In this interview, Dr Paddy Walker, a leading expert on autonomous weapons and military ethics, speaks with Dr Iain Overton, Executive Director of Action on Armed Violence (AOAV), about his latest book, War Without Oversight: Why We Need Humans on the Battlefield.
Dr Walker, an Associate Fellow at the Royal United Services Institute (RUSI) and Managing Director of the Leon Group, has spent years researching the challenges posed by autonomous weapons. A former military officer, he holds a PhD from the University of Buckingham’s School of Humanities, where his research focused on the legal, ethical, and technical dilemmas surrounding the deployment of autonomous weapon systems. His latest book, published by Howgate, builds on his previous work co-authored with Professor Peter Roberts, War’s Changed Landscape? A Primer on Conflict’s Forms and Norms.
With a background that spans academia, policy, and the private sector, Dr Walker is at the forefront of discussions on how emerging technologies are reshaping conflict. As a former co-chair of the London Committee of Human Rights Watch, he has long advocated for greater oversight and accountability in modern warfare.
As autonomous warfare shifts from speculation to operational reality, this interview provides crucial insights into the political, economic, and technological roadblocks that could shape – or stall – the future of AI-driven conflict.
Dr Iain Overton: Your book opens with the first recorded robot-only military manoeuvre in Ukraine. How do you see this moment as defining in terms of future conflicts, and do you think it represents a point of no return for autonomous warfare?
Dr Paddy Walker: I think it is a critical moment, which is why I open with it. That said, I would hesitate to describe it as a complete revolution in military affairs. There is a tendency to declare the advent of any new technology as the defining shift in the way wars are fought, but the reality is often more complex. While the deployment of unmanned systems in Ukraine represents an escalation in the use of AI-driven warfare, it does not mean we have reached a point where human oversight has been entirely removed from combat decision-making.
One of the key distinctions I make in the book is the need to understand the context in which these systems are being deployed. It’s easy to look at these developments and assume they indicate a total transition to autonomous warfare but autonomy in these cases is limited. December’s robot-only development was still curated and overseen by humans. This was certainly not autonomy in action. What is crucial, and what would mark a real tipping point, is the extent to which these systems are being entrusted with target acquisition and engagement decisions. If a system is merely identifying potential targets and passing that information backwards and forwards to a human operator – which very largely remains the case in Ukraine – that’s one thing. If it is selecting and engaging targets independently, without human oversight, that’s something else entirely. And not what we are yet witnessing on the world’s battlefields.
There is therefore an important discussion to be had about whether we are truly at a ‘point of no return.’ As things stand, even in Ukraine, these weapons are still operating with significant constraints, and all of the practical, legal, and ethical barriers to full autonomy remain unanswered. Indeed, the feasibility of completely removing human oversight from lethal decision-making remains an open question, and one that becomes the key research question of my book.
Fundamentally, what we are seeing is not a straight-line march toward full autonomy, but rather an evolution and one that certainly raises pressing concerns about where this technology is heading and what safeguards need to be put in place before we cross a threshold from which we cannot return.
Dr Iain Overton: You argue that autonomous weapon systems challenge fundamental military principles. What do you see as the most overlooked but crucial challenge – technical, ethical, or legal?
Dr Paddy Walker: Each of these challenges – technical, ethical, and legal – presents significant hurdles, but I would argue that the technical challenges are the most overlooked and, paradoxically, the most foundational. Half of my book is dedicated to exploring the technical difficulties of developing fully autonomous weapon systems, because so much of the public debate assumes their viability without critically examining their feasibility.
To properly address this, we need to make a few assumptions. First, that your questions relates to the deployment of fully autonomous weapons – systems capable of making independent decisions on target engagement without human intervention. Second, that these systems can reliably be fielded with a wide range of capabilities, able to operate in complex, dynamic battlefield environments. And third, that they will adhere to the laws of armed conflict, including proportionality, distinction, and military necessity as widely understood and signed up to by parties to the Geneva Convention.
