Generative AI cuts users' motivation and perceived creative skills even as they rate their outputs the same; a randomized experiment shows GenAI assistance can undermine psychological resources that underlie creativity.

Main Finding

Using Generative AI (ChatGPT) in creative tasks led participants to report significantly lower intrinsic task motivation, lower self-assessed domain-relevant skills, and lower self-assessed creativity-relevant skills, while their self-evaluated creative performance did not change. In short: GenAI support coincided with declines in key psychological and skill-related components of self-perceived creativity, even though perceived performance remained stable.

Key Points

• The study tests GenAI’s human-side effects through the lens of Amabile’s Componential Theory of Creativity (intrinsic task motivation, domain-relevant skills, creativity-relevant skills).
• Contrary to some prior expectations, GenAI use decreased participants’ self-perceived:
  • intrinsic task motivation,
  • domain-relevant skills, and
  • creativity-relevant skills.
• Self-evaluated creative performance stayed unchanged despite the declines above.
• Proposed mechanisms (discussed in the paper): reduced autonomy or ownership, passive reliance on AI (lower cognitive engagement), and possible erosion of perceived competence when AI supplies solutions.
• The results highlight a tension in human–AI collaboration: technological augmentation can coexist with declines in the psychological resources that typically drive creativity.

Data & Methods

• Design: Controlled experimental within-subject study.
• Sample: 82 participants.
• Procedure: Each participant completed creative tasks both with and without ChatGPT assistance (within-subject comparison).
• Measures: Self-report scales capturing intrinsic task motivation, domain-relevant skills, creativity-relevant skills, and self-evaluated creative performance.
• Analysis: Statistical tests comparing the GenAI-assisted and non-assisted conditions (paper reports statistically significant decreases on the three component measures; self-evaluated performance did not differ).
• Limitations noted by authors: reliance on short-term laboratory experiment, modest sample size, and self-reported measures (limits on external validity and inference about long-term, objective skill change).

Implications for AI Economics

• Human capital and complementarities
  • GenAI may weaken workers’ perceived human capital (skills and motivation), even if immediate task outputs are maintained. Over time that could reduce effective human–AI complementarity and lower returns to human cognitive capital.
  • Firms that adopt GenAI should consider potential erosion of worker skills and motivation as an implicit cost of automation/augmentation.
• Productivity measurement and interpretation
  • Stable self-evaluated performance alongside declines in perceived skills/motivation suggests productivity metrics might mask underlying skill depreciation and morale effects. Short-term productivity gains could hide long-run declines in creative capacity.
• Labor market dynamics and wages
  • If workers perceive lower domain and creativity skills, this could affect career trajectories, bargaining power, and occupational choice—potentially influencing wage growth in creative occupations and increasing turnover or deskilling.
• Investment in training and job design
  • To sustain long-term innovation, firms may need to invest more in training, human-in-the-loop workflows, and incentive structures that preserve autonomy and active engagement with creative tasks.
  • Organizational redesign (task allocation, participatory AI integration, prompt engineering education) might be necessary to avoid passive reliance on GenAI.
• Policy and governance
  • Policymakers should consider interventions (e.g., training subsidies, guidelines for human-in-the-loop usage) to prevent skill erosion and preserve innovation capacity at the economy level.
  • Monitoring adoption externalities: widespread GenAI use could create negative externalities in worker skills and creative capabilities that merit public attention.
• Research and evaluation
  • Economic assessments of GenAI should include psychological and skill-related outcomes (not only output/productivity) to estimate true social returns.
  • Future cost–benefit and general equilibrium analyses of GenAI adoption should model potential dynamic declines in human creativity and human capital depreciation.

Suggested next empirical steps (for economists and managers): measure objective performance and long-run skill trajectories under repeated GenAI use; test heterogeneity by baseline skill level, task type, and organizational supports; and evaluate interventions (training, autonomy-preserving interfaces) that might mitigate the negative self-perception effects documented here.