Technology and Social Change

Lecture: 11 — Technology, Labor & Automation

Bogdan G. Popescu

bgpopescu@tec.mx

Tecnológico de Monterrey

Three Motivating Puzzles

Puzzle 1 — Productivity vs. Employment

If automation kills jobs, why has employment risen for two centuries?

Puzzle 2 — Growth vs. Inequality

Why does rising output often increase inequality?

Automated warehouse robots at Ocado. Modern logistics automation exemplifies the “automation anxiety” puzzle. Photo: Techwords / Wikimedia Commons (CC BY-SA 4.0).

Three Motivating Puzzles

Puzzle 3 — The Hollowing Middle

Why do middle-skill jobs shrink while extremes grow?

Learning Objectives

Explain the task-based framework (Autor, Acemoglu)
Distinguish automation from augmentation
Identify causal mechanisms linking tech to labor
Analyze how institutions mediate inequality
Evaluate determinism vs. contingency
Apply frameworks to AI, platforms, and policy

Conceptual Framework

Why Simple Automation Models Fail

The “Luddite fallacy” predicts outcomes we do not observe:

Fixed demand assumption — ignores price-driven demand growth
Ignores new task creation — tech creates occupations too
Treats labor as homogeneous — workers differ in tasks
Neglects institutions — same tech, different outcomes

The key error: confusing task displacement with overall employment effects (Acemoglu & Restrepo, 2019).

The Task-Based Framework

Key Insight

Jobs are bundles of tasks. Technology affects tasks, not jobs directly.

┌────────────────────────────────────────────────────────┐
│                    JOB DECOMPOSITION                   │
│                                                        │
│    "Accountant" = { data entry, analysis, client       │
│                     communication, judgment calls }    │
│                                                        │
│    Automation affects these tasks differentially:      │
│    • Data entry: HIGH substitution potential           │
│    • Analysis: PARTIAL augmentation                    │
│    • Communication: LOW substitution potential         │
│    • Judgment: VERY LOW substitution potential         │
└────────────────────────────────────────────────────────┘

Autor’s Task Taxonomy

Task Type	Examples	Automation Risk
Routine Cognitive	Data entry, bookkeeping	HIGH
Routine Manual	Assembly, machine operation	HIGH
Non-Routine Analytical	Research, programming	VARIABLE
Non-Routine Interpersonal	Teaching, negotiation	LOW
Non-Routine Manual	Janitorial, food service	LOW

Source: Autor, Levy & Murnane (2003)

The Ford Model T assembly line (1913): routine manual tasks broken into repetitive steps — the classic target of automation. Source: Wikimedia Commons (public domain).

The Acemoglu-Restrepo Framework

Technology produces dual effects on labor:

Displacement — capital replaces labor in existing tasks
Reinstatement — new tasks where labor has advantage
Productivity — output expands from efficiency gains

The net employment effect is empirically indeterminate — it depends on which force dominates.

Displacement-Reinstatement Model

%%{init: {'theme': 'base', 'themeVariables': {'fontSize': '20px', 'primaryColor': '#4a7c6f', 'primaryTextColor': '#1e293b', 'primaryBorderColor': '#334155', 'lineColor': '#64748b', 'secondaryColor': '#f9fafb', 'tertiaryColor': '#f9fafb'}, 'flowchart': {'useMaxWidth': true, 'nodeSpacing': 40, 'rankSpacing': 50, 'width': 1150, 'height': 650}}}%%
flowchart LR
    A["Technological<br/>Change"] --> B["Displacement<br/>Effect"]
    A --> C["Reinstatement<br/>Effect"]
    A --> D["Productivity<br/>Effect"]
    B --> E["Automates existing tasks<br/>(− Labor Demand)"]
    C --> F["Creates new tasks<br/>(+ Labor Demand)"]
    D --> G["Expands output<br/>(± Labor Demand)"]
    E --> H{"Net Employment<br/>Effect"}
    F --> H
    G --> H
    style A fill:#1e293b,color:#f9fafb,stroke:#334155
    style B fill:#b44527,color:#f9fafb,stroke:#334155
    style C fill:#4a7c6f,color:#f9fafb,stroke:#334155
    style D fill:#b7943a,color:#1e293b,stroke:#334155
    style H fill:#64748b,color:#f9fafb,stroke:#334155

Source: Based on Acemoglu & Restrepo (2019)

Automation vs. Augmentation

Dimension	Automation	Augmentation
Definition	Machines replace human labor	Machines complement human labor
Labor effect	Substitution (demand ↓)	Complementarity (demand ↑)
Skill bias	Reduces routine-skill demand	Raises cognitive-skill demand
Examples	ATMs, assembly robots	CAD software, diagnostic AI

Source: Based on Acemoglu (2021)

ATMs automated cash handling but also expanded bank branch networks, creating new teller roles focused on customer service. Source: Wikimedia Commons.

