I, Warbot

Hurst published Kenneth Payne’s I, Warbot in 2021 into a conversation that had been gathering pace for several years. Paul Scharre’s Army of None had appeared in 2018, Christian Brose’s The Kill Chain in 2020, and a steady stream of policy reports from the Center for a New American Security, RAND, Brookings, and various Pentagon offices had pushed the question of autonomous weapons from speculative ethics into operational planning. Payne, who teaches strategy in the Department of War Studies at King’s College London, came to that conversation from an unusual angle. His earlier book, Strategy, Evolution, and War, had treated war as a fundamentally cognitive activity rooted in evolved human psychology — emotion, reciprocity, risk-tolerance, the asymmetric weighing of losses and gains. I, Warbot extends that thread into the machine age. It asks what happens to a contest defined by human judgment when the agents doing the judging are no longer human at all.

The central argument is that artificial intelligence is not simply a faster, more accurate version of tools militaries already use. The book pushes back gently against the framing that AI is “just” better targeting, better surveillance, better logistics. Payne contends that war, at its core, is a contest of cognition under uncertainty, and that AI systems — particularly the deep-reinforcement-learning agents that astonished observers when DeepMind’s AlphaGo defeated Lee Sedol in 2016 and when AlphaZero subsequently mastered chess, shogi, and Go from self-play within hours — exhibit a different kind of cognition than the human kind. They are tactically brilliant in narrow domains, lack common sense, do not feel fear or honour, and produce moves that human grandmasters describe as alien. Payne’s contention is that the same alien quality will eventually appear on battlefields, and that the side which integrates non-human cognition into its decision loops will hold an advantage that is not merely quantitative but categorical. War, in this telling, becomes the first domain since chess to be transformed not by faster physical tools but by faster and stranger minds.

The book unfolds in a roughly chronological and thematic arc. Early chapters trace the deep evolutionary background of human conflict, drawing on the psychology of cooperation and competition that Payne developed in his earlier work. The argument here is that human warriors are biased decision-makers: they feel fear, want to live, are loyal to in-groups, and are subject to the well-mapped distortions of prospect theory and motivated reasoning. War-fighting has been shaped, throughout history, by the strengths and weaknesses of that wetware. From there the book moves into the long entanglement of computing and warfare — Alan Turing and Bletchley Park, the SAGE air-defence network, the Strategic Computing Initiative of the 1980s, the precision-guided munitions of Desert Storm — and then into the more recent inflection point at which machine learning, particularly deep neural networks trained on vast data, began to outperform hand-coded systems at perception tasks. The middle chapters describe the present moment, in which loitering munitions, drone swarms, autonomous undersea vehicles, and machine-learning-driven intelligence analysis are no longer hypothetical. The closing chapters speculate forward, asking what command, deterrence, escalation, and the conduct of war will look like when machines participate as cognitive agents rather than as tools.

The concrete examples are where the book earns its weight. The Israeli Harpy loitering munition, manufactured by Israel Aerospace Industries and in service since the 1990s, is treated as the first fielded weapon that clearly crosses the threshold into autonomous lethal targeting: it loiters over an area, identifies hostile radar signatures matching a stored library, and dives on them without further human authorization. Payne uses Harpy and its successor Harop to make the point that the autonomous-weapons debate is not about a future that might arrive but about a present that already has. He moves on to the Turkish Bayraktar TB2, which in 2020 in Nagorno-Karabakh demonstrated how a mid-tier state could field combat drones at scale and degrade a conventionally armed adversary’s armoured columns and air defences. The Azerbaijan–Armenia war is treated as a doctrinal warning shot — small, cheap, semi-autonomous airframes prosecuting an unmistakably decisive campaign. The Russian Lancet, the American Switchblade, the Chinese CH-series, and the Israeli Hero family all appear as variations on the same theme.

Underwater the book points at the Boeing Orca extra-large autonomous undersea vehicle and the various unmanned surface vessels emerging from the US Navy’s Ghost Fleet Overlord and Sea Hunter programmes. On the air side it treats the Skyborg autonomy framework, the Loyal Wingman concept developed jointly with Boeing in Australia, and the Mosaic Warfare doctrine pushed by DARPA, in which large numbers of cheap, expendable, networked platforms replace small numbers of exquisite manned systems. The Joint All-Domain Command and Control initiative — the Pentagon’s effort to fuse sensors and shooters across services and domains into a single machine-readable battlespace — is treated as the connective tissue that makes AI-enabled fighting possible. Project Maven, the Department of Defense initiative to apply machine vision to drone footage that became briefly notorious in 2018 when Google employees protested their company’s involvement, anchors the chapter on intelligence analysis.

On the Chinese side Payne discusses the People’s Liberation Army’s stated doctrine of “intelligentization”, a successor to its earlier shift toward “informatization”, and the work of researchers and institutions associated with the National University of Defense Technology. He notes the explicit Chinese ambition, set out in the 2017 New Generation Artificial Intelligence Development Plan, to lead the world in AI by 2030, and the close civil-military fusion through which commercial Chinese AI capability is meant to flow into defence applications. SenseTime, Megvii, iFlytek and the broader Chinese computer-vision and surveillance ecosystem appear as the industrial base that gives that ambition substance. The Russian side gets shorter treatment but includes the Uran-9 unmanned ground combat vehicle, the Lancet and KUB-BLA loitering munitions later seen in Ukraine, and Vladimir Putin’s much-quoted 2017 remark to schoolchildren that whoever leads in AI will rule the world.

