Workers who believe they face automation are more likely to seek training, but information nudges only work when they hit home; in a public administration, simple behavioural fixes to email and meetings cut excess meetings and reduced sickness absence, implying modest productivity and well‑being gains.

Main Finding

Workers’ beliefs about the likelihood their tasks will be automated matter for who takes up training, and simple, well-framed labor-market information can shift training engagement — but its effectiveness depends on perceived relevance, credibility, and worker characteristics (notably age and who initiates training/funding). Aligning workers’ automation perceptions with objective risk and tailoring information interventions raises the effectiveness of policies aimed at reskilling in the face of technological change.

Key Points

• There is an “automation training gap”: mismatches between expert-assessed occupational automation risk and workers’ self-assessed automation exposure are associated with differences in training participation.
• Workers who under-estimate automation risk are less likely to engage in training that could mitigate displacement; over-estimation can also alter engagement patterns.
• Field experiments show that providing framed labor-market information increases short-run indicators of training engagement, but impacts are heterogeneous:
  • Age matters: younger and older workers respond differently to framings.
  • The source, framing, and perceived relevance/credibility of the information condition whether workers act on it.
• Information interventions are conditionally effective — relevance to the individual’s occupation and acceptance of the message are key mediators.
• Funding type and training initiator (employer vs. employee vs. public programs) interact with both perceived risk and the effectiveness of information/take-up.

Data & Methods

• Observational analysis (Chapter 2):
  • Data source: ROA Lifelong Learning Survey (detailed responses on training participation, self-assessed automation risk, and individual/occupational characteristics).
  • External measure: expert-assessed occupational automation risk linked to occupations.
  • Methods: descriptive statistics and regression analyses comparing expert-assessed risk and self-assessed risk as predictors of training participation; heterogeneity checks across funding types and initiators; robustness checks reported.
• Field experiments (Chapters 3 and 4):
  • Randomized controlled designs delivering framed labor-market information to participants (samples drawn via administrative/employment-agency channels and/or survey panels as described).
  • Treatments varied framing (e.g., risk framing, opportunity framing), specificity (occupation-level vs. aggregate), and source credibility cues.
  • Outcomes included measures of training engagement (intentions, information-seeking, sign-ups/registrations or similar short-run behavioral indicators), and measures of information acceptance/relevance.
  • Analyses include intent-to-treat comparisons, subgroup heterogeneity (age, occupation), and checks for randomization balance and robustness.

Implications for AI Economics

• Worker beliefs are a behavioral friction in the labor-market adjustment to AI and automation. Simply expanding training supply is insufficient if workers do not perceive the need.
• Policy levers:
  • Invest in credible, occupation-specific labor-market information campaigns that explicitly address perceived relevance to workers’ tasks.
  • Use behavioral framing and trusted messengers to increase acceptance; tailor messages by age and other observable heterogeneities.
  • Combine information with incentives (subsidies, employer mandates, tailored outreach) because conditional effectiveness implies information alone may be weak for some groups.
• Measurement and monitoring:
  • When modeling labor-market impacts of AI, incorporate belief heterogeneity — not only objective automation risk but also subjective risk perceptions that shape training demand and transition speeds.
  • Evaluate retraining programs by accounting for who initiates and funds training; these institutional features moderate responses to both technology shocks and information interventions.
• Organizational responses:
  • Firms and public agencies should consider proactive communication and on-the-job reskilling pathways; internal signals about role changes can be as important as external labor-market signals.
• Research agenda:
  • Further work should quantify long-term impacts of information + incentive interventions on re-employment, wage trajectories, and occupation transitions under AI adoption scenarios.
  • Deeper investigation into which framings and messenger types scale best across sectors will improve cost-effective policy design.

If you want, I can (a) extract and summarize results and effect sizes from specific chapters/appendices if you can provide the relevant pages or tables, or (b) draft policy recommendations tailored to a specific country or sector (e.g., manufacturing, services, public sector).