Automation vs. Augmentation

The same technology can serve automation or augmentation — this is a design choice, not a technological inevitability (Acemoglu, 2021).

Why Excessive Automation?

The tax system creates bias toward automation:

Labor is heavily taxed (income, payroll, benefits)
Capital faces lower effective tax rates
Firms substitute capital even when socially inefficient
Result: “so-so automation” — displaces without big gains

Exercise 1: Task Decomposition

Prompt: Pick a job you know well (barista, teacher, journalist, nurse). Decompose it into 4–5 tasks. For each task, rate automation potential as HIGH / MEDIUM / LOW. Which tasks might AI affect that traditional automation could not?

Estimated time: 5 minutes. Discuss in pairs, then share.

Causal Mechanisms

The Causal System

%%{init: {'theme': 'base', 'themeVariables': {'fontSize': '20px', 'primaryColor': '#4a7c6f', 'primaryTextColor': '#1e293b', 'primaryBorderColor': '#334155', 'lineColor': '#64748b', 'secondaryColor': '#f9fafb', 'tertiaryColor': '#f9fafb'}, 'flowchart': {'useMaxWidth': true, 'nodeSpacing': 50, 'rankSpacing': 50, 'width': 1150, 'height': 650}}}%%
flowchart LR
    A["Institutions<br/>(taxes, unions,<br/>education)"] -->|shape| B["Technological<br/>Change"]
    B --> C["Task<br/>Reconfiguration"]
    C --> D["Labor Market<br/>Adjustment"]
    D --> E["Distributional<br/>Outcomes"]
    E --> F["Political<br/>Responses<br/>(policy, parties)"]
    F -->|reshape| A

    style A fill:#4a7c6f,color:#f9fafb,stroke:#334155
    style B fill:#1e293b,color:#f9fafb,stroke:#334155
    style F fill:#b44527,color:#f9fafb,stroke:#334155

Causation is recursive: institutions shape technology and are reshaped by its consequences.

Mechanism 1: Skill-Biased Technological Change

Definition

SBTC: Technologies that raise the relative productivity and wages of skilled workers compared to unskilled workers.

Causal pathway: Computer tech → ↑ cognitive-skill demand → ↑ college premium → ↑ inequality

U.S. College Wage Premium, 1963–2022

Figure 1: College vs. high-school earnings gap

Mechanism 2: Routine-Biased Technological Change

SBTC cannot explain job polarization. Autor (2003, 2013) proposes RBTC:

Computers excel at rule-based tasks
Routine tasks (cognitive AND manual) are automatable
Non-routine tasks at BOTH ends are complemented
Middle-skill, routine jobs decline
High-skill AND low-skill service jobs grow

U.S. Job Polarization, 1980–2005

Figure 2: Employment share change by skill percentile

Mechanism 3: Geographic Concentration

Automation effects are spatially concentrated:

Manufacturing automation devastated specific regions
Knowledge economy concentrates in “superstar cities”
Local labor markets adjust slowly
Geographic immobility compounds individual displacement

The abandoned Packard Automotive Plant, Detroit. Deindustrialization devastated specific regions while others thrived. Source: Wikimedia Commons (public domain).

Mechanism 3: Geographic Concentration

Automation effects are spatially concentrated:

The China shock destroyed ~2.4 million U.S. jobs, concentrated in specific commuting zones (Autor, Dorn & Hanson, 2013, 2016).

Mechanism 4: Productivity-Wage Decoupling

Figure 3: Productivity grew far faster than compensation

Institutions & Path Dependence

The Institutional Mediation Thesis

Key Claim

The same technology produces different outcomes across institutional contexts. Effects on employment, wages, and inequality are institutionally contingent.

Cross-National Evidence

Figure 4: High robot density does not guarantee high inequality

Same Technology, Different Outcomes

%%{init: {'theme': 'base', 'themeVariables': {'fontSize': '18px', 'primaryColor': '#4a7c6f', 'primaryTextColor': '#1e293b', 'primaryBorderColor': '#334155', 'lineColor': '#64748b', 'secondaryColor': '#f9fafb', 'tertiaryColor': '#f9fafb'}, 'flowchart': {'useMaxWidth': true, 'nodeSpacing': 30, 'rankSpacing': 45, 'width': 1150, 'height': 650}}}%%
flowchart TD
    A["Industrial Robots<br/>(Same Technology)"] --> B["Germany"]
    A --> C["United States"]
    A --> D["Sweden"]
    B --> E["Strong unions<br/>Vocational ed<br/>Codetermination"]
    C --> F["Weak unions<br/>Limited training<br/>Market-led"]
    D --> G["Strong unions<br/>Active labor<br/>market policy"]
    E --> H["Moderate inequality<br/>Job security"]
    F --> I["Sharp inequality<br/>Job loss"]
    G --> J["Low inequality<br/>Smooth transitions"]
    style A fill:#1e293b,color:#f9fafb,stroke:#334155
    style H fill:#b7943a,color:#1e293b,stroke:#334155
    style I fill:#b44527,color:#f9fafb,stroke:#334155
    style J fill:#4a7c6f,color:#f9fafb,stroke:#334155

