Manufacturing experts see AI as a partner, not a replacement: safety, trust and data governance—especially worker participation in data processes—determine whether AI yields real production gains.

Main Finding

Experts view human–AI collaboration in manufacturing as a mutually supportive partnership: AI contributes analytical, pattern-detection, and automation capabilities while humans supply domain expertise, contextual judgment, and oversight. In manufacturing this partnership is shaped strongly by safety concerns and by data practices—workers' willingness to share knowledge, their participation in data processes, and organizational data governance are key determinants of whether AI delivers production improvements.

Key Points

• Human-centric framing: The collaboration aligns with Industry 5.0’s emphasis on human-centered, resilient socio-technical systems rather than full automation.
• Complementarity: AI and workers are seen as complementary — AI handles data-intensive tasks and pattern extraction; humans provide tacit knowledge, exception handling, and ethical/safety judgments.
• Safety as a defining constraint: Hazardous machinery, complex shop-floor interactions, and regulatory requirements make safety a priority that shapes deployment, oversight, and acceptability of AI tools.
• Benefits conditional on trust and data quality: Expected production and process improvements depend on worker trust in AI outputs and on high-quality, representative, and well-governed data.
• Data practices matter: Worker willingness to share domain knowledge, participation in labeling/annotation or feedback loops, and organizational data governance (access, quality control, privacy, incentives) are central to successful human–AI collaboration.
• Social and organizational barriers: Cultural, incentive, and governance issues (not just technical performance) can limit adoption and effective use of AI on the shop floor.

Data & Methods

• Approach: Qualitative interview study with experts in manufacturing and related domains, analyzed through a socio-technical lens.
• Focus: Perceptions of human–AI collaboration, perceived benefits, challenges, and contextual factors (safety, data practices, trust, organizational governance).
• Limitations inherent to method: Findings reflect expert perceptions and thematic interpretation rather than quantitative causal estimates; generalizability is limited and results are hypothesis-generating.

Implications for AI Economics

• Complementarity and returns to skill: The findings support models where AI increases returns to certain human skills (tacit knowledge, supervision, safety expertise), implying complementarities that influence wage and task structure in manufacturing.
• Adoption frictions and diffusion: Non-technical frictions—trust, worker willingness to contribute data, governance arrangements—are important adoption costs. These frictions can slow diffusion and reduce realized productivity gains relative to technical potential.
• Data as an economic input and coordination problem: High-quality labeled operational data and ongoing human participation are inputs with transaction costs and incentive problems. Firms face investment choices in data collection, annotation, and governance that affect the marginal productivity of AI.
• Measurement challenges: Standard productivity statistics may miss gains concentrated in safety, quality, downtime reduction, or tacit-process improvements; economic evaluation should include these dimensions.
• Policy and firm strategy implications:
  • Invest in training and human capital that complements AI (supervisory, interpretive, safety management skills).
  • Design incentives and governance to encourage worker participation in data processes (compensation, co-design, transparency).
  • Regulate and support safety standards and liability clarity to reduce firm and worker risk aversion to AI use.
  • Support infrastructure for data quality and sharing while protecting privacy and intellectual contributions of workers.
• Research directions: Quantify the productivity impacts of human–AI complementarities in manufacturing, estimate the costs of data acquisition and governance, model incentives for worker data sharing, and evaluate how safety regulation and liability regimes alter adoption and returns.