The biggest technical challenge is the issue of data. Autonomous systems rely on machine-learning algorithms that process vast amounts of data to make decisions. However, battlefield data is fundamentally different from the controlled datasets used in laboratory settings. Combat environments are unpredictable, filled with contradictory information, sensory overload, and rapidly shifting conditions. Unlike humans, who have evolved to process and adapt to uncertainty, machines rely on structured learning. If the data they receive is flawed, obsolete, biased, or incomplete, the decisions they make must certainly be compromised.
One of the first problems that make up this argument therefore revolves around data curation – how do we ensure that an autonomous weapon is processing reliable, accurate information? What happens when a machine encounters an unfamiliar scenario? How do we prevent systems from making catastrophic mistakes due to misidentification? These are not trivial questions. In human decision-making, we rely on experience, judgment, and moral reasoning. These weapons will operate in isolation and where very consequential decisions must be made based solely on algorithms, which are unfortunately only as good as the data they have been trained upon.
Another concern then becomes the issue of adversarial manipulation. We already know that machine learning models can be tricked with minimal data modifications – change a few pixels in an image, and a sophisticated AI can mistake a stop sign for a speed limit sign. In a battlefield scenario, adversaries could exploit these weaknesses to manipulate targeting algorithms. If an AI-controlled weapon system misidentifies a civilian vehicle as a military threat due to a small, deliberate alteration, the consequences will likely be catastrophic.
These weapons’ legal and ethical challenges, while significant, are ultimately determined by this shaky technical foundation. If a machine cannot be trusted to reliably differentiate between combatants and civilians, how can it be legally and ethically deployed? The idea of ‘meaningful human control’ is only meaningful if humans can understand and predict the system’s behaviour. Without addressing fundamental technical limitations, these weapons’ entire justification is quickly eroded.
Dr Iain Overton: You discuss meaningful human control in your book as a statutory requirement for lethal engagements. If this principle were to become a battlefield necessity, how realistic do you think it would be to enforce – particularly in an era where military efficiency often prioritises speed over oversight?
Dr Paddy Walker: To provide some context, the discussion around meaningful human control has been evolving for over a decade. Back in 2012, Human Rights Watch released its Killer Robots report, which became one of the first major public-facing warnings about the potential risks of autonomous weapons. It framed the issue starkly, raising questions about how to regulate this technology without inappropriately restricting military capability.
But I do agree that the phrase meaningful human control itself is problematic. What exactly is meaningful? What constitutes control? And must the human always be in the loop? Even in its current discussions within the Convention on Certain Conventional Weapons (CCW) at the United Nations, it has taken years just to agree on basic definitions. With 194 signatories to the Geneva Conventions participating, debates over semantics have often led to obfuscation rather than clarity.
But if we break it down, one of the most crucial aspects of meaningful human control is independent targeting. For me, this is the red line in losing human supervision in lethal engagement. A quick refresh might be useful. Autonomous systems must make decisions based on data processing, which requires them to abstract real-world objects into simplified, machine-readable forms. The system’s sensors must then match exactly these abstractions to the data training sets upon which they have been programmed in order to determine whether a target qualifies for legally-compliant engagement.
This process introduces multiple layers of potential failure – whether through biased collection or interpretation of data, misidentifications within that data, or the inability of that data subsequently to account for real-world complexities. This is why I argue that meaningful human control might be based upon a statutory framework that is anchored on target profiles.
Weapons must also adhere to legal and ethical principles. At their most basic, these are enshrined in those same Geneva Conventions, and comprise proportionality, distinction, military necessity, and the Martens Clause.(Ed.: “Until a more complete code of the laws of war has been issued, the High Contracting Parties deem it expedient to declare that, in cases not included in the Regulations adopted by them, the inhabitants and the belligerents remain under the protection and the rule of the law of nations, as they result from the usages established among civilised peoples, from the laws of humanity and the dictates of public conscience.”)