Source: Conceptual diagram based on Hall & Soskice (2001); Thelen (2014)

Path Dependence and Lock-In

Brian Arthur’s Path Dependence

Early institutional choices become self-reinforcing, making later changes difficult even when alternatives are superior.

U.S. path: Limited union institutionalization (1930s–50s) → weak safety net → few policy tools today

Germany path: Codetermination (1940s–60s) → strong works councils → technology adoption shaped by worker input

Workers on the VW Crafter assembly line, Wrzesnia. German codetermination gives workers voice in how automation is adopted. Photo: Karlis Dambrans (CC BY 2.0).

Source: Arthur (1989); Pierson (2004)

Mokyr: Technology and Institutional Adaptation

Mokyr (1990, 2002) emphasizes:

Technology alone does not determine outcomes
“Useful knowledge” requires institutional support
Industrial Revolution needed complementary institutions
Patent systems, education, and labor regulation co-evolved
Implication: today’s transition requires institutional innovation

Exercise 2: Institutional Design

Prompt: You are advising Mexico’s government on automation policy. Mexico has moderate union density and a large informal sector. Drawing on the Germany/Sweden/U.S. comparison, which 2–3 institutional features would you prioritize and why?

Estimated time: 5 minutes. Small-group discussion.

Empirical Evidence

Acemoglu & Restrepo (2020): Robots and Jobs

Research design: Variation in robot adoption across U.S. commuting zones; instrumented with European industry robot adoption.

Source: Acemoglu & Restrepo (2020, *JPE*)
Outcome	Effect per Robot / 1,000 Workers
Employment-to-pop. ratio	−0.2 pp
Wages	−0.42%
Manufacturing jobs	Disproportionate decline

Industrial robots in automotive manufacturing. Source: Wikimedia Commons.

The China Shock

Autor, Dorn & Hanson (2013, 2016, 2021):

Import competition caused ~2.4 million U.S. job losses
Effects concentrated in specific local labor markets
Workers did not relocate; many left labor force
Effects persisted for over a decade

Container ship NYK Themis at the Port of Los Angeles. The scale of global trade created concentrated regional shocks. Photo: Downtowngal / Wikimedia Commons (CC BY-SA 4.0).

The China Shock

Autor, Dorn & Hanson (2013, 2016, 2021):

Political consequence: Trade-exposed regions shifted toward populist candidates (Autor et al., 2020).

Frey & Osborne (2017): Automation Risk

Headline finding:

“47% of U.S. employment at high risk of automation.”

Critiques:

Estimates at occupation-level, not task-level
Ignores economic and institutional barriers
OECD re-analysis (task-based): only 9% at high risk
Arntz, Gregory & Zierahn (2016): similar revision

Competing Interpretations

Interpretation 1: Technological Determinism

Technology has autonomous logic driving social outcomes independent of human choice.

Exponential progress creates inexorable pressure
AI represents a qualitative break with past automation
Cognitive tasks now automatable (“this time is different”)
Resistance is futile; adaptation is the only option

Proponents: Brynjolfsson & McAfee (qualified); Kurzweil (strong form)

Interpretation 2: Institutional Contingency

Technology creates possibilities; institutions determine outcomes.

Same technologies → different outcomes across societies
Technology adoption is itself an institutional choice
Policy can redirect technological development
Historical transitions were managed through institutional innovation

Proponents: Acemoglu, Autor, Mokyr, Piketty, Hall & Soskice

The Debate Matrix

Dimension	Determinism	Contingency
Causation	Tech → Outcomes	Tech ↔︎ Institutions
Agency	Limited	Substantial
Policy	Adapt to the inevitable	Shape the development path
AI	Qualitative break	Continuous with history
Prediction	Massive displacement	Depends on choices

Source: Author’s synthesis

An Integrated View

Synthesis

Technology creates pressures and possibilities, but institutions determine the rate, direction, distribution, and adjustment mechanisms.

The question is not “Will AI take our jobs?” but:

“What institutions will we build so progress benefits broadly?”

Modern Relevance

AI and Large Language Models

How does AI fit the task-based framework?

