Technology and Social Change

Lecture 15: AI and International Relations

Bogdan G. Popescu

bgpopescu@tec.mx

Tecnológico de Monterrey

Motivating Puzzles & Objectives

Motivating Puzzles

Puzzle 1: Why does the same technology (AI) produce different governance outcomes in democracies vs. autocracies?

Puzzle 2: Why do US-China chip bans disrupt global supply chains?

Puzzle 3: Why might AI infrastructure matter more than AI algorithms?

Puzzle 4: Why have international efforts to regulate autonomous weapons failed?

US Secretary of State Blinken meets PRC liaison minister Liu Jianchao, Washington, 2024. AI competition is now central to great-power relations. Source: US State Department / Wikimedia Commons (public domain).

Learning Objectives

By the end of this lecture, students should be able to:

1. Apply IR theories (Realism, Liberalism, Constructivism) to AI

2. Explain the geopolitical logic of US-China AI competition

3. Assess AI’s military implications through autonomous weapons and cyber warfare

4. Understand why AI infrastructure is power (chips, energy, data centers)

5. Compare governance models (democratic vs. authoritarian AI)

AI and IR Theory

AI and IR Theory: Three Paradigms

Note

“No single paradigm can comprehensively capture the entirety of AI’s implications for war, trade, and international order.” — Ndzendze & Marwala (2023)

Realism: States seek power in an anarchic system — AI as strategic asset

Liberalism: States cooperate through trade and institutions — AI enables shared governance

Constructivism: Ideas and norms shape world politics — AI reshapes threat perceptions, identities, and the meaning of “AI superpower”

Why These Theories Matter for AI

Each theory predicts different outcomes for AI governance:

Theory	Key Prediction	Policy Implication
Realism	Arms race, decoupling	Export controls, tech nationalism
Liberalism	Cooperation possible	International standards, treaties
Constructivism	Norms shape behavior	“AI superpower” identity politics

The US-China rivalry provides a natural experiment: which theory best explains observed behavior?

The US-China AI Competition

US-China AI Competition

The Evidence: United States

This rivalry closely matches the Realist view: states compete for power under anarchy.

United States

• $500B Stargate Project (2025)
• Export controls on advanced chips
• CHIPS and Science Act ($52.7B)

Stargate data centers, Abilene, Texas

US-China AI Competition

The Evidence: China

China

• China publishes far more AI-related patent documents than the US (2019–2025)
• DeepSeek R1 triggered Nasdaq 3.1% drop
• 295,000 industrial robots installed (2024)

Source: GreyB Patent Database (2019–2025)

AI patent filings, 2019-2025. Source: GreyB

LLM Performance: US vs. China

Source: LMArena Text Arena Elo Ratings (Jul 2024 – Jul 2025)

Security Dilemma in AI Competition

%%{init: {'theme': 'base', 'themeVariables': {'fontSize': '18px', 'primaryColor': '#4a7c6f', 'primaryTextColor': '#1e293b', 'lineColor': '#334155', 'secondaryColor': '#f9fafb'}, 'flowchart': {'useMaxWidth': true, 'nodeSpacing': 40, 'rankSpacing': 45, 'curve': 'basis'}, 'width': 1150, 'height': 650}}%%
flowchart LR
    A["US invests<br/>in AI"] --> B["China perceives<br/>threat"]
    B --> C["China increases<br/>AI investment"]
    C --> D["US perceives<br/>China catching up"]
    D --> E["US increases<br/>controls<br/>(chip bans)"]
    E --> F["China accelerates<br/>domestic<br/>development"]
    F --> B

    style A fill:#4a7c6f,color:#f9fafb,stroke:#334155
    style B fill:#b44527,color:#f9fafb,stroke:#334155
    style C fill:#b44527,color:#f9fafb,stroke:#334155
    style D fill:#4a7c6f,color:#f9fafb,stroke:#334155
    style E fill:#4a7c6f,color:#f9fafb,stroke:#334155
    style F fill:#b44527,color:#f9fafb,stroke:#334155

Key insight: Anarchy creates competition regardless of intentions (Realist mechanism)

The “Digital Cold War”

Note

“As of mid-2025, the geopolitics of AI stands at a crossroads… the world could slide further into fragmentation, with a digital iron curtain separating US-led and China-led tech spheres.” — World Economic Forum (2025)

1. Bloc Formation — Chip 4 Alliance (US, Japan, Taiwan, South Korea) vs. Digital Silk Road

2. Fragile Rules — Nov 2024: US-China pledge that humans—not AI—should control nuclear weapons

3. Strategic Decoupling — April 2025: US banned Nvidia H20 chip exports to China

Exercise 1: Applying IR Theories

Class Discussion (5 min)

Pick one of the following events and explain it using all three IR theories:

1. The US banning Nvidia chip exports to China (April 2025)
2. China’s DeepSeek R1 launch triggering a Nasdaq drop
3. France announcing a $109B “third way” AI plan

Format: State the theory, identify the causal mechanism, and predict what happens next.

