Algorithms of Armageddon

Algorithms of Armageddon arrived in 2024 from the Naval Institute Press, a house better known for memoirs of carrier aviators and reference works on warship classes than for futurist tracts. That provenance matters. Its authors, George Galdorisi and Sam J. Tangredi, are not academics writing about war from a distance. Galdorisi spent a career in naval aviation, commanded a helicopter squadron and an air wing, and now leads strategic assessments at the Naval Information Warfare Center Pacific in San Diego. Tangredi retired a Navy captain, took command of frigates and a destroyer squadron, and holds the Leidos Chair of Future Warfare Studies at the U.S. Naval War College. The book joined a crowded shelf — Paul Scharre’s Four Battlegrounds had landed the previous year, Christian Brose’s Kill Chain in 2020, Kenneth Payne’s I, Warbot the year before that — but it staked out a particular position within that conversation: the practitioner’s view, written for officers and policymakers, with the operational tempo of the Pentagon in mind rather than the rhythm of a seminar room. Admiral James Stavridis, the former NATO supreme allied commander, supplied the foreword, and his presence signals the audience the book is courting.

The central argument is blunt. Artificial intelligence is already reshaping warfare and will redefine it within a decade, and the United States is moving too slowly to keep its edge. Galdorisi and Tangredi do not see AI as a future problem; they see a competition already underway, with China sprinting and Russia improvising. They reject what they describe as the two dominant cultural frames in the West — the Hollywood “Skynet” frame, which treats every advance as a step toward an existential catastrophe, and the academic-ethics frame, which they argue has paralysed defence acquisition by demanding philosophical certainty before operational adoption. Their counter-position is that responsible AI in war is not only possible but already happening in narrow forms, that meaningful human control is compatible with autonomy at the tactical edge, and that the greater moral risk lies in arriving late to a fight against adversaries who have made no such ethical accommodations. The book is, in effect, an argument for speed under guardrails.

The structure follows a deliberate arc. The opening chapters set the strategic stage, framing AI as the latest and arguably the most consequential of the revolutions in military affairs that have punctuated the past century — the machine gun, the tank, the aircraft carrier, the nuclear weapon, precision guidance, networked sensors. The authors trace the history of computing in war from Bletchley Park and ENIAC through the early Cold War decision-support systems, into the precision-strike revolution that culminated in Operation Desert Storm, and onward to the network-centric warfare doctrines of the late 1990s. From there the book pivots to the present, with chapters devoted to the major military AI programmes inside the U.S. Department of Defense, the parallel investments by China and Russia, the application areas where AI is already in service, the ethical and legal questions that surround lethal autonomy, and finally a set of recommendations for how the United States should organise itself to win the competition. A penultimate chapter on what the authors call the “human dimension” — recruiting, training, and retaining the people who will work alongside these systems — is one of the book’s quieter contributions, and one that distinguishes it from purely technical surveys.

The evidence is dense and concrete, drawn from programmes the authors know well. Project Maven, the Pentagon’s pathfinder computer-vision effort that began in 2017 under then-Deputy Secretary Bob Work, gets a careful treatment that includes both its operational successes and the 2018 Google employee walkout that forced the company off the contract. The authors use that episode less to relitigate the politics than to examine how the Department learned, slowly, to build durable partnerships with commercial AI labs. The Joint Artificial Intelligence Center, stood up in 2018 and folded in 2022 into the new Chief Digital and Artificial Intelligence Office under Craig Martell, is mapped in detail, as is the Department’s 2020 adoption of five ethical principles for AI — responsible, equitable, traceable, reliable, and governable — which the authors treat as a serious document rather than a public-relations exercise. Joint All-Domain Command and Control, the doctrine for stitching together sensors and shooters across services and domains, runs as a thread through several chapters; the authors describe it as the connective tissue that any future force will need if AI-enabled decision aids are to scale.

The maritime examples are unsurprisingly the sharpest, given the authors’ backgrounds. Sea Hunter, the medium-displacement unmanned surface vessel that DARPA developed and the Navy now operates as part of its Unmanned Surface Vessel Division One, appears repeatedly as proof that long-endurance autonomous platforms are not science fiction. The Orca extra-large unmanned undersea vehicle built by Boeing receives sustained attention, as does the Ghost Fleet Overlord programme of experimental USVs. The Air Force’s Skyborg programme and the Loyal Wingman concept under the Collaborative Combat Aircraft initiative are presented as the airborne analogues — semi-autonomous, attritable platforms that fly alongside crewed F-35s and B-21s. The book draws on the Replicator initiative, announced by Deputy Secretary Kathleen Hicks in August 2023, which set a target of fielding thousands of all-domain, attritable autonomous systems within two years. Galdorisi and Tangredi treat Replicator as a useful test of whether the Department can in fact procure at scale and at speed, and they are guardedly optimistic about it as a forcing function rather than as a finished answer.

