Artificial Intelligence and the Future of Warfare

Artificial Intelligence and the Future of Warfare arrived in 2021 from Manchester University Press, at a moment when conversations about military AI were shifting from speculation to procurement. James Johnson, a strategic studies scholar based in the United Kingdom whose earlier journal work had mapped the intersection of emerging technology and great-power competition, set out to do something narrower than the broader “rise of the robots” books then on shelves. Rather than survey autonomous weapons across services and countries, Johnson trained the analysis on a single dyad — Washington and Beijing — and on a single durable question of nuclear-era strategy: what happens to strategic stability when a general-purpose technology like AI is layered onto an already tense competition. The book grew out of a cluster of papers Johnson had been publishing on intelligentization, hypersonic systems, and the cyber-nuclear nexus, and it landed alongside Paul Scharre’s Army of None, Kenneth Payne’s I, Warbot, Michael Horowitz’s work at the Center for a New American Security, and a swelling output from RAND and the Center for Security and Emerging Technology at Georgetown. Johnson’s contribution to that conversation is precise rather than panoramic: the question is not whether AI will change war, but whether it will make a nuclear-armed United States and a nuclear-armed China more likely to stumble into a war neither wants.

The central argument is that AI is best understood not as a discrete weapon but as a general-purpose, dual-use technology that diffuses through every layer of the military enterprise, from sensor fusion and intelligence analysis to command-and-control to the autonomous behavior of individual munitions. Because AI is general-purpose, its strategic effects are cross-cutting and difficult to model. Johnson argues that the net effect on US-China strategic stability is likely to be destabilizing, but not for the reasons the popular imagination tends to fixate on. The destabilizing pressure does not come primarily from “killer robots” replacing human soldiers, or from a Skynet-style autonomous nuclear release. It comes from less cinematic places: AI-enhanced intelligence, surveillance, and reconnaissance that erodes the survivability of second-strike nuclear forces; machine-speed decision support that compresses the time available for political leaders to deliberate; the entanglement of conventional and nuclear command systems, where an AI-enabled conventional strike on a Chinese radar or submarine pen could be misread as the opening move of a counterforce campaign; and the cognitive effects on commanders who increasingly trust, distrust, or over-rely on opaque algorithmic outputs. The US-Soviet Cold War, Johnson reminds the reader, had a thick conceptual architecture — mutual assured destruction, second-strike survivability, the “stability-instability paradox” — that took decades to build and codify in treaties. The US-China rivalry has no equivalent, and AI is arriving before the conceptual architecture does.

The book is organized to walk the reader from foundations into specific application domains and then out into governance. Early chapters set the scene by laying out what military AI actually is — machine learning, computer vision, autonomous navigation, decision-support systems — and by reviewing the two countries’ national strategies. On the American side, Johnson tracks the Department of Defense’s Third Offset Strategy under Robert Work, the establishment of the Algorithmic Warfare Cross-Functional Team known publicly as Project Maven, the founding of the Joint Artificial Intelligence Center, and the procurement turn toward AI-enabled platforms across the services. On the Chinese side, he reconstructs the People’s Liberation Army’s doctrine of “intelligentized warfare,” a successor to the earlier framing of “informatized warfare,” and traces it through PLA Academy writings, the State Council’s 2017 New Generation AI Development Plan, and the civil-military fusion policy that channels work from companies like Baidu, SenseTime, and iFlytek into defense applications. Middle chapters then take specific functional domains in turn: AI in intelligence and reconnaissance, AI in conventional precision strike, AI in cyber operations, AI in nuclear command, control, and communications, AI in autonomous weapons systems, and AI in early warning. Each chapter follows a similar pattern — what the technology does, how each side is fielding it, and how the introduction of that capability tilts the deterrence equation. Later chapters move to escalation dynamics and to the question of arms control: whether AI-specific arms control is even possible given that AI is software, dual-use, and developed largely in the private sector.

The concrete heart of the book is the evidence Johnson marshals across those middle chapters, and it is here that the analysis acquires its texture. On intelligence and reconnaissance, he reads AI-enabled ISR as a particular threat to mobile and concealed strategic assets — road-mobile intercontinental ballistic missiles, ballistic-missile submarines, leadership bunkers — because the durable assumption of nuclear deterrence has been that a determined adversary cannot find and target every leg of a triad in the time available. Massed satellite constellations, drone swarms, persistent maritime surveillance, and machine-vision pipelines trained to detect deception change that assumption at the margins, and margins matter when the surviving force is measured in tens. He looks closely at anti-submarine warfare in particular, where the Chinese SSBN force operating from Hainan must transit shallow chokepoints monitored by an increasingly dense American sensor net. On conventional strike, he examines hypersonic glide vehicles and air-breathing hypersonic cruise missiles — Chinese systems like the DF-17 and emerging American programmes — and notes that AI matters here less because it flies the missile than because it processes the targeting data fast enough for the missile’s short flight time to be useful. Hypersonics shorten the decision window; AI shortens it further. On nuclear command and control, Johnson examines Russian and American gestures toward integrating AI into early warning — the Perimetr “Dead Hand” lineage on one side, modernization discussions on the other — and dwells on the dangers of automation bias when a duty officer has minutes to interpret an algorithmic alert. On autonomous weapons, he steps through the debates at the United Nations Convention on Certain Conventional Weapons, the campaigns by the International Committee for Robot Arms Control and Human Rights Watch, and the Pentagon’s Directive 3000.09 standard of “appropriate levels of human judgment.”