So let’s look in more detail at targeting profiles, those conditions that are programmed into these weapons and that must be matched before the weapon unilaterally initiates violence. Tying a statutory framework to such profiles might use the following parameters. First and foremost, those responsible for initiating violence must be able to anticipate those weapon’s effects. Operators should be able to predict the system’s lethal outcomes at all stages of the deployment cycle. Each target’s profile must also be fixed and immutable – autonomous systems should not be allowed to dynamically change their targeting parameters based on received data or the evolving battlefield. Weapons should similarly not target humans directly. Race, gender, age, or other demographic factors can not be used as any part of a weapon’s targeting criteria. The weapons’ operations must be time-limited; autonomous weapons cannot be allowed to engage indefinitely. Their physical area of operation must also be controlled and understood by those responsible for their use. Unsupervised weapons must also not be able to engage in sequential targeting – an autonomous system should not be permitted to move from one engagement to another without fresh human authorisation.
In short, if meaningful human control is to be more than just a buzzword, it needs to be based upon measurable operational constraints. The question is then not just whether this can be enforced, but how authorities define, regulate and police it in a way that prevents unintended escalations while maintaining strategic effectiveness.
Dr Iain Overton: Do you see any historical parallels or lessons learned from past attempts to regulate disruptive military technologies that can inform how we approach the regulation of autonomous weapons today?
Dr Paddy Walker: There is always a sense of presentism in these types of arguments – the belief that we are experiencing something entirely unprecedented or that change is happening at an accelerated rate like never before. However, history shows us that attempts to regulate military technology are not new. The Geneva Conventions and the Hague Conventions, dating back to 1899, have prohibited – or tried to prohibit – several weapons due to their indiscriminate or inhumane nature. Dum-dum bullets, chemical gases, and, more recently, blinding lasers have all been banned under international law. One particularly interesting legal precedent is the Martens Clause which asserts that, even in the absence of specific prohibitions, combatants’ actions should adhere to the ‘dictates of human conscience’ and the principles of humanity. However, the success of any such regulations depends not just on treaties but also on establishing and supporting strong international norms around what is and what is not acceptable on the battlefield.
But the track record for weapons regulation is mixed – attempts to ban landmines and cluster munitions depressingly have hardly affected their use. The delta here usually lies between political will and enforceability. Indeed, the challenge today with regulating autonomous weapons is that their development is being driven not just by national states but also by the technical advances of the world’s private corporations. There is a portfolio of reasons why autonomy is attractive, not least the asymmetric advantage of these platforms for those deploying these means including, of course, emerging non-state actors.
More promising than outright prohibition, perhaps, might be the role of civil society in shaping the debate. Transparency, accountability, and consistent scrutiny from NGOs, investigative journalists, and advocacy groups are vital in ensuring that public awareness and ethical considerations play a role in limiting the deployment of these technologies. The pressure exerted by these groups can often be just as effective as legal frameworks in influencing state behaviour.
Dr Iain Overton: That leads us to the tension you explore in your book between autonomy and accountability in warfare. How do you see accountability functioning in a world where AI-driven weapons may make their own decisions to kill?
Dr Paddy Walker: That is the paradox, isn’t it? The assumption is that autonomous weapons will reduce human involvement in warfare, yet these systems require an extraordinary number of human actors just to function effectively. We see this already with drone warfare – while the pilot may be located thousands of miles away, there is an entire network of intelligence analysts, engineers, and legal advisors that are required to enable those operations.
In my book, I introduce the concept of the ‘Delivery Cohort’- a term that encapsulates the vast web of individuals responsible for developing, deploying, and overseeing autonomous systems. This includes policymakers who approve the use of such systems, engineers and data scientists who design them, legal teams who establish rules of engagement, and logistics personnel who maintain them. Despite the perception that AI will make warfare more efficient, the reality is that it introduces additional layers of complexity. Human oversight is not going away any time soon; the means of configuring and maintaining unsupervised weapons is an area where states’ procurement has hardly touched the surface.
