💻 computer science

Inteligencia artificial y empleo en España: una aproximación territorial y de género a la exposición laboral

This paper proposes a sector-based methodological framework to assess AI's labor exposure in Spain from 2021 to 2023, revealing that metropolitan and service-oriented regions face higher risks and that female employment consistently exhibits greater exposure across all territories.

Original authors: Antoni Mestre, Xavier Naya, Manoli Albert, Vicente Pelechano

Published 2026-05-06

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Antoni Mestre, Xavier Naya, Manoli Albert, Vicente Pelechano

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Picture: A "Weather Map" for Jobs

Imagine the Spanish job market as a vast landscape. For years, people have worried about how Artificial Intelligence (AI) might change this landscape. Will it rain on some jobs and leave others dry? Will it flood certain cities while others stay safe?

This paper, written by a team of researchers, doesn't try to predict the future with a crystal ball. Instead, they built a "Weather Map" (a methodology) to measure how much of the current job market in Spain is exposed to the rain of AI. They aren't saying AI will definitely replace these jobs tomorrow; they are simply measuring how much of the work done in these jobs could be done by AI based on the nature of the tasks.

The Tool: Bridging Two Different Languages

The researchers faced a tricky problem. International studies usually talk about jobs in terms of specific roles (like "Accountant" or "Software Developer"). However, Spain's official statistics talk about jobs in terms of industries (like "Retail," "Healthcare," or "Construction").

It's like trying to compare a recipe for a cake (the international studies) with a list of grocery stores (the Spanish data). You can't easily mix them.

To fix this, the team built a translator bridge (a matrix). They took the international data on which tasks AI can do and mapped them onto Spain's industries. Now, they could look at a province (like Madrid) and say, "Because this province has so many people working in Retail and Administration, it has a high 'AI Exposure Score'."

The Findings: Where the Rain is Heaviest

1. The "City vs. Countryside" Divide
The map shows a clear pattern. The "rain" of AI exposure is heaviest in the big, bustling cities and coastal tourist hubs.

High Exposure Areas: Madrid, Barcelona, Valencia, Málaga, and the islands (Baleares and Canary Islands).
Why? These places are full of "brainy" jobs: offices, banks, shops, hotels, and schools. These jobs involve a lot of writing, organizing, talking to customers, and processing data—things AI is very good at.
Low Exposure Areas: Rural provinces in the interior (like Soria, Teruel, or Zamora).
Why? These areas rely more on farming, traditional manufacturing, and construction. These jobs involve physical labor, moving heavy objects, or working with machines in ways that current AI (which is mostly "digital brain" power) cannot easily touch.

2. The Gender Gap: A Structural Imbalance
One of the most striking findings is that women are consistently more exposed to AI than men across the entire country.

The Analogy: Imagine two different types of boats. Men, on average, are working on boats that are built with wood and steel (construction, transport, heavy industry)—these are harder for AI to replace. Women, on average, are working on boats made of paper and glass (education, healthcare, administration, retail)—these are more transparent and easier for AI to see through and influence.
The Result: In every single province, the "AI Exposure Score" for women is higher than for men. This isn't because women are less skilled; it's because they are concentrated in sectors where AI is currently most applicable.

3. Stability: The Map Doesn't Change Much
The researchers looked at data from 2021 and 2022. They found that the map looked almost identical in both years.

The Takeaway: This isn't a sudden storm that hit overnight. The "exposure" is built into the very structure of the economy. A province doesn't become "high risk" just because of a bad month; it's high risk because its economy has been built that way for decades. The structure is solid, and it changes very slowly.

What This Means (and What It Doesn't)

The authors are very careful to say: "This is a map of potential, not a prediction of doom."

Exposure $\neq$ Replacement: Just because a job is "exposed" to AI doesn't mean the person will lose their job. It might mean they get a "super-powerful assistant" (like a very smart co-pilot) that helps them work faster.
No Crystal Ball: The paper admits it is a "preliminary" study. It's a first draft. It doesn't know how fast companies will actually buy AI, or how laws will change, or how people will adapt.

The Bottom Line

This paper provides a compass for Spain. It tells us that:

Geography matters: Big cities and service hubs are the front line of AI change.
Gender matters: Women are structurally more likely to be in roles that AI can touch, which requires special attention to ensure they aren't left behind.
Structure matters: The economy is built on deep foundations. We can't expect the landscape to change overnight, but we can use this map to start planning how to build better shelters (training, policy, and support) for the people living in the areas where the "AI rain" is heaviest.

In short, the researchers have handed us a flashlight to see where the shadows are, so we can start preparing for the light of the future.

Technical Summary: Territorial and Gender Perspectives on Artificial Intelligence and Employment in Spain

1. Problem Statement

The rapid expansion of Generative Artificial Intelligence (AI) has fundamentally reconfigured debates regarding the future of work, productivity, and social cohesion. While international literature has established robust methodologies for measuring occupational exposure to AI (e.g., using SOC, ISCO, or ESCO classifications), a significant methodological gap exists for the Spanish context. Spain's official labor statistics are primarily disaggregated by economic activity sectors (CNAE) at the provincial and autonomous community levels, rather than by specific occupations.

This misalignment prevents the direct application of international AI exposure indices to Spanish regional data, hindering the development of targeted public policies, union strategies, and territorial planning. Furthermore, existing literature often overlooks the intersection of territorial disparities and gender-specific vulnerabilities, despite evidence suggesting that women are overrepresented in administrative and service sectors highly susceptible to AI automation. The paper addresses the urgent need for an objective, comparable tool to measure AI exposure in Spain that respects its specific statistical structure (CNAE) and allows for granular territorial and gender analysis.