AI automates cognitive, non-routine tasks (new frontier)
RBTC extension: writing, analysis, coding now affected
Strong complementarity potential for knowledge workers
Speed of change faster than previous transitions
Novel capabilities make prediction difficult

Platform Labor and the Gig Economy

Platform Dynamics

Digital platforms restructure employment relationships in ways that often shift risk to workers.

Independent contractor status avoids labor protections
Algorithmic management replaces human supervisors
Task-level labor markets atomize work
Surveillance and control without employment rights

Examples: Uber, DoorDash, Mechanical Turk, Upwork

Policy Responses: A Typology

Approach	Examples	Logic
Laissez-faire	Minimal intervention	Markets adjust
Social insurance	Retraining, UI benefits	Cushion transitions
UBI	Universal basic income	Decouple income from work
Taxation	Robot taxes, wealth taxes	Slow automation; redistribute
Tech policy	Direct R&D toward augmentation	Change the technology path
Labor institutions	Strengthen unions, codetermination	Rebalance power

Source: Synthesized from Acemoglu (2021); Korinek & Stiglitz (2021)

The Political Economy of Automation

Automation exposure predicts populist voting (Autor et al., 2020)
Robot exposure correlates with Trump support (Frey et al., 2018)
Rising inequality fuels political instability (Piketty, 2020)
Mechanism: displacement → status anxiety → anti-establishment politics

Rodrik (2018) argues that populism is, in part, a political response to the uneven distributional consequences of globalization and technological change.

Exercise 3: AI Policy Debate

Prompt: Should governments impose a tax on AI systems that replace human workers? Split into two groups: one argues FOR (citing Acemoglu’s “so-so automation”), the other AGAINST (citing innovation effects). Each group prepares 3 arguments.

Estimated time: 5 minutes. Structured debate format.

Synthesis & Conclusion

Key Takeaways

Jobs = bundles of tasks — tech affects tasks differentially
Dual effects: displacement (−) and reinstatement (+) are both real
RBTC explains polarization — routine tasks automated regardless of skill
Institutions mediate outcomes — same tech, different inequality
Adjustment is slow — markets do not rapidly equilibrate
AI extends the frontier — cognitive tasks now affected; policy is choice

Discussion Questions

Which mechanism (SBTC, RBTC, geography, institutions) best explains contemporary inequality? Why?
Why might Germany experience lower inequality from automation than the U.S.?
Is AI qualitatively different from previous automation, or a continuation?
How should policymakers evaluate the trade-off between a robot tax and innovation?
If automation concentrates gains among owners, what principles should guide redistribution?

References

Acemoglu, D. (2021). Harms of AI. NBER WP 29247.
Acemoglu, D., & Restrepo, P. (2019). Automation and new tasks. JEP, 33(2), 3–30.
Acemoglu, D., & Restrepo, P. (2020). Robots and jobs. JPE, 128(6), 2188–2244.
Arntz, M., Gregory, T., & Zierahn, U. (2016). Risk of automation for OECD jobs. OECD WP 189.
Arthur, W. B. (1989). Competing technologies and lock-in. EJ, 99(394), 116–131.
Autor, D. H., & Dorn, D. (2013). Low-skill service jobs and polarization. AER, 103(5), 1553–1597.
Autor, D. H., Dorn, D., & Hanson, G. H. (2013). The China syndrome. AER, 103(6), 2121–2168.
Autor, D. H., Dorn, D., & Hanson, G. H. (2016). The China shock. Ann. Rev. Econ., 8, 205–240.
Autor, D. H., Dorn, D., Hanson, G. H., & Majlesi, K. (2020). Importing political polarization. AER, 110(10), 3139–3183.
Autor, D. H., Levy, F., & Murnane, R. J. (2003). Skill content of tech change. QJE, 118(4), 1279–1333.
Brynjolfsson, E., & McAfee, A. (2014). The second machine age. W. W. Norton.
Frey, C. B., Berger, T., & Chen, C. (2018). Political machinery. Oxford Rev. Econ. Policy, 34(3), 418–442.
Frey, C. B., & Osborne, M. A. (2017). Future of employment. Tech. Forecasting & Soc. Change, 114, 254–280.

References

Goldin, C., & Katz, L. F. (2008). The race between education and technology. Harvard UP.
Hall, P. A., & Soskice, D. (2001). Varieties of capitalism. Oxford UP.
Korinek, A., & Stiglitz, J. E. (2021). AI and development strategies. NBER WP 28453.
Mokyr, J. (1990). The lever of riches. Oxford UP.
Mokyr, J. (2002). The gifts of Athena. Princeton UP.
Pierson, P. (2004). Politics in time. Princeton UP.
Piketty, T. (2020). Capital and ideology. Harvard UP.
Rodrik, D. (2018). Populism and globalization. J. Int. Bus. Policy, 1(1–2), 12–33.
Thelen, K. (2014). Varieties of liberalization. Cambridge UP.