Why We Need Humans on the Battlefield
Howgate Publishing Limited, 2025, 352 pp, £19.99
Author: Paddy Walker
Reviewed by: Callum Hamilton and Benjamin Wood
The integration of artificial intelligence (AI) into weapons systems has become the new paradigm for major militaries, motivated by desires to overcome human limitations and to achieve superior operational effects. States, militaries and non-government organisations alike are considering the prospects of integrating AI and robotic and autonomous systems (RAS) in military operations. This has triggered international debate about the legality and ethics of autonomous weapons systems (AWS)[1]—‘killer robots’ capable of ‘selecting targets and initiating attacks without human involvement’.[2]
Paddy Walker’s War Without Oversight contributes to this conversation by presenting a multidisciplinary argument against the military use of AWS. The book analyses the development and use of AWS, assessing both practicality and feasibility across legal, operational and technological dimensions. In doing so, it articulates the opportunities and challenges relating to the development of lawful and operationally effective autonomous functions in weapons systems. By introducing the ‘Delivery Cohort’[3] concept, Walker enhances the text’s analysis and broadens the discussion beyond developers and end users to the wider ecosystem of actors that shape AWS capability outcomes. In a field often dominated by hype and futurist speculation, the provision of counterfactuals to claims of the inevitable ubiquity of AWS in military operations adds necessary nuance to anchor the AWS debate in what is possible and practical. Walker concludes that ‘truly unsupervised’ AWS are neither practical nor feasible for military use, and that autonomy in the targeting cycle must be limited to preserve human judgement.
The book’s focus on end-to-end autonomy in the targeting cycle means its scope exceeds what is immediately relevant to capability planners, commanders or developers, who generally work with more limited autonomisation efforts. Although the book never defines its intended audience, its argument for a legislated standard of ‘meaningful human control’ (MHC) clearly aims to shape the views of policymakers and international advocacy groups. This gives it particular relevance to forums such as the Group of Governmental Experts on Lethal Autonomous Weapons Systems[4] and the Summit on Responsible Artificial Intelligence in the Military Domain (REAIM).[5] The book contends that the principles emerging from these forums have had limited practical effect, largely due to a lack of consensus on how they should be operationalised and because of the fragmented international and interdisciplinary understanding of the subject. To counter this, Walker recommends establishing mechanisms to ensure that the use of AWS is subject to MHC applied to ‘when, where and how weapons are used; what or whom they are used against; and the effects of their use’.[6]
War Without Oversight also examines the drivers behind increasing weapon autonomy and the extent to which it is here to stay. Walker explores how elements of the character of warfare both shape and are shaped by the use of AWS, particularly through the capability offset provided by small, inexpensive RAS platforms that support force dispersal and mass. Walker argues that this offset has become increasingly important as personnel and exquisite platforms grow more vulnerable and more costly to sustain on a per‑unit basis. The book illustrates this trend through examples from Ukraine, where the ‘fragility’ of modern military hardware has seen autonomy used to enhance platform resilience to electronic warfare and reduce reliance on radio-based control. This shows how weapon autonomy is currently being applied, mainly to functions for which humans are poorly optimised or otherwise limited in comparison, including traditional command, control and communications models. These sections reflect a view of military operations clearly informed by contemporary conflicts, particularly the ongoing Russo-Ukrainian War.
The book further explores the strengths and weaknesses of various deployment models for autonomous systems. Walker emphasises that no single set of deployment rules can equip AWS to navigate the inherent uncertainty of military environments. Although the book recognises that AWS deployment must be aligned to fit between a system’s capabilities and its intended use case, it does not examine the architectural implications of such an approach or reconcile it with a unified standard for MHC. The author notes in later sections—without expanding on the point—that there is a need for a process that can continually determine an appropriate battlefield standard aligned with variations in context rather than depending on a single, fixed and fragile one. Further analysis of this issue would have added significant value to the book, given its centrality to operationalising military RAS and AI responsibly.
Environmental uncertainty introduces further challenges to prevailing narratives of technological feasibility of AWS across hardware, software, firmware and ‘wetware’, the latter being ‘the human element of information technology architecture … describing programmers’ efforts to replicate the essentials of human intelligence and to effect this through code and code alone’.[7] Walker argues that the technological complexity involved in developing a system capable of achieving a semantic understanding of its environment makes it impracticable to develop and deploy ‘truly unsupervised’ AWS without unacceptable sacrifices relating to output quality, legal compliance, and exposure to risk. This argument is bolstered by the text’s exploration of non-technological limitations to AWS deployment, including adherence to international humanitarian law (IHL), accountability for AWS use, reluctance to adopt innovative but unproven technologies, and the proliferation of countermeasures. One of the book’s findings is that, regardless of whether it is technologically feasible to do so, the unsupervised nature of AWS would prevent their use in accordance with legal and operational expectations.
