The effects of generative AI usage on employee knowledge behavior: a perspective based on the Automation–Augmentation Paradox
Citation. Liu, Z., Gao, C., & Yu, C. (2026). The effects of generative AI usage on employee knowledge behavior: a perspective based on the Automation–Augmentation Paradox. Frontiers in Psychology, 17, 1847724. https://doi.org/10.3389/fpsyg.2026.1847724
Summary. In a three-wave survey of 428 knowledge workers in Shanghai, generative-AI use predicted both more knowledge sharing and more knowledge hiding. The authors model these as competing appraisal pathways: AI use was associated with self-efficacy, which supported knowledge sharing, and with job insecurity, which supported knowledge hiding. The sharing association was larger in this sample, but the opposing indirect pathways were similar enough in magnitude that their total indirect effect was not significant.
Key claims
- Generative-AI use positively predicted knowledge sharing (β = 0.337, p < 0.001) and knowledge hiding (β = 0.100, p = 0.040); the authors’ knowledge-sharing-minus-hiding relative-intensity score was also positively predicted (β = 0.257, p < 0.001).
- AI use predicted self-efficacy (coefficient = 0.120, p = 0.007), whose indirect association with knowledge-behavior relative intensity was positive (bootstrap 95% CI [0.0040, 0.0544]).
- AI use predicted job insecurity (coefficient = 0.154, p = 0.001), whose indirect association with knowledge-behavior relative intensity was negative (bootstrap 95% CI [-0.0535, -0.0039]); the two indirect pathways were opposite in direction and similar in absolute magnitude.
- Competitive psychological climate strengthened both the AI-use → job-insecurity and AI-use → self-efficacy paths, rather than simply making the work environment uniformly more threatening.
Evidence & limitations
- Peer-reviewed Frontiers in Psychology original research article; Crossref confirms the published online date (2026-06-22), journal-article type, and CC BY 4.0 license.
- The study used three waves of self-report survey data from 428 knowledge workers in one Shanghai economic and technological development zone, analyzed with hierarchical regression and PROCESS bootstrap mediation; temporal separation strengthens ordering relative to a single-wave survey but does not establish causal effects of AI use.
- The authors’ Knowledge Behavior Relative Intensity is a difference score between distinct knowledge-sharing and knowledge-hiding measures. The six-factor CFA distinguished the measures, and ratio/winsorization checks were directionally consistent, but the net score should not erase the distinct behaviors or be treated as a settled construct.
- The observed stronger sharing association is a setting-specific result, not evidence that AI adoption reliably improves organizational knowledge flow, collaboration, or change support across organizations.