China gets a chapter of its own, organised around the concept of military-civil fusion as articulated by Xi Jinping after 2017. The authors lay out the People’s Liberation Army’s investments in unmanned platforms, the development of intelligentized warfare doctrine, and the role of national champions such as iFlytek and SenseTime alongside the better-known Huawei and DJI. They detail Beijing’s stated goal, set out in the 2017 New Generation Artificial Intelligence Development Plan, of becoming the world leader in AI by 2030 — and they note the discipline with which that plan has translated into research budgets, talent pipelines, and procurement. Russia is treated as a different kind of problem: less capable in advanced AI than China but more willing to use what it has, and learning rapidly in Ukraine. The book points to the proliferation of Lancet loitering munitions, the use of Iranian-supplied Shahed-136 drones against Ukrainian infrastructure, and the Ukrainian counter-investment in domestic drone production and AI-enabled targeting as a live laboratory for what algorithmic warfare looks like in 2023 and 2024.

On the ethics and law of lethal autonomy, the authors engage seriously with the critics — Stuart Russell, Toby Walsh, Mary Wareham of the Campaign to Stop Killer Robots, and the long-running Convention on Certain Conventional Weapons group of governmental experts in Geneva. They lay out the case for a preemptive ban on lethal autonomous weapons systems and then argue against it on three grounds. First, they contend that the proposed bans rest on definitions of autonomy so broad that they would catch existing defensive systems such as Aegis, Phalanx, and the Iron Dome, all of which already operate in autonomous modes against threats too fast for human decision loops. Second, they argue that the ban would be unverifiable because the autonomy of a weapon is a software state rather than a hardware feature. Third, they make the realist point that adversaries who have signalled no interest in such restrictions will field these systems regardless. Their preferred frame is the Department of Defense’s 2012 directive 3000.09, updated in January 2023, which requires “appropriate levels of human judgment” in the use of force and which they treat as a workable policy that should be exported rather than abandoned.

The book closes with recommendations. The authors press for faster acquisition reform of the kind Will Roper attempted at the Air Force; deeper integration with commercial AI firms and venture-backed defence companies like Anduril, Shield AI, and Palantir; new career fields for digital talent inside the services; and a willingness to break with industrial-age requirements documents in favour of iterative software-style development. They argue that the United States should compete on data infrastructure as aggressively as on models, and they single out the Pentagon’s cloud strategy — the Joint Warfighting Cloud Capability awarded to multiple vendors in 2022 — as a critical enabler. They want allies and partners closer to the work, with AUKUS Pillar 2 named as a model for technology-sharing arrangements that move at the pace of software rather than treaty diplomacy.

Reception in the defence-policy community has been broadly favourable, particularly among readers who already share the authors’ impatience. Reviewers in Proceedings, the Naval Institute’s own journal, and in War on the Rocks have praised the operational specificity and the discipline of writing about AI without lapsing into either utopia or dystopia. The book has been adopted on reading lists at war colleges and command-and-staff courses. Critics have pushed back on three fronts. Some have argued that the authors are too sanguine about the technical readiness of large-scale autonomous coordination, pointing out that systems demonstrated in clean tests have a long history of failing in contested electromagnetic environments. Others have noted that the ethical chapter, while substantive, leans heavily on official Department of Defense framing and gives less voice to the Geneva critics than a balanced treatment would require. A third strand of criticism comes from those who think the book underweights the role of the private cloud providers — Microsoft, Amazon, Google — in shaping what the Department can actually deploy, and the corresponding political risks of dependence on a small number of vendors. The arrival of large generative models since the book went to press has aged some of the technical sections; the authors discuss transformer architectures but predate the rapid integration of foundation models into staff planning, intelligence analysis, and code generation that has marked 2023 and 2024.

For a reader working through the field, Algorithms of Armageddon sits naturally alongside Paul Scharre’s Four Battlegrounds, with which it shares a strategic-competition frame, and Christian Brose’s Kill Chain, with which it shares the acquisition-reform critique. Where Scharre writes from the policy-research community and Brose from staff experience on Capitol Hill, Galdorisi and Tangredi write from inside the operational fleet and the war-college lectern. The book pairs less well with the more philosophical works in the field — Kenneth Payne’s I, Warbot, Daniel Trusilo’s Autonomous Weapons Systems and the Moral Equality of Combatants, or the legal arguments of Rebecca Crootof — but it engages with them respectfully and gives a reader a clear sense of where the practitioner community is willing to compromise and where it is not. What the book does not offer is a deep technical primer; readers wanting to understand backpropagation, reinforcement learning, or the failure modes of large language models should look elsewhere.

What will age best in the book is its insistence on the bureaucratic and human dimensions of military AI adoption — the parts that are about people, procurement, and partnerships rather than about algorithms. Those problems will look much the same in five years. What is already dated is the technical horizon: the authors wrote before foundation models became central to enterprise software, and a future edition will need to grapple with the implications of generative AI for command staffs, intelligence production, and information operations. The book’s framing of the competition with China and Russia has, if anything, been reinforced by events since publication, and the Ukrainian theatre has continued to validate many of its tactical predictions about drone-saturated battlefields. The book reads, in 2024 and into 2025, as a serious operator’s argument for moving with urgency — and as a record of how the people responsible for fielding these systems think about the work in front of them.