Four Battlegrounds — Scharre (2023)
BLUF
Four Battlegrounds: Power in the Age of Artificial Intelligence (2023) extends Scharre’s tactical analysis in Army of None to the level of great-power strategic competition. The work’s central argument — that the AI race between the United States and the People’s Republic of China is primarily an institutional contest, not a technology race — is the most analytically significant contribution to English-language AI-competition literature since the 2017 PRC AI development plan. The four-battleground framework (data, compute, talent, institutions) provides the analytical vocabulary for assessing AI competition dynamics across the vault’s active tracking of US-China strategic rivalry.
Bibliographic Information
| Field | Detail |
|---|---|
| Author | Paul Scharre |
| Full Title | Four Battlegrounds: Power in the Age of Artificial Intelligence |
| Publisher | W.W. Norton & Company, New York |
| Year | 2023 |
| Director | Paul Scharre serves as Executive Vice President and Director of Studies, Center for a New American Security (CNAS) |
| Companion Work | Army of None - Scharre (2018) |
Core Arguments
1. The Four Battlegrounds Framework
Scharre argues that AI power — the capacity of a state to translate AI capabilities into strategic outcomes — is determined across four distinct competitive domains. Assessment: The framework’s analytical value is its disaggregation: each battleground has different competitive dynamics, different state and non-state actors in advantageous positions, and different policy levers available to competing powers. No state dominates all four; the US-China competition is a complex interaction of asymmetric advantages and vulnerabilities across each domain.
Battleground 1 — Data: AI systems require training data. Structural advantages accrue to actors controlling large, high-quality, diverse datasets. Fact: The PRC holds quantitative advantages in certain data domains: a massive population, limited privacy regulation permitting state-accessible data collection at scale, and direct state access to commercial platform data under the 2017 National Intelligence Law. Assessment: US advantages are qualitative and architectural: global internet infrastructure, the diversity of English-language training corpora (which provides representational breadth not easily replicated by Chinese-language corpora), and the disproportionate share of global scientific and technical literature published in English. Gap: Scharre does not resolve whether quantitative or qualitative data advantages prove more decisive for frontier AI development; this remains empirically open.
Battleground 2 — Compute: Training frontier AI models requires semiconductor fabrication capability, GPU cluster infrastructure, and the energy systems to power them. Fact: The US-led semiconductor export control regime — principally Bureau of Industry and Security (BIS) Entity List restrictions on Nvidia H100/A100 exports and ASML EUV lithography systems — is the primary active front in compute competition as of 2023. Assessment: The compute battleground is the most directly contestable through state policy action, and the Biden administration’s October 2022 and October 2023 chip export controls represent the most consequential technology policy actions in the US-China competition since Huawei’s 2019 Entity List designation.
Battleground 3 — Talent: AI research and engineering talent is globally scarce, internationally mobile, and disproportionately trained at US research universities. Assessment: This creates a structural paradox for US competition strategy: US universities train a substantial share of the world’s top AI researchers, a significant portion of whom are PRC nationals. Those who return to China constitute a technology transfer mechanism that no export control regime can fully address. Those who remain in the US constitute the primary US competitive asset in frontier AI development. Immigration policy is therefore a core AI competition policy variable.
Battleground 4 — Institutions: The ability to deploy AI systems at operational scale requires institutional trust — from users, operators, downstream communities, and regulatory authorities. Assessment: This is Scharre’s most original and analytically significant contribution. Building institutional frameworks that allow AI capabilities to translate into reliable operational deployment takes time, requires legal infrastructure, and depends on organizational cultures and governance traditions that cannot be acquired rapidly. A state that possesses AI capabilities without the institutional capacity to deploy them reliably will not gain strategic advantage from those capabilities.
2. The Institutional Contest Thesis
Assessment: Scharre’s central argument is that the conventional framing of the US-China AI competition — as a technology race in which the winner is the state with the most capable AI systems — is analytically misleading. The decisive variable is not capability acquisition but deployment capacity: the organizational, legal, and cultural infrastructure required to translate AI capabilities into reliable operational use at scale. Assessment: This argument converges precisely with Michael C. Horowitz’s “adoption capacity” theory of military innovation diffusion, which holds that the state first to develop a technology rarely derives permanent advantage — the decisive advantage accrues to the state first to develop the organizational framework for reliable employment. The US military’s historically superior ability to integrate new technologies into joint operational doctrine is the relevant institutional asset; the question Scharre raises is whether that capacity is being brought to bear on AI integration at adequate speed.
