Michael C. Horowitz

BLUF

Michael C. Horowitz is an American political scientist whose “adoption-capacity” theory of military innovation — developed in academic work and tested against direct service in the Office of the Secretary of Defense — provides this vault with its primary theoretical framework for analyzing why AI-enabled military capabilities spread unevenly and why doctrine consistently lags technology acquisition. His work is directly applicable to every vault investigation involving Lethal Autonomous Weapons Systems, Algorithmic Warfare, and algorithmic targeting systems. The vault requires this note because Horowitz bridges academic military innovation theory and operational AI governance in ways that no other single analyst in the network currently covers — he is the essential reference point for understanding why the IDF’s Gospel/Lavender targeting architecture, US TITAN integration, and Chinese AI-enabled ISR all follow different diffusion trajectories despite similar underlying technology.

Core Contributions

The Diffusion of Military Power (2010)

Horowitz’s foundational academic text introduces adoption-capacity theory to explain variation in military innovation diffusion. The central argument: the ability to successfully adopt a new military technology is a function of two independently variable types of capital:

• Financial capital: The resources required to acquire the technology (purchase platforms, fund R&D, integrate systems). This is the more easily satisfied condition — states with resource access can clear the financial threshold relatively quickly.
• Organizational capital: The institutional knowledge, doctrinal adaptation, and human capital required to integrate the technology into effective military operations. This is the binding constraint — it requires rewriting doctrine, retraining personnel, reorganizing command structures, and tolerating the institutional disruption that accompanies any major capability change.

Fact: The adoption-capacity framework generates a specific and testable prediction: states with high financial capital but low organizational capital will acquire hardware without achieving effective integration, producing capability-doctrine gaps that create distinctive operational vulnerabilities. This pattern — sophisticated platforms deployed without supporting doctrine — is observable in multiple current contexts and is directly applicable to vault investigations of AI weapons integration.

Assessment: The 2010 text’s analysis of historical cases (armored warfare adoption, aircraft carrier diffusion, submarine proliferation) establishes the theoretical rigor of the framework before it is applied to AI. Analysts should engage the historical cases as well as the AI applications — the framework’s predictive power rests on its historical validation, not solely on its intuitive plausibility.

AI and Military Innovation

Building directly on adoption-capacity theory, Horowitz has systematically applied the framework to artificial intelligence as a military technology. His key analytical findings:

The AI diffusion paradox: AI will diffuse more rapidly than previous transformative military technologies because commercial AI development dramatically lowers the financial capital threshold — states and non-state actors can access foundational AI capabilities through commercial APIs, open-source models, and dual-use platforms without requiring indigenous R&D programs. However, effective military integration will lag because the organizational capital requirements — rewriting targeting doctrine, training operators, establishing accountability structures, integrating AI recommendations into command decision loops — are independent of commercial availability and take institutional time to develop. Assessment: This paradox is directly observable in current vault-relevant cases: the IDF Gospel/Lavender systems represent high-financial-capital, high-organizational-capital integration (long-term investment in doctrine alongside capability); many other states acquiring similar targeting AI will face the capability-doctrine gap that adoption-capacity theory predicts.

AI’s actual military impact domains: Horowitz has argued consistently that AI’s near-term military impact is concentrated in specific capability domains: intelligence, surveillance, and reconnaissance (ISR) data processing and target identification; logistics optimization and predictive maintenance; cyber operations (both offensive pattern recognition and defensive anomaly detection). The “robot soldier” scenarios that dominate public discourse are significantly less near-term than ISR and logistics applications. Fact: This assessment aligns with observable deployment patterns across all major military AI programs tracked in the vault.

The Lethal Autonomous Weapons Systems Governance Problem

Horowitz has been a central analytical voice in the UN-level debate over LAWS governance, engaging seriously with both the military utility arguments and the accountability concerns. His analytical contribution to this debate:

The “meaningful human control” operationalization problem: The primary proposed governance standard — that autonomous weapons systems must remain under “meaningful human control” — is, in Horowitz’s assessment, insufficiently defined to be actionable. Fact: The standard does not specify: what constitutes a human decision versus a human ratification of a machine decision; what minimum review time creates accountability rather than rubber-stamping; which elements of a targeting sequence require human judgment and which can be delegated to autonomous systems. Horowitz has argued that governance frameworks need to answer these specific operational questions rather than invoking a general “human control” standard that different actors will interpret to permit whatever they already do.

Accountability gap analysis: When an autonomous system executes a targeting decision that results in a violation of International Humanitarian Law (IHL), the existing legal accountability structures have no clear point of application. The operator who approved the mission profile, the engineer who designed the system, the commander who authorized deployment, and the AI system itself each represent partial accountability nodes — but none cleanly satisfies the traditional legal concept of command responsibility. Horowitz has documented this gap systematically and assessed it as a fundamental legal architecture problem requiring multilateral negotiation, not unilateral restraint by leading states.

