Complex human-AI tasks can lower employee engagement by raising tech-learning anxiety, but confidence with AI and humble leaders blunt the harm.

Main Finding

Human-AI collaboration task complexity reduces employee work engagement by increasing employees’ anxiety about learning AI-related skills (HAI-C tech-learning anxiety). AI self-efficacy (a context-specific personal resource) buffers this negative indirect effect, and humble leadership (a job resource) mitigates the harm indirectly by increasing employees’ AI self-efficacy. Results come from a three-wave longitudinal survey of 497 employees using hierarchical regression and bootstrapping.

Key Points

• The study is framed with the Job Demands–Resources (JD-R) model: HAI-C task complexity = job demand; humble leadership = job resource; AI self-efficacy = personal resource.
• Hypotheses tested:
  • H1: HAI-C task complexity positively → HAI-C tech-learning anxiety.
  • H2: HAI-C tech-learning anxiety negatively → work engagement.
  • H3: HAI-C tech-learning anxiety mediates the negative effect of HAI-C task complexity on work engagement.
  • Moderation: AI self-efficacy weakens the negative indirect effect; humble leadership strengthens AI self-efficacy and thus indirectly reduces the negative effect.
• Empirical results support the hypothesized mediation and moderated-mediation: task complexity → greater tech-learning anxiety → lower engagement; AI self-efficacy buffers this pathway; humble leadership promotes AI self-efficacy and thus alleviates the negative chain.
• The paper highlights that human-AI collaboration can be both motivating and depleting depending on contextual resources (links JD-R and challenge–hindrance stressor perspectives).

Data & Methods

• Design: three-wave longitudinal survey (matched responses across waves) to reduce common-method bias.
• Sample: 497 employees (matched across three waves; authors affiliated with a Chinese university—sample provenance not exhaustively described in the excerpt).
• Analysis: hierarchical regression and bootstrapping to test mediation and moderated-mediation effects.
• Key measured constructs: HAI-C task complexity, HAI-C tech-learning anxiety, work engagement (vigor/dedication/absorption), AI self-efficacy, humble leadership.
• Limitations noted implicitly: no detailed sectoral breakdown or causal identification beyond longitudinal survey; effect sizes and robustness tests not included in the excerpt.

Implications for AI Economics

• Productivity and human capital:
  • Increased AI-driven task complexity can reduce worker engagement (and likely productivity) unless organizations invest in resources that reduce learning anxiety or raise AI-specific self-efficacy.
  • Models of AI adoption should incorporate psychological frictions—learning anxiety and reduced engagement—as endogenous costs of deployment.
• Organizational investment decisions:
  • Firms face trade-offs: benefits from AI (efficiency, better decisions) versus costs from increased cognitive demands and engagement losses. Investments in targeted training and leadership development (humble leadership behaviors) can be cost-effective complements to technology adoption.
  • Measuring and valuing AI self-efficacy and learning-anxiety-reduction interventions should be part of cost–benefit analyses for AI rollout.
• Labor-market and distributional effects:
  • Heterogeneity matters: workers with low AI self-efficacy or in organizations lacking supportive leadership are more likely to suffer engagement losses—this can exacerbate productivity gaps and affect wages, turnover, and retraining needs.
  • Policy implications include subsidizing firm-level training, leadership programs, and certification pathways to reduce skill-upgrade frictions.
• Modeling suggestions for economists:
  • Incorporate non-pecuniary psychological costs (learning anxiety, engagement) into dynamic models of technology adoption and human capital accumulation.
  • Allow for complementarities between managerial practices (leadership style), firm investments in training, and worker-specific AI self-efficacy in production functions.
  • Consider endogenous adjustment of task complexity (task redesign or modularization) as a policy lever to reduce human costs of AI deployment.
• Empirical agenda:
  • Quantify how engagement losses translate into productivity, absenteeism, turnover, and wages.
  • Test heterogeneous treatment effects by skill level, sector, and country, and evaluate randomized or quasi-experimental interventions (training, leadership coaching, task redesign) to estimate returns on investments that raise AI self-efficacy or reduce tech-learning anxiety.

Limitations to keep in mind: the findings are survey-based (albeit longitudinal) and focus on psychological outcomes rather than direct productivity measures; generalizability across countries/sectors requires further empirical study.