Army of None — Scharre (2018)
BLUF
Army of None: Autonomous Weapons and the Future of War (2018) is the definitive English-language analytical treatment of lethal autonomous weapons systems (LAWS). Paul Scharre, drawing on practitioner experience as a 75th Rangers veteran and DOD Directive 3000.09 drafter, provides the foundational taxonomy of autonomy in warfare, identifies the structural accountability gap in IHL enforcement for machine-initiated targeting decisions, and documents the flash war escalation risk from autonomous system interaction. Post-publication developments — principally the IDF Gospel and Lavender programs, Project Maven, and TITAN — confirm and in several cases exceed Scharre’s analytical concerns.
Bibliographic Information
| Field | Detail |
|---|---|
| Author | Paul Scharre |
| Full Title | Army of None: Autonomous Weapons and the Future of War |
| Publisher | W.W. Norton & Company, New York |
| Year | 2018 |
| Winner | Council on Foreign Relations Arthur Ross Book Award, Gold Medal (2019) |
| Companion Work | Four Battlegrounds - Scharre (2023) |
Core Arguments
1. Autonomy as a Spectrum, Not a Binary
Fact: Policy debate on autonomous weapons frequently treats “autonomous” as a binary property — a weapon either is or is not autonomous. Scharre demonstrates this framing is analytically incoherent. Assessment: Autonomy in weapons systems varies across at least three independent dimensions: who selects the target (human identification vs. algorithmic identification), who authorizes the engagement (human approval vs. automated authorization), and what physical actions are automated vs. human-controlled. A system that requires human target identification but authorizes engagement automatically is partially autonomous in a meaningful sense that the binary framing cannot capture. Governance discussions and arms control proposals require this disaggregation to be legally or operationally meaningful.
2. The Human-Control Taxonomy
Scharre establishes the standard reference taxonomy for human oversight levels, now the baseline in LAWS governance literature:
| Level | Definition | Example Class |
|---|---|---|
| Human-in-the-loop | Human approves each individual engagement | Current manned platforms, most armed UAV doctrine |
| Human-on-the-loop | Human can override; system acts automatically within parameters | Phalanx CIWS, Iron Dome fire control |
| Human-out-of-the-loop | System selects and engages with no human involvement at any stage | Theoretical fully autonomous weapon; contested empirically |
Assessment: The human-on-the-loop category is where most operationally deployed systems currently sit and where the accountability gap is most acute. The operator nominally retains override authority but in practice cannot exercise meaningful review at machine decision speeds.
3. The Structural Accountability Gap
Assessment: This is Scharre’s most legally and strategically significant finding. When an autonomous system violates IHL — engages a protected civilian, fails proportionality analysis, disregards a surrender signal — no individual has made the targeting decision that constitutes the violation. The commander who authorized system deployment cannot be held accountable for a specific targeting decision the system made without human input. The programmer who wrote the engagement logic cannot be held accountable for how the algorithm applied that logic in a specific context the programmer did not design for. The operator who monitored the system cannot be held accountable for a decision made faster than human reaction time. Fact: This gap is structural, not a function of inadequate rules or poor training, and it has no resolution within current IHL doctrine. Gap: Scharre identifies the accountability gap clearly but does not propose a satisfactory legal architecture to close it. Post-publication proposals (strict liability for deploying commanders; pre-certification review boards) remain contested.
4. The Flash War Risk
Assessment: Autonomous systems engaging adversary autonomous systems at machine speed — in timeframes ranging from milliseconds to seconds — can produce escalation to general war faster than any political decision loop can intervene. Scharre draws the analogy explicitly to the 2010 Flash Crash in equity markets, where algorithmic trading systems interacting without human intervention produced a 1,000-point Dow collapse in minutes before circuit breakers halted trading. Assessment: The military equivalent involves no circuit-breaker equivalent in current doctrine or treaty law. The risk is not accidental escalation of a human-authorized conflict but the initiation of a conflict in which no human made the decision to begin hostilities. This risk is categorically distinct from the accident risks analyzed in nuclear safety literature and requires a separate analytical framework.
