观)
若您希望,我可以:
请选择您想优先深入的部分(如:表格逐项解析、模型数学细节证明的逐行释义,或关于如何将该发现迁移到特定行业/组织的政策建议)。
2511.18582.pdf