Human workers as physical AI data generators: A socio-technical governance model for sustainable manufacturing transformation

The convergence of Physical Artificial Intelligence (Physical AI) with labour-intensive manufacturing has created expanding demand for high-quality recordings of human physical activity, with the market projected to grow from roughly US$4 billion in 2024 toward US$60 billion by the mid-2030s. Yet the prevailing approach to acquiring this data remains surveillance-oriented and productivity-centred, neglecting the socio-technical dimensions on which both worker welfare and dataset quality depend. This review re-frames manual workers not as passive subjects of monitoring but as intentional Physical AI Data Generators, whose tacit knowledge, ergonomic state and psychological experience constitute foundational data assets. Drawing on a PRISMA 2020-guided synthesis of 187 sources across industrial engineering, human-factors engineering, computer vision, occupational psychology and AI ethics, the study develops the Socio-Technical Physical AI Data Governance Framework, which integrates productivity metrics, ergonomic-risk assessment and surveillance perception within a single decision engine coupling the Fuzzy Best-Worst Method, Bayesian-network inference and Fuzzy Failure Mode and Effects Analysis. The evidence indicates that electronic performance monitoring is reliably associated with elevated worker strain; that computer-vision and wearable pipelines can automate RULA, REBA and OWAS scoring but degrade under occlusion; and that a human-centred governance configuration is expected to outperform conventional surveillance across productivity, ergonomic, trust, data-quality and ethical-compliance dimensions. The principal contribution is the articulation of Physical AI Data Governance as a distinct domain with an operationalised variable taxonomy and integrated methodology. The framework is conceptual and requires empirical validation but offers an actionable basis for human-centred data collection across manufacturing sectors and enterprise scales.