Correcting AI for innovation: how networks shape knowledge correction capability in hybrid systems

Purpose This study aims to address the gap that AI-generated content (AIGC) often contains systematic errors, undermining the assumption that external knowledge can be directly absorbed. We conceptualize knowledge correction capability as an AI-specific, upstream mechanism with depth-oriented (within-domain precision) and breadth-oriented (cross-domain validity) dimensions, and examine how network embeddedness enables them to accelerate new product development (NPD). Design/methodology/approach Drawing on Hybrid Intelligence, we test a conceptual model with survey and archival data from 248 publicly listed manufacturing firms in China. Two forms of external embeddedness – trade associations and nonprofits/NGOs (NPONGOs) – are related to depth/breadth correction and NPD speed using structural equation modeling. Findings In this research paper we find that both depth- and breadth-oriented correction positively relate to NPD speed. Association embeddedness primarily strengthens depth-oriented correction through vertical standards and professional specialization, whereas NPONGO embeddedness fosters breadth-oriented correction via cross-domain knowledge integration and plural references. Breadth-oriented correction plays a comparatively stronger mediating role between network embeddedness and innovation speed. Practical implications This paper aims to implement human-in-the-loop validation checklists, configure a dual network portfolio (associations for vertical precision; NPONGOs for cross-domain validity), and institutionalize feedback loops that feed correction experience into prompts, datasets, and knowledge bases. Originality/value This study advances a correction-first view by positioning knowledge correction as a pre-absorption micro-foundation distinct from absorptive capacity, and reframes network embeddedness as a cognitive calibration system, enhancing the visibility and correctability of AI-generated errors.