To your question, this complexity also complicates accountability. If an autonomous weapon commits an atrocity, who is responsible? The programmer who wrote the algorithm? The commander who deployed the system? The politician who sanctioned its use? Unlike conventional weapons, where responsibility is usually quite clearly delineated, AI-driven warfare blurs these lines.
Additionally, building autonomous systems that can adhere to legal and ethical standards presents significant challenges. That the idea looks attractive in a PowerPoint presentation is a million miles away from fielding compliant, useful means on the battlefield. It is this whole notion of a ‘Revolution in Expectation’. We’re seen these platforms in films and therefore they can be built. This is risible. Take countries’ Rules of engagement; for example, Norway recently updated its own framework in what has become a 1,600-page document. How do you encode this to command a weapon’s notion of what it can and cannot do? How can such nuanced legalese be reflected in an AI system? How do you ensure that a machine interprets ambiguous battlefield scenarios correctly? These are the questions that make accountability in autonomous warfare exceptionally difficult to ensure.
Ultimately, while military leaders may push for greater autonomy in weapon systems, the need for human oversight remains essential, all driven by a mix of practical and behavioural reasons, for technical and legal reasons. The challenge is not just in preventing AI from making lethal mistakes but in ensuring that the entire chain of human actors remains accountable in the deployment of weapons without human oversight or veto.
Dr Iain Overton: You highlight technological progress as often chaotic, non-linear, and not necessarily inevitable. Do you see any major roadblocks – technological, political, or economic – that could slow down or even prevent the widespread deployment of autonomous weapons systems?
Dr Paddy Walker: Absolutely. If we start with the political factors, democratic nations, in particular, are highly susceptible to public opinion and media scrutiny. Imagine the fallout if an autonomous system malfunctions and causes unintended civilian casualties – the public outcry would be enormous. We’ve seen before how one catastrophic event, amplified by the press, can change government policy overnight. Just as the enemy has a say in how wars are fought, so too does society. A strong enough backlash could force policymakers to rethink autonomous weapons programs.
From an economic standpoint, there’s a misconception that autonomous weapons will always provide cost-savings. While drone technology has undoubtedly become cheaper – first-person view drones are available for as little as $300 – truly autonomous weapons require extensive R&D, far-reaching integration into existing military systems, and long-term logistical support. The transition to autonomy isn’t just a matter of acquiring the technology; it necessitates a wholesale shift in military doctrine, training, and infrastructure. That’s an expensive and complex transformation – the more so for an untested military asset whose advantage may be unexpectedly short-lived – and nations may be reluctant to invest so heavily in unproven systems where technical capability is moving so quickly.
The technical challenges, however, are perhaps the most underestimated. Machine learning algorithms depend on data, but battlefield conditions are unpredictable and highly adversarial. Unlike the controlled testing environments of a university workbench, war zones involve sensory overload, a sea of contradictory information, all undercut by the intentional deception of opponents. AI struggles with incomplete, contradictory and duplicatory inputs. Very small manipulations of a dataset can completely alter an algorithm’s outputs. For instance, a weapon system that cannot reliably distinguish between combatants and civilians is an unacceptable proposition.
Then there’s the issue of what is termed ‘local minima’ – the tendency for AI systems to settle into suboptimal decision-making patterns without the ability to correct themselves. Unlike human soldiers, who learn from experience and adapt fluidly, autonomous systems can become stuck in rigid cycles that fail under novel conditions. In a battlefield scenario, this would clearly be unacceptable from any parties’ standpoint.