3. Complementarity with Army of None
Assessment: Four Battlegrounds and Army of None together provide a vertically integrated analytical framework. Army of None operates at the platform and weapons-system level: how does a specific autonomous system work, what are its failure modes, who is accountable for its decisions? Four Battlegrounds operates at the great-power competition level: what determines which state derives strategic advantage from AI, and across what dimensions is that competition occurring? Neither work is complete without the other for analysts tracking both AI governance and US-China strategic rivalry.
Structure
| Part | Focus | Analytical Contribution |
|---|---|---|
| I — Data | Training data competition | PRC vs. US data position; quality vs. quantity tradeoffs |
| II — Compute | Semiconductor and infrastructure competition | Export control strategy; ASML/TSMC choke points |
| III — Talent | Human capital competition | Immigration policy as AI policy; reverse brain drain dynamics |
| IV — Institutions | Organizational and governance competition | Deployment capacity; the institutional contest thesis |
Methodological Significance
Scharre draws on interviews with senior US government officials, PRC AI researchers (primarily through secondary accounts), and defense industry practitioners. Fact: The book’s publication coincided with the US October 2022 chip export controls, making several of its compute battleground projections immediately relevant to active policy. The institutional contest thesis represents a methodological shift from capability-counting — the dominant approach in much AI competition analysis — toward an organizational sociology of AI power.
Critical Assessments
Assessment: The framework’s four battlegrounds are analytically useful but not exhaustive. Energy infrastructure (compute depends on power; data center construction is energy-constrained) and international standards-setting (the competition for AI governance norms in bodies like ISO, ITU, and the UN) receive less sustained attention than the four named domains, despite being directly relevant to AI strategic competition. The institutional contest thesis is persuasive but the book does not specify measurable indicators of institutional AI deployment capacity, which limits its operationalizability for structured intelligence assessment. Gap: The work does not treat the AI competition’s intersection with military autonomy governance — the subject of Army of None — in sufficient depth, leaving the connection between strategic competition and weapons-system-level analysis largely implicit.
Contemporary Relevance for This Vault
Four Battlegrounds provides the strategic-competition framework for assessments involving People’s Republic of China and AI competition dynamics tracked under Algorithmic Warfare. The compute battleground analysis is directly relevant to understanding the strategic logic of US semiconductor export controls and PRC responses (domestic chip acceleration, SMIC capacity expansion). The institutional contest thesis provides the analytical basis for assessments of whether PRC AI acquisitions in specific domains will translate into operational military advantage — a question the vault’s investigations of IDF and US TITAN programs approach from a different angle.
Key Connections
- Paul Scharre
- Army of None - Scharre (2018)
- Michael C. Horowitz
- Great Power Competition
- People’s Republic of China
- Russian Federation
- Algorithmic Warfare
- TITAN
- US-China Strategic Competition
- Foundational Books
Sources
- Scharre, Paul. Four Battlegrounds: Power in the Age of Artificial Intelligence. New York: W.W. Norton & Company, 2023.
- Scharre, Paul. Army of None: Autonomous Weapons and the Future of War. New York: W.W. Norton & Company, 2018. — Companion work; tactical level of the same analytical project.
- Horowitz, Michael C. The Diffusion of Military Power: Causes and Consequences for International Politics. Princeton: Princeton University Press, 2010. — Adoption-capacity theory; converges with Scharre’s institutional contest thesis.
- US Department of Commerce, Bureau of Industry and Security. Export Administration Regulations: Commerce Control List Supplement No. 4 to Part 774 — Advanced Computing Items. October 7, 2022. [primary, regulatory] — The policy action the book’s compute battleground analysis contextualizes.
- Roberts, Hal, et al. Measuring the AI Compute Ecosystem. Georgetown Center for Security and Emerging Technology (CSET), 2023. — Empirical supplement to compute battleground analysis.
- State Council of the People’s Republic of China. Next Generation Artificial Intelligence Development Plan [新一代人工智能发展规划]. July 8, 2017. [primary, state] — PRC strategic AI planning document; the policy target of the US-China competition Scharre analyzes.