The generative AI duality trap: a knowledge governance perspective on knowledge diversity and epistemic convergence in AI-augmented organisations
Hui Chun Chan, Kuo Ming ChuPurpose
This study aims to examine a knowledge management (KM) puzzle: why generative artificial intelligence (GenAI) use relates to knowledge diversity expansion in some organisations and epistemic convergence in others. The authors develop the GenAI duality trap, a cross-level knowledge governance framework in which artificial intelligence (AI)-mediated knowledge outcomes track governance design, not technological capability alone.
Design/methodology/approach
A multi-source design combines organisational surveys (n = 67), employee measures (n = 1,247) and sentence-bidirectional encoder representations from transformers(SBERT)-derived semantic diversity indices. Partial least squares structural equation modelling (PLS-SEM) and partial least squares multi-group analysis (PLS-MGA) test nine hypotheses across three institutional regimes (Bradford, 2023), with cluster-robust and multilevel diagnostics for the nested data.
Findings
Knowledge governance relates to diversity through human–AI interaction quality. Transparency, decentralised access and AI literacy investment are associated with critical engagement, which, in turn, relates to perceived and semantic diversity. Institutional regime moderates all three pathways, most weakly for access, with associations strongest in market-driven and attenuated in state-driven regimes.
Research limitations/implications
Knowledge governance theory is extended into the GenAI era, with interaction quality as the micro-mechanism associated with knowledge outcomes and natural language processing (NLP)-derived semantic diversity complementing perceptual measures. The cross-sectional design constrains causal inference, and epistemic convergence is inferred from comparatively low diversity rather than measured as a distinct longitudinal process.
Practical implications
KM managers may treat AI governance as a knowledge-variety lever: transparency, distributed access and literacy investment relate to whether GenAI use accompanies enrichment or compression of the knowledge base.
Originality/value
The paper introduces the duality trap framework, among the first cross-level knowledge governance models linking institutional regimes, organisational AI governance and human–AI interaction to knowledge diversity and demonstrates computational diversity metrics for KM against recent evidence on GenAI-related homogenisation (Anderson et al., 2024; Doshi and Hauser, 2024; Padmakumar and He, 2024).