AI at War

Sam J. Tangredi’s edited volume arrived in 2021 from the Naval Institute Press, the in-house publisher of a service that had spent the previous decade watching the People’s Liberation Army Navy commission warships at a pace not seen since the Cold War. Tangredi, a retired US Navy captain who holds the Leidos Chair of Future Warfare Studies at the US Naval War College in Newport, framed the project as a corrective. Most of the conversation about artificial intelligence in war up to that point had been written by people who thought in terms of land combat, drones overhead in counterinsurgency theatres, and the targeting cycles of joint terminal attack controllers. Naval warfare, with its enormous spatial scales, its sensor-saturated environments, its long stand-off engagements, and its peculiar problem of identifying friend from foe in international waters, had been comparatively under-served. The book gathers serving and retired officers, civilian analysts from federally funded research and development centres, academic specialists in robotics and computer science, and a handful of industry voices, and asks them to address a single question: what does AI actually do for navies, and what is it likely to do soon.

The argument that emerges across the chapters is not that artificial intelligence will hand victory to the side that builds the smartest algorithm. It is closer to the opposite. Tangredi and his contributors push back against the more breathless commentary about robot fleets and machine-speed war, and insist that AI’s first and largest contribution to naval power will be in the unglamorous middle of the kill chain: ingesting the deluge of sensor data that already overwhelms shipboard watchstanders, recognising patterns in acoustic and electromagnetic signatures, fusing intelligence across platforms that today cannot talk to each other, and freeing human operators to make the small number of decisions that actually require judgement. The hard problem, the book argues, is institutional. The US Navy has demonstrated capable autonomous prototypes for years. It has not figured out how to buy them in numbers, how to integrate them into a force structure still dominated by manned capital ships, how to train sailors to trust them, or how to write doctrine that accounts for them. The People’s Liberation Army Navy, working with the rest of the Chinese state behind it, faces fewer of those frictions and is moving faster than its slower hardware base would suggest.

The book is structured as an edited volume rather than a single-authored monograph, and that shape matters. Early chapters lay out what AI and machine learning actually are, in language pitched at a flag officer who has never trained a neural network. The contributors are careful to separate genuine current capabilities — narrow pattern recognition, image classification, route optimisation, anomaly detection in sensor feeds — from speculative general intelligence. They walk through the supervised, unsupervised, and reinforcement learning families, explain why deep learning has taken off in the past decade, and give the reader enough vocabulary to follow the rest of the book without being snowed by jargon. A middle block of chapters turns to specific naval mission areas: anti-submarine warfare, mine countermeasures, intelligence-surveillance-reconnaissance, electronic warfare, command and control, logistics, and personnel. Each chapter takes a mission and asks where AI is already producing a measurable gain, where it is being tried experimentally, and where the hard problems sit. A final block addresses obstacles: data infrastructure, the security of the algorithms themselves, the ethics of delegating lethal decisions to machines, the legal frameworks under which autonomous platforms operate at sea, and the human factors of getting a sailor to trust a recommendation produced by a black box.

The concrete heart of the book is its catalogue of programmes and platforms. The contributors return repeatedly to the Navy’s family of unmanned surface vessels — the medium displacement Sea Hunter built by Leidos under the DARPA Anti-Submarine Warfare Continuous Trail Unmanned Vessel programme, and the larger Ghost Fleet Overlord ships acquired by the Strategic Capabilities Office and transferred to the Surface Development Squadron. The argument is that these hulls are valuable not because they make the human sailor obsolete but because they let the Navy disaggregate its sensor and shooter functions across cheaper, more numerous, more attritable platforms — what the surface warfare community had begun to call distributed maritime operations. The book treats the Boeing Orca extra-large unmanned undersea vehicle as the analogous bet under water: a long-endurance, modular platform that can carry mines, conduct surveillance, or function as a forward node in an undersea sensing network without putting submariners in harm’s way. Sea-based ISR comes in for sustained attention, with discussions of how machine vision algorithms — descendants of the Project Maven work originally aimed at processing full-motion video from over land — could be turned on the much larger and noisier problem of identifying vessels in the open ocean from a constellation of satellite, aerial, and shipborne sensors.

Anti-submarine warfare receives some of the most detailed treatment in the volume, because it is the mission where AI’s pattern-recognition strengths map most cleanly onto an enduring problem. Modern diesel-electric submarines, including the Chinese Yuan class and the Russian Kilo and Lada families, are very quiet, and the acoustic signatures available to a sonar operator are buried in a great deal of biological and oceanographic noise. The contributors describe how machine learning trained on historical acoustic data sets can flag candidate contacts faster and more consistently than human operators, and how that capability scales when distributed across a network of unmanned surveillance buoys, undersea gliders, and persistent surface platforms. Mine countermeasures benefit from similar logic — autonomous undersea vehicles equipped with side-scan sonar can survey a contested approach lane, classify the seabed for mine-like objects, and either neutralise them or hand the contact off to an explosive ordnance disposal team. The book points to the Navy’s Knifefish programme as an early production example of the pattern, while noting that the institutional appetite for replacing manned mine countermeasure ships has historically been weaker than the technical case for doing so.

