Tacit Knowledge Extraction via Logic Augmented Generation and Active Inference

Researchers introduce a neuro-symbolic framework combining Logic-Augmented Generation and Active Inference to extract and formalize tacit knowledge into machine-interpretable Knowledge Graphs. The approach addresses a critical gap in knowledge engineering by capturing implicit assumptions and contextual expertise from procedural domains like manufacturing, demonstrated through analysis of assembly repair videos.

The challenge of formalizing tacit knowledge represents a fundamental bottleneck in knowledge engineering and AI systems deployment. Traditional approaches struggle because expert judgment, embodied skills, and contextual reasoning exist primarily in human minds rather than documented procedures. This research tackles the problem through a hybrid neuro-symbolic approach that leverages both neural language models and formal logic systems to bridge the gap between human expertise and machine-readable representations.

The framework's significance lies in its practical applicability to industrial domains where procedural knowledge directly impacts operational efficiency and quality. Manufacturing and repair processes exemplify this challenge—experts perform complex tasks intuitively based on years of experience, yet this knowledge transfers poorly to new workers or AI systems. By using instructional videos as proxy data, the researchers create a reproducible methodology that avoids relying solely on expert interviews or manual documentation, which often miss crucial contextual details.

For industry stakeholders, this work opens pathways toward automating knowledge capture and transfer at scale. Manufacturing facilities, maintenance operations, and other procedural-heavy sectors could dramatically reduce onboarding time and standardize quality by encoding expert tacit knowledge into queryable, reasoned-over systems. The improvement in completeness and semantic quality demonstrated suggests the approach handles real-world complexity better than purely neural or purely symbolic alternatives alone.

Looking ahead, the critical question is whether this methodology scales beyond controlled environments to capture the full depth of expert judgment in high-stakes domains. Success here could reshape how enterprises approach knowledge management and AI integration.

• →Neuro-symbolic framework successfully combines neural generation with formal logic to extract tacit knowledge from procedural domains.
• →Approach demonstrates improved completeness and semantic quality compared to existing knowledge engineering pipelines.
• →Uses instructional videos as reproducible proxy data, avoiding reliance on manual expert interviews alone.
• →Enables formalization of implicit assumptions, contextual constraints, and embodied skills into machine-interpretable Knowledge Graphs.
• →Addresses critical capability gap for manufacturing, maintenance, and other industrial sectors dependent on procedural expertise transfer.