Pentagon Service (2022–2024)

As Deputy Assistant Secretary of Defense for Force Development and Emerging Capabilities, Horowitz moved from academic analysis to policy implementation. Fact: This role placed him inside the development of US Department of Defense AI adoption policy, including frameworks for responsible AI integration in military operations and the department’s autonomous weapons review process. His public outputs from this period are constrained by security obligations, but his post-government writing will constitute primary documentation of internal US policy deliberations on AI warfare.

Analytical Framework

Horowitz’s methodology is comparative historical institutionalism applied to military technology. He derives generalizable propositions from systematic comparison of historical innovation adoption cases, then applies those propositions to current technology trajectories. The framework’s distinctive feature is its explicit separation of capability acquisition (a function of resources) from capability integration (a function of organizational adaptation) — a separation that most technology-focused military analysis collapses, producing overestimates of how quickly new technologies change operational outcomes.

The framework is deliberately non-deterministic: it predicts the conditions under which effective adoption is likely or unlikely but does not specify outcomes. States with high organizational capital may still adopt new technologies slowly due to bureaucratic resistance; states with low organizational capital may achieve partial integration that is militarily significant despite doctrine gaps. The adoption-capacity variables identify binding constraints, not certainties.

Analytical Positioning

Within this vault’s author network, Horowitz occupies the AI-and-military-innovation tier with a theoretical-empirical orientation that distinguishes him from primarily ethical or strategic-risk framings of the same subject. His adoption-capacity framework is the primary theoretical instrument for analyzing AI weapons diffusion; it should be the first analytical reference for any vault investigation involving state acquisition or deployment of autonomous military capabilities.

Horowitz is analytically distinct from Paul Scharre — whose work focuses more heavily on the strategic risks and ethical dimensions of autonomous weapons and whose Army of None (2018) provides the broader ethical and strategic risk survey — by his innovation-diffusion theoretical frame and his focus on comparative state behavior rather than case-by-case weapons ethics. Both are required for complete analysis of the LAWS domain. Gap: A Scharre profile note is not currently in the vault; cross-link pending creation.

Relative to the Algorithmic Warfare concept note in the vault, Horowitz provides the theoretical framework for understanding why AI military integration follows the patterns documented there. His adoption-capacity theory should be cited as the analytical frame whenever the vault note on AI warfare addresses diffusion variance across states.

For investigations involving the People’s Republic of China, Horowitz’s framework predicts that China’s AI military integration will be constrained not by financial capital (which PRC military modernization has in abundance) but by organizational capital — specifically, the doctrinal and command-structure adaptations required to integrate AI recommendations into PLA decision loops. Assessment: Early evidence suggests PLA AI integration is progressing faster on ISR applications (where existing targeting doctrine provides a ready integration path) than on autonomous decision-support in contested environments (where new command accountability structures are required). This is the predicted adoption-capacity pattern.

Limitation: Horowitz’s government service period (2022–2024) creates potential future constraints on his public analytical output for the near term. Additionally, the adoption-capacity framework was developed and validated primarily against large-state military innovation cases; its applicability to non-state actors, hybrid forces, and sub-state armed groups operating AI capabilities has not been systematically tested. Gap: Application of the framework to non-state actor AI adoption (e.g., drone warfare by Hezbollah, Houthi ISR capability development) remains underdeveloped in the literature.

Key Works

• The Diffusion of Military Power: Causes and Consequences for International Politics (Princeton University Press, 2010) — foundational adoption-capacity theory text.
• “Artificial Intelligence, International Competition, and the Balance of Power” (Texas National Security Review, 2018) — primary AI-warfare framework article.
• “Do Emerging Military Technologies Matter for International Politics?” (Annual Review of Political Science, 2020) — survey and synthesis of military innovation theory applied to AI.
• “The Promise and Peril of Military Applications of Artificial Intelligence” (Bulletin of the Atomic Scientists, 2018)
• Various policy briefs, Center for a New American Security (CNAS) — multiple publications on LAWS governance and AI policy.

Key Connections

• Lethal Autonomous Weapons Systems
• Algorithmic Warfare
• People’s Republic of China
• Russian Federation
• Hybrid Warfare

Sources

• The Diffusion of Military Power (Princeton, 2010) — High (primary text)
• University of Pennsylvania / Perry World House faculty biography — High (current, institutional)
• DOD service dates and title — High (confirmed public record, Senate confirmation hearing record)
• CNAS and journal publication record — High (public, verifiable)
• Birth year estimate — Medium (inferred from academic career timeline; not publicly stated)
• Post-government analytical constraints assessment — Assessment (structural inference, not documented)