Command and control is the other dominant theme. Several chapters take up the Joint All-Domain Command and Control concept, which the Department of Defense had begun to articulate in the late 2010s as a way of stitching together sensors, shooters, and decision-makers across the services and across all five operating domains. The naval contribution to that vision rests on AI in two ways: as the connective tissue that translates between previously incompatible data formats and tactical data links, and as the recommendation engine that helps a strike group commander prioritise targets and sequence engagements faster than a Chinese fleet operating under its own version of the same logic. The contributors are careful here. They do not argue that the algorithm replaces the commander. They argue that the commander who can ingest a fused common operational picture, query it in natural language, and receive a course-of-action recommendation in seconds will defeat a commander who is still being briefed by staff officers reading paper charts.

The China chapters are the book’s centre of gravity, and they sit somewhat uncomfortably alongside the more technical material. The contributors describe a Chinese naval and academic establishment that is enthusiastic about AI in a way the US Navy has not been, with the People’s Liberation Army talking openly about “intelligentised warfare” as a stage beyond the informatised warfare it pursued in the 2000s and 2010s. The volume points to Chinese investment in unmanned surface and undersea platforms, to commercial AI champions like Baidu, SenseTime, and iFlytek whose work bleeds into defence applications under the country’s military-civil fusion strategy, and to a maritime militia that increasingly uses commercial vessels equipped with sensors and communications gear as an outer surveillance layer in the South and East China Seas. The contributors do not predict a Chinese victory, but they treat the contest as real and the US institutional response as inadequate. The Pentagon’s Joint Artificial Intelligence Center, stood up in 2018, gets credit for at least naming the problem; the procurement system gets less credit for solving it.

The ethics and human factors chapters round out the volume. The contributors engage seriously with the autonomous weapons debate that Paul Scharre, Heather Roff, and others had been conducting for several years by the time the book appeared, but they do so from a naval perspective that is slightly less alarmist. Their case is partly that the maritime environment is structurally cleaner from a legal standpoint than the urban land environments where most of the autonomous weapons concern is focused: a hostile warship in international waters is identifiable in a way that an insurgent in a village is not, and the rules of engagement for naval combat have a longer and more settled doctrinal lineage. Their case is also partly that human-on-the-loop and human-in-the-loop architectures, in which the algorithm proposes and the human disposes, are both technically feasible and operationally desirable for the kinds of engagements navies actually conduct. The book treats the prospect of fully autonomous lethal engagement at sea as something the United States is unlikely to authorise and unlikely to need.

The reception of the book has been steady rather than spectacular, in the way of edited volumes published by service presses. It has become a standard reference on professional reading lists at the Naval War College and the US Naval Academy, and it is cited in the unclassified literature on distributed maritime operations and unmanned naval systems. Reviewers in the naval and defence press have generally taken the contributors’ technical claims at face value while pushing back on two points. The first is that the book, like much of the institutional Navy, is more confident about the strategic logic of unmanned platforms than the budget and acquisition system has so far been willing to ratify; several years after publication, the medium and large unmanned surface vessel programmes are still moving more slowly through congressional approval than the contributors’ arguments would suggest they should. The second is that the China material, while substantively right, sometimes reads as more confident about PLA Navy capabilities than the open-source evidence strictly supports — a familiar pattern in the China-watching literature of the period. Subsequent work by the Center for Strategic and Budgetary Assessments, the RAND Corporation, and the Center for a New American Security has gone deeper on specific operational concepts like mosaic warfare and decision-centric warfare, building on but also complicating the picture the book paints.

For a reader assembling a working library on artificial intelligence in war, the volume fits a specific slot. Paul Scharre’s Army of None covers the autonomous weapons ethics question in greater depth and across a broader range of services. Christian Brose’s The Kill Chain makes the institutional and acquisition-reform argument more sharply and at greater length. Books like Kenneth Payne’s I, Warbot or Ben Buchanan and Andrew Imbrie’s The New Fire address the broader machine-learning revolution and its strategic implications more directly. None of those is a naval book. Tangredi’s volume is, and that specificity is its main value. A reader who wants to understand why the surface, submarine, and aviation communities of the US Navy think about AI in the way they do, what specific platforms and programmes are being held up as proof of concept, and how the China challenge looks when viewed from a Newport classroom rather than from Washington, will find more of that material here than in any single comparable book.

Parts of the volume have aged faster than others. The specific programmatic details — which unmanned surface vessel prototype is in which phase, how many of which platform the Navy is requesting in which budget year — were already in motion when the book went to press and have continued to shift. The broader framing has held up better. The argument that AI’s first contribution to naval power will be in sensor fusion and decision support rather than in autonomous lethal engagement, the argument that the institutional and acquisition obstacles matter more than the technical ones, and the argument that the Chinese state is moving faster on military AI than the US national security establishment has been willing to internalise — these are still the load-bearing claims of the naval AI conversation, and they are still under-served elsewhere. The book reads, several years on, less as a snapshot of a moment and more as the opening statement of a debate that the Navy is only beginning to have with itself.