AI and Military Applications

From Competition to the Battlefield

If AI is a core arena of great-power rivalry, its most dangerous application is military force.

• AI is increasingly used in targeting and weapons systems
• This raises legal and ethical questions about killing by machines

One of the central debates is over Lethal Autonomous Weapons Systems (LAWS).

An MQ-9 Reaper unmanned aerial vehicle — one of the most widely deployed military drones. Source: Wikimedia Commons (public domain, US Air Force).

Lethal Autonomous Weapons Systems (LAWS)

Source: https://www.youtube.com/watch?v=X7MqE-vqnSw

1. How does the video define autonomous weapon systems?
2. In which regions are autonomous weapons already being used?
3. What does “automation bias” mean, and why is it dangerous?
4. What international response is discussed in the video?

LAWS: Definitions and Regulation

Note

CCW Definition (Nov 2024): “An integrated combination of weapons and technological components that enable the system to identify and/or select, and engage a target, without intervention by a human user.”

• 166 UN votes in favor of regulation (Dec 2024)
• 3 opposed (Russia, DPRK, Belarus)
• 2026 target for binding treaty

Key issues: Human control, accountability gaps, IHL compliance

The UN General Assembly Hall, where 166 nations voted in favor of autonomous weapons regulation. Source: Wikimedia Commons (public domain).

Why LAWS Regulation Fails

%%{init: {'theme': 'base', 'themeVariables': {'fontSize': '18px', 'primaryColor': '#b44527', 'primaryTextColor': '#1e293b', 'lineColor': '#334155', 'secondaryColor': '#f9fafb'}, 'flowchart': {'useMaxWidth': true, 'nodeSpacing': 40, 'rankSpacing': 50, 'curve': 'basis'}, 'width': 1150, 'height': 650}}%%
flowchart LR
    A["All states want<br/>to avoid an<br/>arms race"] --> B["Each state fears<br/>disadvantage if<br/>others develop LAWS"]
    B --> C["Verification is<br/>nearly impossible<br/>(software, not hardware)"]
    C --> D["Defection is the<br/>dominant strategy"]
    D --> E["Result: 166 UN votes<br/>in favor, but<br/>no binding treaty"]

    style A fill:#4a7c6f,color:#f9fafb,stroke:#334155
    style B fill:#b7943a,color:#1e293b,stroke:#334155
    style C fill:#b7943a,color:#1e293b,stroke:#334155
    style D fill:#b44527,color:#f9fafb,stroke:#334155
    style E fill:#b44527,color:#f9fafb,stroke:#334155

Key insight: Unlike nuclear weapons (visible, countable), AI is embedded in software — verification is structurally difficult.

From Autonomous Weapons to Cyber Warfare

LAWS represents one dimension of AI’s military impact. AI is also transforming cyberspace.

• AI doesn’t just power weapons — it enables new forms of attack
• Cyber operations are now a primary tool of great-power competition

The ODNI’s 2025 Annual Threat Assessment identifies China as “the most active and persistent cyber threat to U.S. government, private-sector, and critical infrastructure networks.”

AI and Cybersecurity: The Dual-Use Problem

AI creates a fundamental tension: the same capabilities empower both defenders and attackers.

For Defenders:

• Predictive threat intelligence
• Automated incident response

Cyber security at the UK Ministry of Defence. AI empowers both attackers and defenders in the escalating cyber arms race. Source: Chris Roberts / UK MOD / Wikimedia Commons (OGL v1.0).

AI and Cybersecurity: The Dual-Use Problem

AI creates a fundamental tension: the same capabilities empower both defenders and attackers.

For Attackers:

• Adaptive malware that evades detection
• Real-time vulnerability exploitation

Cyber security at the UK Ministry of Defence. AI empowers both attackers and defenders in the escalating cyber arms race. Source: Chris Roberts / UK MOD / Wikimedia Commons (OGL v1.0).

Cybercrime Costs Are Exploding

Source: Cybersecurity Ventures

Defense Is Adapting: Dwell Time Declining

Source: Mandiant M-Trends Reports

Exercise 2: Autonomous Weapons Debate

Small Group Discussion (5 min)

1. Should machines ever make life-or-death decisions in war?
2. Does autonomous weapons development make war more or less likely?
3. Can international law realistically regulate LAWS?