Additionally, AI systems need to constantly update their understanding of each machine’s immediate environment, known as its ‘initial representation’ which is programmed at deployment and set – manually, I should add – under each weapon’s initial configuration. When that weapon is deployed, its sensors gather real-time data in order, quite reasonably, to have that asset adapt to changing conditions. But it is the degree of change to that representation that is problematic. To what degree should the Delivery Cohort’s initial programming be overwritten to account for new data? How are these machines to be anchored? Indeed, how much can an agent’s ‘perception’ shift before it is no longer fit for purpose, no longer reliable or predictable? The upshot here is that the system is very likely to misinterpret its surroundings due to that new data, then making flawed decisions with catastrophic consequences. Unlike humans, who instinctively recalibrate their judgment, AI has no innate ‘common sense’ to fall back on.
Another major concern is error correction, the more so in the likely communications-denied environment that will characterise autonomous weapons. Once such systems are deployed, they must of course operate without human oversight. But malfunctions behind enemy lines or in conditions of electronic warfare must still be diagnosed and fixed remotely. Unlike conventional weapons, which can be recalled or repaired by technicians, an autonomous system may be left to function – possibly with faulty decision-making – until it exhausts its power reserves and mandate. Indeed, how even to patch an AI-driven weapon in real-time is an enduringly challenging issue.
Historical precedent also demonstrates that military technology rarely progresses in a straight line. Be it the F-35 fighter jet or multiple NASA space programs, working to the edge of technical know-how usually incurs material delays and cost overruns, not least because of difficulties integrating complex systems. Autonomous weapons will face similar hurdles, particularly when states realise that these technologies demand more human oversight – not less – in order to function predictably.
So, while the push for autonomous warfare is strong, numerous roadblocks – whether technological limitations, economic constraints, or political backlash – could slow its adoption significantly. The notion that we are inevitably heading toward fully autonomous war is, in my view, premature.
Dr Iain Overton: You mentioned the military-industrial complex, procurement timelines, and dual-use technologies as major drivers of ‘killer robot’ proliferation. How do you see corporate and commercial interests shaping the future of warfare?
Dr Paddy Walker: There has been a profound shift in how military technology is developed and acquired. In the past, governments would issue specifications, and domestic arms manufacturers would build weapons to order. Now, the dynamic has changed entirely. Instead of a ‘pull’ system where governments dictate requirements, procurement is instead beholden to a ‘push’ system, where commercial companies are first developing cutting-edge and usually dual-use technologies before these are then marketed to the military.
A telling example has been in the development of AI-assisted targeting and decision aids, an example being their deployment by the Israeli Defense Forces. Investigations such as that researched by +972 Magazine have revealed systems like ‘Gospel’ and ‘Lavender’- AI-driven programs used to generate lists of potential targets. In past conflicts, the IDF might have identified around 5,000 targets during an entire military operation. Now, with AI-assisted analytics, they are reportedly able to identify nearly forty thousand targets in a month. And where did this capability come from? Not from traditional military R&D pipelines, but from the commercial sector. Private technology companies operating in this space have been pivotal in advancing these capabilities. The pace and depth of smartphone technology provides another example to confirm this trend with R&D investments made in consumer electronics – batteries, high-resolution cameras, LiDAR navigation, and more – all feeding into military applications.
This shift creates significant challenges. Unlike government-controlled military procurement, the private sector is set up to generate surpluses. With out capital constraint and able to recruit excellent human talent, they often move at a much faster pace. Driven by commercial success, their technologies unsurprisingly outstrips regulatory oversight before governments have had time to respond to each new development. Furthermore, tech companies have few restrictions to selling on such AI-driven capabilities. This clearly raises ethical and security concerns. If commercial firms, rather than national governments, become the primary drivers of military AI, who is responsible for ensuring that these technologies are deployed responsibly? Who sets the ethical boundaries? And who holds private actors accountable when autonomous systems are exported to authoritarian regimes or non-state actors?
As we move forward, corporate interests will continue to shape the landscape of warfare. The question is whether regulation and oversight mechanisms can keep pace with the commercialisation of military AI before their spread and use is endemic.
