A Multi-Level Assessment Model for Generative AI's Occupational Impact
Year: 2026
Type: conference-paper
Abstract: Generative artificial intelligence is rapidly reshaping the global labor market and exerting highly heterogeneous impacts on different occupations. Quantifying these differentiated impacts has therefore become an urgent task. Existing methods mostly rely on static and single-dimensional evaluation. They fail to capture how augmentation effects shift into substitution effects as the technology evolves, and they rarely explain what drives occupational transformation, which leaves the analytical view incomplete. To address these limitations, this paper proposes a progressive Multi-Level Gen-AI Occupational Impact Assessment Model, named ML-OIAM. The core idea is to link static impact quantification, dynamic trend projection, and causal driver attribution into one complete analytical chain. ML-OIAM consists of three key components. First, we build a Weighted Multi-dimensional Impact Index Model that uses task-level data and applies the Analytic Hierarchy Process to quantify substitution risk, which improves both the accuracy and the interpretability of the assessment. Second, we design a scenario-based occupational trend projection model and introduce a time-varying coefficient to capture how augmentation effects transition into substitution effects under different AI development scenarios. Third, we apply a grey relational attribution model to identify the key factors driving occupational transformation, thereby filling the attribution gap under small-sample conditions. The three components work synergistically, enabling ML-OIAM to achieve both comprehensive evaluation and robust prediction. Experiments on three representative occupations demonstrate that ML-OIAM outperforms three mainstream baseline methods across all evaluation dimensions, and ablation experiments further verify the indispensability of each component.
Authors: Lingyu Ji, Yutong Zhou, Zhuohan Wang, Jian-Yu Li
Institution: Nankai University
Language: English
Subfield: Safety Research
Field: Social Sciences
Domain: Social Sciences
Open Access status: hybrid
OpenAlex ID: W7201869030