Use these concepts: Security dilemma, collective action failure, verification problems

AI Infrastructure

Semiconductor Wars: The Critical Chokepoint

Export Controls as Weapon

• Jan 15, 2025: BIS issued a proposed “AI Diffusion” export-control framework (global licensing tiers for advanced AI chips)
• May 13, 2025: BIS rescinded the AI Diffusion rule before implementation (export controls still exist, but not under that specific framework)

Semiconductor Manufacturing Shift

Source: Semiconductor Industry Association (SIA)

How Chip Chokepoints Create Leverage

%%{init: {'theme': 'base', 'themeVariables': {'fontSize': '17px', 'primaryColor': '#4a7c6f', 'primaryTextColor': '#1e293b', 'lineColor': '#334155', 'secondaryColor': '#f9fafb'}, 'flowchart': {'useMaxWidth': true, 'nodeSpacing': 30, 'rankSpacing': 40, 'curve': 'basis'}, 'width': 1150, 'height': 650}}%%
flowchart LR
    A["Chip Design<br/>(US dominance)<br/>Nvidia, AMD"] --> D["CHOKEPOINT<br/>Any node can<br/>be weaponized"]
    B["Chip Fabrication<br/>(Taiwan dominance)<br/>TSMC: 92% of advanced nodes"] --> D
    C["Chip Equipment<br/>(Netherlands)<br/>ASML: 100% EUV"] --> D
    D --> E["US export<br/>controls"]
    D --> F["Taiwan<br/>invasion risk"]
    D --> G["ASML restricted<br/>from China sales"]

    style A fill:#4a7c6f,color:#f9fafb,stroke:#334155
    style B fill:#b7943a,color:#1e293b,stroke:#334155
    style C fill:#b44527,color:#f9fafb,stroke:#334155
    style D fill:#1e293b,color:#f9fafb,stroke:#334155
    style E fill:#64748b,color:#f9fafb,stroke:#334155
    style F fill:#64748b,color:#f9fafb,stroke:#334155
    style G fill:#64748b,color:#f9fafb,stroke:#334155

Key insight: Unlike oil, chips cannot be stockpiled — technology ages. Control is ongoing.

Energy: The New Constraint on AI Power

The “Electron Gap”

• US data center demand will double by 2030 (426 TWh)
• That is 9% of total US electricity demand
• China builds infrastructure faster without public opposition

Data Center Energy vs. Countries

Source: Enerdata, IEA (2020 data)

Why Location Matters

Data Center Geography

• US hosts 45%+ of world’s data centers
• Northern Virginia = largest market globally
• Countries compete for data centers like factories

Location determines: latency, energy access, regulatory control

US data center demand capacity by county

Big Tech Concentration in AI

The AI Stack Is Bottlenecked

Concentration is not just “few apps.” It is control over inputs needed for frontier AI:

• Compute: Data centers and cloud access

• Chips: Specialized accelerators and supply chains

• Data + Distribution: Platform integration, default channels

• Talent + Research: Labs, benchmarks, conference incentives

A semiconductor clean room at NASA Glenn Research Center. Advanced chip fabrication requires extreme precision and massive capital investment — creating natural chokepoints in the AI supply chain. Source: NASA / Wikimedia Commons (public domain).

How Big Tech Gains Power

1) Gatekeeping — Control of infrastructure

Big tech firms control cloud servers, AI tools, and payments — deciding who can build AI.

2) Lock-in — Hard to leave

Once firms use one platform, data and workflows become tied to it.

3) Private rule-making — Setting the standards

Big tech shapes safety standards and benchmarks, influencing governments and markets.

From Concentration to Political Influence

%%{init: {'theme': 'base', 'themeVariables': {'fontSize': '18px', 'primaryColor': '#4a7c6f', 'primaryTextColor': '#1e293b', 'lineColor': '#334155', 'secondaryColor': '#f9fafb'}, 'flowchart': {'useMaxWidth': true, 'nodeSpacing': 40, 'rankSpacing': 50, 'curve': 'basis'}, 'width': 1150, 'height': 650}}%%
flowchart LR
    A["Economies<br/>of Scale<br/>(training costs<br/>$100M+)"] --> B["Market<br/>Dominance<br/>(3-5 firms<br/>control stack)"]
    B --> C["Resource<br/>Asymmetry<br/>(lobbying,<br/>revolving door)"]
    C --> D["Political<br/>Influence<br/>(regulatory<br/>capture)"]

    style A fill:#4a7c6f,color:#f9fafb,stroke:#334155
    style B fill:#b7943a,color:#1e293b,stroke:#334155
    style C fill:#b44527,color:#f9fafb,stroke:#334155
    style D fill:#1e293b,color:#f9fafb,stroke:#334155

Potential checks: Antitrust (slow), open-source (resource-constrained), international competition, democratic mobilization

What Can Governments Do?