5. The “Sorcerer’s Apprentice” Problem
Assessment: Complex autonomous systems in dynamic adversarial environments produce emergent behaviors that designers did not anticipate and that operators cannot reliably predict or constrain. Scharre’s historical case studies — including friendly fire incidents from early automated defensive systems — demonstrate that emergent failure modes are not a theoretical concern but a documented operational pattern. Assessment: The problem intensifies under adversarial conditions, where an opponent is specifically attempting to cause emergent failures through sensor spoofing, jamming, or novel environmental presentations. The behavior of autonomous weapons systems under adversarial exploitation of their emergent failure modes is fundamentally unpredictable in advance of deployment.
Structure
| Part | Focus | Analytical Contribution |
|---|---|---|
| I — What Is a Robot? | Technical baseline on autonomy | Definitional precision; spectrum taxonomy |
| II — The Robots Are Coming | Survey of current/near-future capabilities (as of 2018) | Empirical baseline; now significantly outdated |
| III — The Killer Robot Problem | Core governance analysis | Accountability gap; flash war risk; sorcerer’s apprentice |
| IV — Should We Ban Killer Robots? | Arms control options | Prescriptive; least developed section |
Methodological Significance
Scharre combines practitioner authority with systematic research across technical, legal, and strategic domains. The human-control taxonomy has been adopted in subsequent UN Group of Governmental Experts (GGE) discussions on LAWS and in the ICRC’s position papers on autonomous weapons. Fact: The book’s conceptual framework — particularly the three-level taxonomy and the accountability gap analysis — is now the standard reference in English-language LAWS governance literature.
Critical Assessments
Assessment: Published in 2018, the book predates the documented deployment of the IDF Gospel targeting system (operational 2019–2024), the Lavender mass-casualty targeting algorithm (operational 2023–2024), Project Maven’s operational maturation, and the US Army’s TITAN ground system. Each of these post-publication developments both confirms Scharre’s concerns — the accountability gap is now empirically instantiated in documented operational programs, not merely a theoretical risk — and exceeds his predictions on deployment pace and scale. The book’s arms control prescriptions remain analytically underdeveloped relative to its diagnostic power. Part IV on arms control proposals does not adequately address the verification challenges that make LAWS arms control qualitatively harder than nuclear or chemical weapons control.
Contemporary Relevance for This Vault
Army of None provides the foundational analytical framework for the vault’s active investigation of the IDF Kill Machine and for the AI targeting systems tracked in Emerging & Dual-Use Technologies. The accountability gap framework is directly applicable to the analytical assessments of Gospel and Lavender: neither system satisfies the IHL individual accountability requirement, and both instantiate exactly the structural gap Scharre identifies. The flash war risk analysis bears on the Ukraine theater’s drone warfare dynamics, where algorithmic engagement cycles are compressing to timeframes below reliable human-on-the-loop review. Companion reading: Four Battlegrounds - Scharre (2023) for the strategic-competition-level extension of this analysis.
Key Connections
- Paul Scharre
- Michael C. Horowitz
- Four Battlegrounds - Scharre (2023)
- Lethal Autonomous Weapons Systems
- Algorithmic Warfare
- Lavender
- The Gospel
- TITAN
- The IDF’s Kill Machine
- Ukraine War
- Gaza War
- Foundational Books
Sources
- Scharre, Paul. Army of None: Autonomous Weapons and the Future of War. New York: W.W. Norton & Company, 2018.
- ICRC. Autonomy in Weapon Systems: Technical, Military, Legal and Humanitarian Aspects. Geneva: ICRC, 2016. — Parallel analytical track at international law level.
- United Nations Group of Governmental Experts (GGE). Report on Lethal Autonomous Weapons Systems. UN Doc. CCW/GGE.1/2018/3, 2018. — Policy context; Scharre’s taxonomy appears in GGE submissions.
- Human Rights Watch and IHRC. Losing Humanity: The Case Against Killer Robots. 2012. — Pre-publication advocacy framing that Scharre’s analytical framework subsequently superseded.
- Abraham, Yuval, et al. “‘Lavender’: The AI Machine Directing Israel’s Bombing Spree in Gaza.” +972 Magazine / Local Call, April 3, 2024. [primary, investigative] — Post-publication empirical instantiation of Scharre’s accountability gap analysis.