Dr Iain Overton: Some proponents argue that autonomous weapons could lead to what might be described as ‘cleaner’ warfare, with fewer civilian casualties. Given your research, do you see any merit in this claim, or is it just another myth of technological utopianism?
Dr Paddy Walker: This argument is often made by proponents of these systems, but I remain sceptical. The idea of ‘cleaner’ warfare assumes that machines will be more precise than humans, less prone to emotional decision-making, and therefore better able to distinguish between legitimate military targets and civilians. The theory here is that a perfectly calibrated AI system with flawless data inputs could, in some cases, reduce collateral damage. However, the reality is far messier.
A justification put forward for this position is the notion of an ‘ethical governor’ – a proposed AI system that acts as an arbitrator and restraint on all of the decision-making around initiating lethality, ensuring compliance with international humanitarian law before allowing a weapon to engage. The computer scientist Ronald Arkin, for instance, was one of the first academics to suggest that such a system could make AI-driven weapons more ethical even than human soldiers, who are empirically prone to panic, to hold biases, or have their actions determined by heuristics such as revenge. Again, the concept might look promising on paper but, as soon as real-world complexity is introduced, the idea quickly breaks down. First, autonomous weapons rely on data-driven pattern recognition. Unlike a human, who can assess context with flexibility and intuition, AI systems operate on rigid rules and training sets. Any flaws in the sensed data – biases, missing variables, duplicatory or obsolete feedback that jars with the agent’s training data, adversarial inputs and suchlike – will likely lead to catastrophic miscalculation. An AI that misidentifies a surrendering soldier as an active combatant will trigger unlawful lethality with no human in the loop to mediate that decision.
Another issue is system failure. What happens if an autonomous weapon is damaged mid-operation? Does it continue engaging targets, even if its sensors are compromised? AI by definition lacks the self-preservation instinct of a human soldier – while this might lead to appropriately conservative behaviour, it also means the system may continue a mission regardless of sensed data to the contrary. If it fails, does it fail in a ‘safe’ way – by ceasing operations – or does it fail dangerously, continuing to execute its programmed objectives despite battlefield conditions that point otherwise? These are not abstract concerns; they are real risks that have not been adequately addressed.
We also must consider the strategic implications. If one side deploys autonomous weapons under the assumption that they will be ‘cleaner,’ does that create an incentive to use them more frequently? The barrier to engaging in conflict could be lowered if political leaders believe they are engaging in a form of ‘surgical’ warfare, free of human error. But as history has shown, every intended deployment of precision warfare – from drone strikes to smart munitions – has occasioned civilian casualties and unintended consequences. Evidence suggests that any assumption that it will be different this time with AI is clearly misguided. Ultimately, the claim that autonomous weapons will reduce civilian harm is based on a best-case scenario that wilfully ignores technical realities as well as the historical precedents of what has occurred in the recent past.
Dr Iain Overton: That leads to a broader question about unintended consequences. Warfare has always been shaped by human instincts, emotions, and errors. If we remove or significantly reduce human agency from lethal decision-making, what unforeseen consequences might emerge that we haven’t fully considered?
Dr Paddy Walker: An immediate concern is how non-state actors might exploit this technology. Now, we tend to think about autonomous weapons in terms of state military forces, but the reality is that this technology is proliferating. If a commercially available drone can be modified to function as a rudimentary autonomous weapon, what’s stopping non-state groups from doing the same? The threshold for developing and deploying these systems is much lower than many assume.
Another concern is plausible deniability in these weapons’ use. Autonomous systems, by their nature, remove direct human decision-making from lethal actions. This opens the door for parties to use these weapons in ways that obscure accountability. If an autonomous drone strikes a civilian convoy, was it a programming error, an adversarial hack, or deliberate targeting? Whose drone was it that undertook this error? The answer may never be clear. Governments and military leaders will likely absolve themselves of responsibility by blaming ‘unpredictable’ AI behaviour, making it harder to attribute war crimes or hold decision-makers accountable.
