Human–AI Task Allocation and Auditor Judgment: A U-Shaped Relationship Between AI Reliance and Audit Quality
Аннотация
Human–AI task allocation (HATA) is a central concept of the auditing process, an important determinant of audit quality. Sharing cognitive resources is a new collaborative framework based on human–machine interaction, which is one of the foundations of intelligent auditing. This paper proposes a quantitative model of task distribution resources considering the adaptive behavior of auditors. To this end, the quadratic regression model is applied to describe the relationship between the degree of AI reliance and auditors’ professional judgment, confidence level, and final decision consistency for the same type of audit scenario. Extending the analysis within the empirical framework both to the auditor as a decision-making agent and to AI systems, the study quantifies the curvature of the U-shaped relationship within the interaction domain on audit quality. In the experiment, two groups of auditors were analyzed for the same audit task; then, a round of evaluation was carried out after moving the allocation threshold, and two rounds of comparison were performed for two samples with the same complexity. Our results demonstrate that, at certain levels, moderate reliance on AI improves both individual and collective audit outcomes. Moreover, our results suggest that overreliance and underreliance on automation are detrimental to the consistency of professional judgment over time. In particular, the discussion has been extended further with reference to the need to implement transparent frameworks, high standards for the validation of algorithmic judgments across the auditing field and its regulatory bodies.