Much of the book’s analysis stops short of synthesising recommendations for how autonomy in weapons systems should be practically controlled. This is partly because it focuses on AWS with ‘humans out of the loop’, rather than the more immediate trend toward autonomous systems with narrow functions that are expected to shape the near-term conduct of warfare under ‘human-in-the-loop’ or ‘human-on-the-loop’ control models. Consequently, the book’s emphasis on the most advanced and speculative forms of AWS makes its core arguments less immediately relevant to practitioners and to the military use of RAS and AI today. As a result, its discussion of how MHC might be achieved for hypothetical AWS misses the opportunity to explore the transitional steps towards such systems that could support the development of MHC as a practical framework.
In regard to concepts of ethics, control and agency, many of War Without Oversight’s assumptions go untested despite diverging interpretations continuing to frustrate international and interdisciplinary consensus. To illustrate, at one point the author claims that AWS ‘are the manifestation of handing over the decision to kill to a computer’.[8] This claim assumes that machines enabled by probabilistic algorithms actually make decisions when performing actions. This assumption diverges from the practitioners’ view that autonomous systems merely implement the commander’s decision to deploy the system, according to the range of potential behaviours and outcomes which the Delivery Cohort accepts and has oversight of through development and training. The presentation of such an assumption as ‘a given’ weakens the text’s arguments and cannot be readily contested by readers who are not engaged in the AWS debate.
The book further reflects a lack of consensus within prominent AWS-focused ‘communities of practice’, inheriting their conceptual limitations, interdisciplinary challenges and ambiguous terminology. The author defines AWS in several ways, so it is sometimes difficult to determine whether arguments are relevant to AWS, narrow autonomy in weapons systems, or both. Likewise, the terms ‘supervision’ and ‘oversight’ are often used interchangeably, with neither being defined explicitly. ‘Control’—the essential function of MHC—is not defined until the book’s penultimate chapter. This omission limits the capacity of the text to explore how to operationalise MHC and uplift the Delivery Cohort’s ability to address AWS challenges throughout the lifecycle, rather than as an afterthought for technologically flawed or limited systems.
This review’s key criticism of War Without Oversight is that it does not explore how to practically operationalise MHC of AWS, or autonomy in weapons systems more broadly. Walker calls for the addition of a statutory requirement for MHC, and makes various recommendations for the principles of MHC as a framework. These recommendations include the need to address control requirements on a case-by-case basis; generate capacity for timely intervention; and apply MHC principles before, during and after use of force. However, these principles are not accompanied by an exploration of how they might be translated into practice. Nor does the text include the legal doctrinal analysis required to show that existing IHL principles are insufficient to regulate AWS. The book’s penultimate chapter raises further questions about how to establish such a requirement for MHC. In what legislation or treaty should it be enshrined? How will states navigate differing concepts of human control, and distinct motivations for pursuing weapon autonomy? To what extent should MHC frameworks apply to extant uncrewed systems? How and by what metrics would adherence to MHC be measured, and by who? What confidence would signatories have that other states will comply? The text does not provide practical guidance on approaching these challenges.
It is worth noting that, by aiming to comprehensively address the topic of AWS from a multidisciplinary perspective, the author set himself a near-impossible task. It is to his credit, then, that the book’s exploration of the drivers and limitations that affect military AWS is so thorough in its treatment of issues and perspectives. However, much like the conversations that the concept of AWS generates among practitioners and scholars alike, the text does not delineate a path to actualise and practically implement the principles it espouses. As military interest and investment in autonomy increase, the extent to which principles such as MHC will inform the future use of AWS will be determined by the ability of countries and international bodies to ensure and incentivise adherence to these principles during conflict. War Without Oversight provides a wealth of relevant information that will support further research and operationalisation of principles for the use of autonomy in weapons systems.
Endnotes
[1] Describing weapons systems that are enabled by end-to-end autonomy in the targeting cycle, and distinct from RAS with lower levels of autonomy.
[2] Paddy Walker, War Without Oversight: Why We Need Humans on the Battlefield (Howgate Publishing Limited, 2025), pp. 99–100.
[3] The text defines the term ‘Delivery Cohort’ as ‘a device to convey the parties involved in delivering the deployment of AWS and will include, inter alia, the following taskings: neurophysiologists to coordinate AWS networks; psychologists to coordinate learning and cognition; biologists for adaption strategies; engineers for control routines; logisticians; roboticists; electrical specialists; behaviorists; politicians, civil servants and diplomats; people working in non-governmental organisations; sociologists; lawyers; company directors; weaponists; military tacticians; manufacturers; professionals involved in miniaturisation, simulation, configuration, coding, power supply and modularity; specialists in sensors, in distributed and decentralised routines; ethicists; data scientists and data engineers; and specialists in tooling and calibration’. Walker, War Without Oversight, pp. 11–12.
[4] This group is formulating an international legal instrument in the area of AWS for High Contracting Parties to the Convention on Certain Conventional Weapons.
[5] Where states have committed to a call to action, blueprint for action and pathway to action related to the military use of AI, informed by a specialist global commission.
[6] Walker, War Without Oversight, p. 32.
[7] Ibid., p. 180.
[8] Ibid., p. 6.