2. Methodology

The study proposes a preliminary, replicable methodology to bridge the gap between international occupational AI metrics and Spanish sectoral data. The approach involves three main steps:

2.1. Construction of the AI–CNAE Incidence Matrix

The authors adapt the "AI Applicability Score" from Tomlinson et al. (2025), which is based on over 200,000 real-world interactions with generative AI tools (e.g., Microsoft Copilot). Instead of using occupational codes directly, the authors construct a mapping matrix that translates these occupational capabilities into the Spanish CNAE (2-digit) classification.

Factor Assignment: Each CNAE sector is assigned an "incidence factor" ranging from 0.06 to 0.30.
Logic: Sectors dominated by manual, physical, or routine tasks (e.g., Agriculture, Construction) receive low factors. Sectors characterized by administrative, linguistic, analytical, or synthesis tasks (e.g., Finance, Professional Services, Information & Communications) receive higher factors.
Synthesis: The matrix represents a weighted synthesis of international evidence and the specific occupational composition of Spanish sectors.

2.2. Data Sources and Calculation Model

The study utilizes microdata from the Spanish National Statistics Institute (INE) for the years 2021 and 2022, specifically the Annual Census of Occupied Population, disaggregated by province, sector (CNAE), and sex.

Sectoral Exposure: For each province $p$ and sector $c$ , the "AI-adjusted employment" is calculated as:
$IA\_employment(p,c) = employment(p,c) \times factor_{IA}(c)$
Territorial Indicator: The aggregate exposure for a province is defined as the ratio of total AI-adjusted employment to total provincial employment:
$IA\text{-}share(p) = \frac{\sum_{c} IA\_employment(p,c)}{\sum_{c} employment(p,c)}$
Gender Disaggregation: The same calculation is performed separately for male and female employment to derive gender-specific exposure indices ( $IA\text{-}share_{female}$ and $IA\text{-}share_{male}$ ).

2.3. Limitations

The authors explicitly state that this is a preliminary study measuring potential structural exposure, not actual substitution or job loss. The model assumes a homogeneous occupational composition within each CNAE sector, which may mask internal heterogeneity. It does not account for dynamic effects such as the creation of new tasks, productivity gains, or the speed of corporate adoption.

3. Key Contributions

Methodological Adaptation: The paper provides the first systematic translation of international AI exposure metrics into the Spanish CNAE framework, enabling provincial-level analysis where previously only national or broad regional estimates were possible.
Territorial Cartography: It generates a detailed map of AI exposure across all 52 Spanish provinces, identifying structural vulnerabilities based on economic specialization.
Gender Analysis: It quantifies a persistent structural gap in AI exposure between men and women, revealing that the female workforce is systematically more exposed to AI-driven transformation across all territories.
Policy Tool: It offers a transparent, open methodology for unions, policymakers, and researchers to monitor trends and design "just transition" strategies aligned with the National AI Strategy (ENIA).

4. Results

The analysis of 2021–2022 data reveals several consistent patterns:

4.1. Aggregate Exposure

Spain exhibits a moderate-to-high aggregate exposure to AI, with the "IA-share" oscillating between 18% and 22% of total employment. This indicates that a significant portion of the Spanish labor market is engaged in tasks where generative AI has high applicability.

4.2. Territorial Polarization

A clear "center-mediterranean corridor" of high exposure is identified, including Madrid, Barcelona, Valencia, Alicante, Málaga, and the Canary Islands (Las Palmas and Santa Cruz de Tenerife). These provinces exceed 21% exposure.

Drivers: High exposure correlates with a concentration of advanced services, finance, professional consulting, information/telecommunications, and retail/commerce.
Low Exposure: Provinces in the interior (e.g., Soria, Teruel, Zamora, Palencia) and those with strong agricultural or traditional manufacturing bases show lower exposure (approx. 17.5%–18.5%).
Stability: The ranking of provinces remained remarkably stable between 2021 and 2022, confirming that exposure is driven by deep-seated structural economic factors rather than short-term cyclical fluctuations.

4.3. Gender Gap

In every single province, female employment shows a higher exposure to AI than male employment.

Magnitude: The gap generally ranges from 1.5 to 3 percentage points, though it is wider in highly tertiaryized regions.
Causes: Women are disproportionately concentrated in sectors with high AI applicability, such as education, healthcare, administration, and social services. Conversely, men are more represented in construction, transport, and extractive industries, which have lower AI applicability factors.
Implication: This represents a structural vulnerability for the female workforce, suggesting that AI adoption could disproportionately impact female employment patterns unless mitigated by specific policies.

5. Significance and Claims

The paper positions itself not as a definitive prediction of job losses, but as a diagnostic tool for strategic planning. Its primary significance lies in:

Bridging the Data Gap: Enabling the use of standard Spanish statistical data (INE/CNAE) to answer questions about AI impact that were previously unanswerable at the sub-national level.
Highlighting Structural Inequalities: Demonstrating that AI exposure is not random but follows existing lines of territorial specialization and gender segregation in the labor market.
Informing the "Just Transition": Providing a baseline for the Spanish government and social partners to design training, reskilling, and protection policies that are geographically and demographically targeted.

The authors conclude that while the immediate impact of AI is likely complementary (enhancing productivity), the structural exposure identified in this study highlights where the risks of displacement and the need for adaptation are greatest. They emphasize that future work should refine the model with more granular occupational data and incorporate dynamic adoption rates, but the current framework offers a necessary first step for inclusive and equitable technological governance in Spain.