The book engages seriously with the doctrinal and ethical literature on meaningful human control. Payne walks through the positions associated with the Campaign to Stop Killer Robots, the Article 36 reviews mandated by Additional Protocol I to the Geneva Conventions, and the Convention on Certain Conventional Weapons discussions in Geneva that, by 2021, had failed for nearly a decade to produce binding restrictions on lethal autonomy. He treats the moral case for keeping a human in the loop with respect but is not persuaded by the operational realism of doing so when reaction times collapse below human cognitive limits. The example he repeatedly returns to is air defence: the Aegis combat system, the Patriot, the Iron Dome — all already operate in modes where human operators authorise the system to engage threats faster than humans can think. The autonomous threshold, in this reading, has already been crossed in defence; the only open question is how far it crosses into offence.

Where I, Warbot becomes most distinctive is in its chapters on creativity and strategic genius. Payne is interested in whether AI can do what Clausewitz called coup d’oeil — the swift intuitive grasp of a battlefield that great commanders are said to possess. He uses the AlphaGo move 37 against Lee Sedol, a move human professionals had judged a mistake and which then turned out to be game-winning, as the canonical example of machine creativity in an adversarial setting. He extends the argument to wargaming. DeepMind’s later AlphaStar agent, which mastered the real-time-strategy game StarCraft II, is treated as a closer analogue to military command because it operates in continuous time with imperfect information and a mix of economic, logistical, and tactical decisions. The book speculates that machine commanders, freed of fear and political accountability, may execute strategies humans would dismiss as reckless and may converge on doctrines that look, to human observers, simultaneously brilliant and terrifying.

That last point becomes the bridge into the book’s anxieties about escalation. War, Payne reminds the reader, is governed historically by human caution. The Cuban Missile Crisis ended where it did partly because Kennedy and Khrushchev shared an intuition that this was getting too dangerous; Vasili Arkhipov refused to authorize the launch of a nuclear torpedo aboard B-59 in 1962 because he, a human, could be persuaded by argument and doubt. Machine agents have no such intuition. The book considers what happens when machine speed compresses crises that previously took days into seconds, and when machine-to-machine signalling replaces the slow, ambiguous, deliberately costly diplomatic signalling that has stabilised great-power competition. It does not predict catastrophe, but it dwells on the loss of the friction that used to give humans time to choose otherwise.

Reception in the strategic-studies community was favourable. Reviewers in International Affairs, Survival, and the RUSI Journal credited the book for taking the cognitive dimension seriously and for resisting the policy-brief tendency to flatten AI into a procurement question. Some pushed back on its readiness to treat learning-based systems as genuinely creative rather than as sophisticated function-fitters; others questioned whether the AlphaGo analogy survives the move from a closed-world game to the open, adversarial, deceptive, weather-and-politics-laden world of actual war. Several reviewers wanted more on logistics, sustainment, and the unglamorous back end of military AI — the place where the technology has actually been delivering operational value. Since publication, the war in Ukraine that began in February 2022 has provided an enormous natural experiment for many of the book’s claims. The widespread use of FPV drones, of Lancet and Switchblade loitering munitions, of commercial satellite imagery for targeting, and of Palantir-style data-fusion software at scale has confirmed the broad trajectory Payne sketched. The pace at which Ukrainian and Russian operators have iterated on tactics, software, and counter-drone measures has also vindicated his point that the side with faster learning loops wins.

For a reader working through the literature on AI and war today, I, Warbot occupies a particular niche. It is less a policy primer than Scharre’s Army of None and less an industrial-strategy polemic than Brose’s The Kill Chain. It is closer in spirit to books that ask what war is for and what cognition is — sitting near Christopher Coker’s Future War, Antoine Bousquet’s The Eye of War, and the more philosophical chapters of Stuart Russell’s Human Compatible. It pairs naturally with later works such as Scharre’s Four Battlegrounds and with the growing body of writing on the Ukraine drone war by analysts at the Royal United Services Institute and the Centre for European Policy Analysis. A reader who wants to know what the autonomous-weapons treaty fight looks like in detail will need a different book; a reader who wants to know what kind of mind is appearing on the battlefield will find this one well chosen.

What is likely to age well is the analytical frame. Payne’s insistence that the interesting variable is cognition, not horsepower, and that the human side of war is itself a particular cognitive architecture among possible others, is a way of seeing the field that survives most specific technology turnovers. The chapters on Clausewitzian judgment, on the psychology of risk, on the difference between game-tree search and lived strategic intuition, will still be readable in a decade. What is already dated, or will be soon, is the specific industrial landscape — the model names, the demonstration programmes, the company logos. The Skyborg framework has moved on, Project Maven has been transferred between agencies, the language of “intelligentization” has been overtaken by newer Chinese formulations, and the autonomous systems being fielded in Ukraine were barely visible when the manuscript closed. That is the predicament of any book on a fast-moving field, and the book accepts it. Its bet is that the underlying argument about minds — human and otherwise — meeting in war will outlast the particular machines through which the meeting first happens. On the evidence of the years since publication, that bet looks well placed.