AI-mediated knowledge behavior

AI-mediated knowledge behavior tracks how AI changes workers’ discretionary choices to contribute, share, protect, withhold, or strategically deploy what they know. It is distinct from collaboration structure: a network can become denser while individual workers still protect scarce expertise, and an AI tool can lower the cost of articulating knowledge without changing whether sharing feels safe or worthwhile.

2026-liu-effects-generative-ai-knowledge-behavior gives the wiki an AI-specific dual-pathway case. Across three survey waves with 428 Chinese knowledge workers, greater generative-AI use predicted both more knowledge sharing and more knowledge hiding. The positive association with sharing was larger (β = 0.337 versus β = 0.100), but the explanatory value is not a simple net-positive story: self-efficacy and job insecurity carried opposing indirect pathways of similar magnitude. AI can therefore make knowledge easier to produce and share while simultaneously making the worker’s distinctive expertise feel more defensible.

The useful mechanism is appraisal, not the tool alone. In this study, AI use was associated with self-efficacy, which supported sharing, and with job insecurity, which supported hiding; competitive psychological climate strengthened both paths. A competitive setting does not simply make AI threatening or supportive. It can intensify the perceived opportunity to become more capable and the perceived risk that one’s value is becoming replaceable at the same time.

For implementation, this separates a real change-management question from generic adoption messaging. Organizations that want AI to improve collective knowledge flow need to measure whether workers see AI as expanding their ability to contribute, threatening the value of their expertise, or both. Training, tool access, and visible use are not evidence that discretionary knowledge contribution will follow, and claims that AI improves collaboration need to distinguish individual sharing/hiding behavior from observed team coordination or network ties.

Connections

• Connects to behavioral-human-centered-ai because challenge and threat appraisals can produce different knowledge behaviors under the same AI-use condition.
• Connects to employee-change-support because employees may comply with an AI rollout while still withholding the discretionary knowledge contribution and peer support needed for change to travel.
• Connects to ai-mediated-organizational-networks because knowledge sharing/hiding is a behavioral mechanism that can shape network ties, but it is not itself evidence that network structure changed.
• Connects to ai-mediated-teamwork because teams rely on knowledge contribution and expertise exchange, while the source does not test team performance, coordination, or trust.
• Connects to automation-and-substitution because perceived replaceability can translate an automation appraisal into defensive knowledge behavior before any job-level displacement occurs.

Contradictions & open questions

• 2026-liu-effects-generative-ai-knowledge-behavior found a larger sharing than hiding association, but this single organizational setting does not show that AI usually improves knowledge flow; its competitive-mediation result says the underlying empowerment and threat pathways can offset one another.
• The study’s Knowledge Behavior Relative Intensity difference score creates a useful integrative indicator, but knowledge sharing and knowledge hiding were empirically distinct constructs. Organizations should not treat a favorable net score as evidence that no hiding risk remains.
• Three-wave self-report data improve temporal separation but do not show whether AI use caused the appraisals or behavior, whether management interventions can change either pathway, or whether individual self-reports correspond to observed sharing, reuse, collaboration, or network centrality.
• Competitive psychological climate strengthened both pathways in one Chinese knowledge-worker sample. The boundary conditions across occupations, cultures, incentive systems, AI capability profiles, and change-management designs remain untested.