Women view workplace AI as riskier than men and withdraw support faster as the chance of employment gains falls; higher female risk aversion and greater exposure to AI-related risks partly explain the gap, suggesting unchecked AI policy could deepen gender inequalities and provoke political backlash.

Main Finding

Women are more skeptical than men about AI adoption in the workplace because they (1) are, on average, more risk averse and (2) face greater exposure to AI-related risks. Experimental variation shows this relationship is causal: when the probability of net positive employment effects of AI falls, women’s support for firms adopting AI declines more sharply than men’s. Open-ended responses confirm women express greater uncertainty about AI’s benefits and more often expect little or no benefit.

Key Points

• Consistent gender gap: women perceive AI as riskier than men across a large North American sample.
• Two primary drivers identified:
  • Higher general risk aversion among women.
  • Greater direct or perceived exposure to AI-related risks (e.g., job displacement, wage pressure).
• Causal evidence from an experiment: lowering the stated likelihood of net-positive employment effects reduces women’s support for AI adoption more than men’s.
• Qualitative topic modeling: women’s open-ended comments show more uncertainty and emphasize limited or negative benefits.
• Policy relevance: ignoring gender-specific perceptions and exposures could worsen existing labor market inequalities and provoke political backlash to AI deployment.

Data & Methods

• Data: original survey of ~3,000 respondents in Canada and the United States.
• Measures:
  • Perceived risk of AI adoption.
  • General risk aversion (standard survey measures).
  • Self-reported or inferred exposure to AI-related workplace risks.
  • Open-ended questions about expected benefits/harms.
• Causal test: experiment embedded in the survey that manipulates the probability that AI will have net positive employment effects; interaction analysis shows differential gender response.
• Text analysis: structural topic modeling applied to open-ended responses to identify themes by gender.
• Identification strategy: combination of observational decomposition (to attribute gap to risk preferences and exposure) and experimental variation (to establish causal link between perceived employment risk and support for AI).

Implications for AI Economics

• Distributional effects: gender differences in attitudes and exposure imply AI adoption may have unequal welfare consequences and could amplify gender gaps in employment and income if rollout and compensation are not gender-aware.
• Policy design:
  • Incorporate gender-differentiated risk assessments into AI impact evaluations.
  • Targeted retraining, searching assistance, and income-support policies for groups with higher exposure.
  • Transparent communication of employment-impact evidence to reduce uncertainty and mitigate asymmetric backlash.
  • Include women in AI governance and workplace decision-making to surface and address exposure concerns.
• Political economy: failing to address gender-specific risks may reduce public and political support for AI adoption, slowing beneficial diffusion or prompting restrictive regulation.
• Research priorities: measure industry- and occupation-level heterogeneity, track long-run labor outcomes by gender, and test interventions (e.g., tailored information, compensation schemes) to narrow the gender gap in AI acceptance.