There’s also the risk that reducing human involvement in warfare could lower the barriers to conflict itself. If leaders believe they can engage in warfare without significant human losses, they may be more willing to do so. We’ve already seen how the introduction of drone warfare has enabled states to conduct military operations in places where they might have previously hesitated. With autonomous weapons, that effect could be magnified. The idea of ‘risk-free’ war is an illusion, but the perception that AI can fight battles without human cost could make war seem like a more viable policy tool.
On a longer timescale, we must consider how incremental changes in military technology can lead to tipping points. Autonomous weapons are not likely to be introduced all at once; instead, they will be deployed incrementally and in limited capacities, gradually taking on more roles and uses as their Delivery Cohort grow in confidence. But at some point, the cumulative effect of those incremental changes could fundamentally alter how war is waged. Just as drones started as reconnaissance tools before evolving into frontline strike systems, the purpose of introducing machine learning into autonomous violence must be for the weapon to learn, to use examples of a known case to make decisions about a present case and, while starting with limited functions and limited goals, quickly take on lethal decision-making roles that were never originally intended.
The last point I’d make is about force multiplication. The nature of conflict is changing – what we’re seeing in Ukraine today is just the beginning. Right now, drone swarms are being used in attacks numbering in the hundreds, but soon, we could be looking at engagements involving thousands with lethal munitions camouflaged by hundreds of decoys to overwhelm an adversary’s air defence. If that happens, how do conventional military defences respond? How does deterrence function in a world where autonomous systems are mass-produced, rapidly deployed, and difficult to counter? These are strategic questions that have yet to be answered.
In short, removing human agency from war doesn’t just change how battles are fought; it changes how wars begin, how they escalate, and how they end. Those unintended, second order consequences are what worry me the most.
Dr Iain Overton: If you had the power to implement one non-negotiable principle for regulating autonomous weapon systems, what would it be and why?
Dr Paddy Walker: If I could enforce just one principle, it would be those strict limits on autonomous targeting profiles discussed above. No weapon system should be allowed to engage targets without direct, meaningful human oversight. The idea that an AI system can be programmed with ethical safeguards is, in my view, fundamentally flawed. Machines operate on probabilities, not judgment. That is a critical distinction, and one that makes their independent use of lethal force inherently dangerous where a compliant weapon must work first time and every time in a predictable, explainable fashion and in a manner that is based upon doctrine and the passing practices of the human soldier.
So, to answer your question, that set of core restrictions should immediately be put in place: autonomous weapons must not target humans; they must not engage in open-ended missions with self-generated objectives; they must not operate indefinitely without human review. Any engagement decision should require human validation before lethal force is used.
Beyond that, I would push for greater transparency in military AI development. Too much of this work is happening behind closed doors, enabled by private corporations and their research teams with little public accountability. If we don’t establish clear ethical and legal constraints now, we risk seeing precedents that will be impossible to reverse. The history of military technology shows that once a new capability is introduced, it rarely goes away. We need to act before these systems become an entrenched part of modern warfare.
Dr Iain Overton: So, we’re not necessarily heading for an apocalyptic wasteland of autonomous warfare?
Dr Paddy Walker: Not necessarily, but I do think we’re on the verge of some fundamental changes. The most dangerous assumption we can make is that these systems will work as advertised. Autonomy in warfare is far more difficult to achieve than many realise, and even if it is achieved, the consequences may not be what people expect. It’s not just about whether AI can function in war – it’s about whether we can control what comes next.
Dr Iain Overton: Paddy, this has been an incredibly insightful conversation. I hope this serves as a compelling reason for AOAV’s readers to pick up your book.
Dr Paddy Walker: I appreciate that, Iain. Thank you – and yes, it’s available now via Howgate Publishing in both paperback and hardback. Many thanks.
Did you find this story interesting? Please support AOAV's work and donate.
Donate