Automation has been a major driver of rising wage inequality in Spain: without task displacement the Gini would be roughly 21.5% lower, with losses concentrated in the middle of the distribution. AI exposure also raises inequality, but primarily by boosting top‑end wages rather than displacing middle earners.

Main Finding

The paper develops a task-sensitive, continuous "Unfixed automation" index and a distributional counterfactual method, and applies them to Spanish microdata (2000–2019). It finds that automation is a major driver of rising wage inequality in Spain: without task displacement captured by their preferred automation measure the 2019 Gini would be about 21.5% lower. Early AI exposure (2015–2019) is also associated with higher inequality (a 2019 Gini about 9.9% smaller in the absence of AI). Mechanisms differ: automation depresses wages mainly in the middle (and bottom) of the distribution, while AI raises wages at the top. Automation substantially widens the education premium but has little effect on the gender gap in the period studied.

Key Points

• New measurement: The Unfixed automation index is continuous and allows task routinization scores to change over time (it is sensitive to both employment shifts across occupations and changes in the degree of task routinization within occupations), addressing boundary and baseline problems of earlier indices.
• AI measures: The study uses two established AI exposure measures (Felten et al., 2018; Webb, 2020) to proxy early AI impacts while remaining agnostic on substitution vs augmentation.
• Quantified distributional impacts (Spain, 2000–2019):
  • Automation: absent task displacement ⇒ Gini ~21.5% lower in 2019.
  • Automation also: top 10% wage share would be ~3.9% smaller without automation; middle 50% share ~0.15% higher; bottom 10% share ~2.2% higher.
  • Education premium: absent automation, the 2019 education premium would be ~43% smaller.
  • AI (2015–2019): absence of AI ⇒ Gini ~9.9% smaller in 2019; AI tends to increase top-end wages.
• Heterogeneity:
  • Automation hits younger, low-educated workers hardest.
  • Older and medium/high-educated women are often neutral or positively affected.
  • Regional differences: Navarra, Aragón, Murcia show larger adverse automation impacts.
• Other factors (markups, trade, offshorability, employment rates, sectoral shares) are included and play secondary, period-specific roles; markups and “superstar” firm dynamics are important confounders to separate from pure technological effects.

Data & Methods

• Data: Spanish microdata of workers aged 25–65 covering 2000–2019. Units of analysis are demographic groups defined by age brackets, gender, education, and region.
• Regression strategy:
  • Dependent variable: change in log average wage for each demographic group between periods.
  • Key regressors: Unfixed automation index, AI exposure indices (Felten; Webb), plus controls (markups, trade openness, offshorability, employment rates, sectoral composition, education).
  • Time spans: main long-run 2000–2019; decade panel (2000–2010–2019); AI analysis focused on 2015–2019.
• Identification and robustness:
  • Instrumental variables: shift-share (Bartik) instruments for automation and AI following Ferri (2022) to address endogeneity and reverse causality.
  • Additional robustness: bounding procedure per Cinelli and Haznett (2020).
• Distributional counterfactual:
  • Use IV estimates to construct counterfactual wage distributions that remove the effect of a specific technology variable, then compare observed vs counterfactual using Gini, MLD, and wage shares (top10, middle 25–75, bottom50, bottom10).
• Measures of automation/AI:
  • Unfixed automation (new): continuous, time-varying task scores.
  • Classical RTI (Autor & Dorn) and Lewandowski et al. (2019) indices are discussed and compared.
  • AI: Felten et al. (2018) occupational impact score and Webb (2020) exposure score.

Implications for AI Economics

• Measurement matters: task-sensitive, continuous measures (like the Unfixed index) that allow task content to evolve are better suited to evaluate technological impacts on wage distributions than static occupation-to-task baselines. Researchers studying AI/economic inequality should adopt task-dynamic measures where possible.
• Heterogeneous mechanisms: automation and AI have distinct distributional footprints (automation → middle/low-wage pressure; early AI → top-wage gains). Policy and research should distinguish these mechanisms rather than treating “technology” as homogeneous.
• Policy directions:
  • Skill formation and retraining: findings reinforce the need for targeted upskilling/reskilling policies for middle-skill and low-skill workers vulnerable to automation.
  • Inclusive innovation: policies that capture and redistribute value from AI (taxation of rents, stronger wage bargaining, regulation of market power) may be necessary because technological gains can concentrate at the top.
  • Spatial and demographic targeting: because impacts vary by region, age, and education, place-based and cohort-specific interventions can be more effective than uniform policies.
• Empirical strategy advice:
  • Use shift-share (Bartik) instruments and robustness checks (e.g., bounding methods) to strengthen causal claims about technology and inequality.
  • Include controls for market power (markups) and globalization to avoid conflating technology-driven and market-structure-driven inequality.
  • Complement regression estimates with counterfactual distributional exercises to quantify the full distributional consequences, not only mean effects.
• Limitations and cautions:
  • Occupational task mappings assume similar task content across countries (authors map U.S. O*NET measures to Spain), which helps identification but may miss local task differences.
  • AI measures used capture exposure or current impact rather than the full range of possible future automation from advanced AI; results reflect early AI adoption dynamics.

Overall, the paper provides a replicable framework and empirical evidence that task-displacing automation has been a major contributor to rising wage inequality in Spain, while early AI exposure amplifies top-end gains—insights that are directly relevant for researchers and policymakers studying the distributional consequences of AI and automation.