Notes on a sociology of generative knowledge: Human-AI epistemic stewardship

This article develops a sociology of generative knowledge for the age of AI. It treats contemporary systems as hybrid actors within sociotechnical networks and reframes classical anchors – Mannheim’s situatedness, Merton’s norms, and Latour’s distributed agency – for model-mediated inquiry. Generative knowledge is defined as knowledge organized to produce further knowledge through iterative, tool-mediated, and socially embedded processes. On this basis, the paper advances epistemic stewardship as a practical orientation that sustains human agency while harnessing computational acceleration. Stewardship is operationalized through transparency-by-design, provenance and traceability, calibrated trust, independent verification and red teaming, structured challenge routines, inclusivity in data and participation, and proportional delegation to machines. The account clarifies gains in discovery and education alongside risks from opacity, automation bias, feedback loops, and cognitive drift, and it specifies institutional reforms: standardized disclosure artifacts, replication triggers for AI-assisted claims, revised authorship taxonomies, and equitable access to compute, benchmarks, and community-governed datasets. A research agenda follows, calling for comparative evaluations of stewardship designs, longitudinal studies of hybrid practice, and field-specific protocols that link evaluation to adoption thresholds. The result is a framework that integrates social theory with design and governance guidance, aiming to convert acceleration into certified advance while keeping responsibility legible and contestable.