He returns repeatedly to the problem of entanglement. American and Chinese conventional and nuclear forces share early-warning satellites, share command networks, and in some cases share basing — Chinese DF-26 missiles are explicitly dual-capable, and the People’s Liberation Army Rocket Force commingles conventional and nuclear units. An AI-enabled American precision strike intended to suppress Chinese conventional missiles in a Taiwan contingency could plausibly degrade nuclear early-warning infrastructure as a by-product, and the algorithms helping Chinese commanders make sense of the strike will be looking at incomplete sensor data under time pressure. Johnson draws here on Caitlin Talmadge’s work on inadvertent escalation, on Fiona Cunningham and Taylor Fravel on Chinese nuclear doctrine, and on Forrest Morgan’s RAND modelling. The reader comes away with a clear picture of why the AI-enhanced version of a US-China conventional war could escalate faster and less deliberately than its non-AI cousin.

The book has been read in the field as a careful, slightly worried mid-decade synthesis. Reviewers have credited Johnson with taking strategic stability seriously as a category rather than letting it collapse into “arms races good or bad,” and with treating the Chinese side of the dyad on its own doctrinal terms rather than as a mirror of American thinking. Where colleagues have pushed back, they have pushed in two directions. One line of disagreement, associated with more techno-skeptic voices like Heather Roff and elements of Scharre’s later writing in Four Battlegrounds, holds that contemporary machine learning is more brittle than Johnson’s worst cases imply — that the actual deployed systems hallucinate, fail under distribution shift, and would not survive contact with a peer adversary’s electronic warfare, and that this brittleness is itself stabilizing because no commander would entrust a nuclear-adjacent decision to such a system. The other line, associated with scholars like Kenneth Payne and with Chinese writing on intelligentization, holds that Johnson, if anything, understates the transformation, because he treats AI as an overlay on existing platforms rather than as an enabler of entirely new operational concepts like Mosaic Warfare, swarming, and human-machine teaming at the small-unit level. Johnson himself has continued this conversation in subsequent books, including AI and the Bomb in 2023, which deepens the nuclear thread of the 2021 volume.

For a reader trying to assemble a working library on AI and war, this book sits in the academic, strategic-studies corner rather than the popular-trade corner. It pairs naturally with Scharre’s Army of None, which provides the broader survey of autonomous weapons and the human-in-the-loop debate; with Payne’s I, Warbot and Strategy, Evolution, and War for the cognitive and theoretical layer; with Horowitz’s research on diffusion of military technology; with Elsa Kania’s reporting on Chinese intelligentization; and with the Congressional and CSET grey literature on specific procurement programmes. Where Johnson is strongest is on the US-China dyad and on the nuclear-conventional bleed; where he is necessarily thinner is on Russia, on the European defence industrial picture, on the Middle Eastern proving grounds where loitering munitions and Turkish-built drones have actually been used in anger, and on the Ukraine theatre that opened up after the book went to press. Anyone reading the book in the mid-2020s will want to bring those theatres back in from outside sources.

What is likely to age well in the book is the framework. Johnson’s insistence on treating AI as a general-purpose technology with effects that ripple through ISR, C2, and decision-making, rather than as a single weapon class, has been borne out by procurement patterns since 2021. So has his focus on entanglement and on the compression of decision time as the chief escalation risks. The discussion of automation bias and of opaque algorithmic outputs reads, if anything, more sharply now that large language models have entered the operational conversation. What is already dated is the inventory: specific programmes have been renamed, reorganized, or absorbed; Project Maven has moved out of the Pentagon and into a different bureaucratic home; the Joint AI Center has been folded into the Chief Digital and AI Office; hypersonic test schedules on both sides have slipped and accelerated; Chinese civil-military fusion has been reshaped by export controls on chips and lithography. A reader picking up the book today is best served by treating its framing chapters as the lasting contribution and its programme-by-programme detail as a useful 2020 snapshot rather than a current map. The harder question the book poses — whether the institutions of arms control and crisis management can catch up with a technology that is being fielded faster than it is being understood — is still open, and that is the question the book asks the reader to keep asking.