Leveraging Large Language Models for Tacit Knowledge Discovery in Organizational Contexts
Citation
Zuin, G., Mastelini, S., Loures, T., & Veloso, A. (2025). Leveraging Large Language Models for Tacit Knowledge Discovery in Organizational Contexts. arXiv:2507.03811.
Core Framing
The paper reframes tacit knowledge capture as a network traversal problem rather than a knowledge conversion process. Instead of focusing on how to get experts to externalize knowledge (the SECI approach), they ask: can an LLM agent reconstruct complete knowledge by piecing together fragments distributed across an organization?
Key insight: You don’t need to find the domain expert (“patient zero”) if you can aggregate partial knowledge from people who received fragments through informal channels.
Key Findings
- LLM agent achieved 94.9% full-knowledge recall across 864 simulations
- Agent often reconstructed complete knowledge without ever contacting the original domain specialist
- Self-critical feedback scores correlated strongly with external evaluation metrics (0.73 Spearman correlation)
- Knowledge dissemination modeled as SI (Susceptible-Infectious) epidemic with waning infectivity
- Informal connections significantly improved knowledge retrieval quality
Methodology
- Simulated organizational hierarchies with varying parameters (depth, size, informal connections)
- Modeled knowledge spread using epidemic models
- Agent used prompt-chaining with self-critique loops
- Evaluated using METEOR, G-Eval, and self-critical scoring
Theoretical Grounding
- Weber’s bureaucratic organization theory (formal hierarchies)
- Blau’s social exchange theory (informal dynamics)
- Polanyi’s tacit knowledge concept
- Nonaka & Takeuchi’s SECI model
Atoms Extracted
- 05-atom—knowledge-epidemic-model
- 05-atom—patient-zero-independence
- 05-atom—self-critique-correlation
- 06-atom—waning-knowledge-infectivity
- 07-atom—informal-networks-knowledge-quality
Molecules Extracted
Organism Potential
Strong angle for an article connecting this to SECI framework limitations and the role of AI in tacit knowledge capture. The finding that you can reconstruct knowledge without finding the expert challenges conventional assumptions about knowledge management.