%%{init: {'theme': 'base', 'themeVariables': {'fontSize': '18px', 'primaryColor': '#4a7c6f', 'primaryTextColor': '#ffffff', 'primaryBorderColor': '#334155', 'lineColor': '#334155', 'secondaryColor': '#f9fafb', 'tertiaryColor': '#f9fafb'}, 'flowchart': {'useMaxWidth': true, 'nodeSpacing': 30, 'rankSpacing': 40, 'htmlLabels': true, 'width': 1150, 'height': 650}}}%%
flowchart LR
A["Technology<br/>Innovation"] --> B["Output per<br/>Worker Rises"]
B --> C{"Condition 1:<br/>Raises Marginal<br/>Productivity?"}
C -->|"Yes: Complement"| D["Labor<br/>Demand Rises"]
C -->|"No: Substitute"| E["Displacement:<br/>Labor Demand Falls"]
D --> F{"Condition 2:<br/>Bargaining<br/>Power?"}
F -->|"Yes"| G["Wages Rise<br/>Shared Prosperity"]
F -->|"No"| H["Wages Stagnate<br/>Inequality Rises"]
Technology and Social Change
Lecture 14: Artificial Intelligence and the Political Economy of Technological Change
Bogdan G. Popescu 
Tecnológico de Monterrey
AI and the Political Economy of Technological Change
Will AI create shared prosperity—or concentrated gains?
From steam engines to large language models: mechanisms, not events
Understanding AI requires political economy, not just computer science
Same fundamental questions as the Industrial Revolution
The Engels Pause: Then and Now
Industrial Revolution (1790–1840):
- Productivity rose substantially during early industrialization
- Real wages stagnated for 50+ years
- Workers lacked bargaining power (unions illegal, labor surplus)
Contemporary parallel (post-1980):
- Productivity continued rising
- Median wages stagnated
- Union density declined sharply
Study 1: Results—Biggest Gains for Novice Workers
Figure 1: AI productivity gains are largest for lower-skill workers
Study 2: AI Users Were 40% Faster
Figure 2: AI users completed professional writing tasks significantly faster
Study 2: AI Compresses the Performance Gap
Figure 3: AI flattens the quality gap between strong and weak writers
Study 3: Robots and Jobs (Acemoglu & Restrepo, 2020)
U.S. local labor markets and industrial robots
- +1 robot per 1,000 workers leads to:
- Employment rate falls 0.20 pp
- Average wages fall 0.42%
- Effects strongest in manufacturing regions
Implication: Displacement was NOT offset by task creation locally
Source: Acemoglu & Restrepo (2020, Fig. 6)
The Race Between Education and Technology
Goldin & Katz (2008): Inequality as a “race”
- Education outpaces technology (1900–1980): inequality falls
- Technology outpaces education (post-1980): inequality rises
AI challenge: Capabilities advance faster than education can adapt
Policy implication: Retraining can help—but only if it keeps pace
AI Exposure: Who Is Affected?
Eloundou, Manning, Mishkin & Rock (2024):
- ~80% of U.S. workers have at least 10% of tasks exposed to LLMs
- ~19% have at least 50% of tasks exposed
- Unlike robots, AI hits high-skill cognitive workers hardest
This inverts the usual automation pattern
AI Exposure by Occupation Group
Figure 4: AI exposure is highest for cognitive occupations
Capability vs. Reality: The AI Adoption Gap
Figure 5: Theoretical AI capability far exceeds actual observed usage across all occupational categories
Most Exposed Occupations: Top 10
Figure 6: Computer programmers and customer service reps lead in observed AI task coverage
Early Warning: Young Worker Hiring Slowdown
Figure 7: Job finding rates for young workers in AI-exposed occupations show early decline
Study 5: The Jagged Technological Frontier
Dell’Acqua et al. (2023): 758 BCG consultants in a field experiment
- Inside AI’s frontier (creative, analytical tasks):
- 12% more tasks completed, 25% faster, 40% higher quality
- Outside AI’s frontier (tasks needing real-world judgment):
- 19 percentage points less likely to produce correct answers
Key insight: AI capability is uneven—workers must learn where it helps and where it harms
Institutional Responses: What Might Work?
| Policy | Condition Targeted |
|---|---|
| Education/retraining | Condition 1: maintain marginal productivity |
| Collective bargaining | Condition 2: restore rent sharing |
| Income support (UBI) | Cushion displacement transitions |
| Antitrust/competition | Reduce market power concentration |
| AI governance | Shape technology development direction |
No single policy suffices—combinations are needed, as during industrialization
Source: Author’s illustration based on Acemoglu & Johnson (2023)
Appendix: Macro Projections Bar Chart
Figure 8: Estimates of AI’s macroeconomic impact vary by an order of magnitude