AI tends toward dominance because it needs huge investment, data, chips, and energy.

Four government strategies:

1. Antitrust laws — block mergers and monopolies
2. Interoperability rules — reduce platform lock-in
3. Public AI systems — reduce dependence on private firms
4. Safety and audit rules — force public standards

Each strategy faces political and practical constraints.

Democracy vs. Authoritarianism

Two Visions of AI Governance

Democratic Model (US/EU)

• Innovation with guardrails
• Market-driven development
• Concentration of power in Big Tech
• Growth vs. worker protections tension

Authoritarian Model (China)

• State licensing: AI must align with state
• Strategic infrastructure approach
• Rapid deployment, no public opposition

Critical questions: Who controls training data? Who sets safety standards? Who benefits?

Comparing AI Governance Models

Dimension	Democratic (US/EU)	Authoritarian (China)
Speed	Slower (deliberation)	Faster (top-down)
Distribution	Market-determined	State-directed
Accountability	Courts, elections	Party control
Innovation	Decentralized	Coordinated
Rights protection	Higher (in theory)	Lower

Neither model automatically produces better outcomes — institutions and political choices matter

Exercise 3: Governance Trade-offs

Think-Pair-Share (5 min)

Imagine you are advising a mid-income democracy (e.g., Mexico, Brazil, Indonesia) on AI policy.

1. Would you adopt the US model, China model, or a hybrid?
2. What are the trade-offs in speed, rights, and innovation?
3. What role should the state play vs. private firms?

Use concepts: Institutional lag, distributional conflict, path dependence

Historical Perspective

Lessons from History for AI Governance

What eventually shared Industrial Revolution gains?

• Union legalization (1824)
• Factory Acts (1833+)
• Suffrage expansion (1832, 1867, 1884)
• Public education (1870)

Implication: Technology alone does not ensure shared prosperity. Institutional adaptation determines who benefits.

The Great Chartist Meeting on Kennington Common (1848) — mass mobilization for democratic reform during industrialization. Source: Wikimedia Commons (public domain).

Lessons from History for AI Governance

What eventually shared Industrial Revolution gains?

• Union legalization (1824)
• Factory Acts (1833+)
• Suffrage expansion (1832, 1867, 1884)
• Public education (1870)

Goldin & Katz (2008): Inequality depends on the “race between education and technology.”

The Great Chartist Meeting on Kennington Common (1848) — mass mobilization for democratic reform during industrialization. Source: Wikimedia Commons (public domain).

Conclusion

Key Takeaways

• AI is a core dimension of great-power competition, reshaping security, influence, and status

• Military uses intensify ethical and legal dilemmas, from LAWS to cyber operations

• Physical infrastructure is geopolitical power, with chips, data centers, and electricity as chokepoints

• Competing governance models reflect deeper regime differences

• History suggests institutional adaptation will determine who benefits from AI

Analytical Framework

Use these concepts to structure your analysis:

Concept	Application
Security dilemma	Why states develop weapons even when all prefer not to
Collective action	Why treaties fail despite shared interests
Chokepoints	Where leverage exists in supply chains
Path dependence	Why early advantages compound over time
Institutional lag	Why political responses trail technology
Distributional conflict	Who gains, who loses, and how this shapes politics

Good answers trace causal mechanisms, not just describe outcomes

References

• Acemoglu, D., & Johnson, S. (2023). Power and Progress. PublicAffairs.
• Brynjolfsson, E., & McAfee, A. (2014). The Second Machine Age. W. W. Norton.
• Frey, C. B. (2019). The Technology Trap. Princeton UP.
• Goldin, C., & Katz, L. F. (2008). The Race between Education and Technology. Harvard UP.
• GreyB. (2025). AI patent landscape report.
• Mandiant / Google Cloud. (2024). M-Trends 2024: Special report.
• Masanet, E., et al. (2020). Recalibrating global data center energy-use estimates. Science, 367(6481), 984–986.
• Ndzendze, B., & Marwala, T. (2023). Artificial Intelligence and International Relations Theories. Palgrave Macmillan.
• ODNI. (2025). Annual Threat Assessment of the U.S. Intelligence Community.
• SIA/BCG. (2024). Emerging resilience in the semiconductor supply chain.
• Stanford HAI. (2025). AI Index Report 2025.
• Statista Market Insights. (2025). Estimated cost of cybercrime worldwide.
• World Economic Forum. (2025). AI geopolitics and data in the era of technological rivalry.