Abstract
Contemporary organizations operate in environments characterized by normative complexity, information overload, and demands for rapid, auditable decisions. In this context, organizational learning and knowledge management become central to institutional governance, particularly in public administration, judiciary, and healthcare. Although AI systems based on large language models (LLMs) have transformed how knowledge is produced and applied, their integration into organizational learning processes remains theoretically underdeveloped. Conventional approaches treat AI as an isolated query interface, lacking mechanisms for epistemological control, institutional traceability, or cumulative learning, generating risks of decision-making opacity and information hallucinations.
This paper addresses how to integrate AI systems into organizational learning in a reliable, auditable, and institutionally responsible manner. Drawing on organizational learning theory, the knowledge-based view of the firm, the SECI model, and sociotechnical systems theory, the work proposes a computational framework articulated across five dimensions: (1) an institutional knowledge layer with versioned repositories and normative validity metrics; (2) a semantic retrieval layer structuring knowledge for contextual queries; (3) multiple specialized cognitive agents acting as supervised organizational specialists; (4) a supervision and validation layer verifying consistency and normative compliance; and (5) an organizational memory layer preserving decisions, justifications, and corrections for longitudinal reuse.
Methodologically, the work combines a computational artifact with a case study in a Brazilian public institution, where the system processed large volumes of normative and organizational documents. Evaluation encompasses epistemic quality of responses, measured by consistency, traceability, and reduction of hallucinations; organizational impact, assessed through analytical efficiency and decision support; and organizational learning capacity, observed through memory accumulation and progressive performance improvement.
Preliminary results indicate that the framework improves efficiency in accessing institutional knowledge while introducing cognitive governance mechanisms that transform AI interactions into cumulative learning processes. Theoretical contributions include a reconfiguration of LLMs as components of sociotechnical organizational learning systems. Practically, the framework offers a replicable model for organizations seeking to integrate AI responsibly, with potential impact on public policies, information risk management, and critical decision-making processes.