This is an interview with Antoine Bousquet. Antoine Bousquet is an Associate Professor in Political Science at the Swedish Defence University. His research sits at the intersection of war and security, the history and philosophy of science and technology, and social and political theory in the digital age. He is the author of the influential The Scientific Way of Warfare: Order and Chaos on the Battlefields of Modernity (2022) (now in its second edition), and the award-winning The Eye of War: Military Perception from the Telescope to the Drone (2018). His work has explored subjects including Cold War computing, the revolution in military affairs, and the conceptualisation of war and technological change. He is currently writing a monograph on the intellectual reception of nuclear weapons, spanning the fields of science, strategy, philosophy, and political theory.
DIFFRACTIONS: In your article “Researching ‘Digital War’”, you suggest we may be on the cusp of an “analogue revival” in computing, with neuromorphic and analogue circuits offering advantages for AI and autonomous systems. How might this technical shift complicate or transform the research agenda for scholars studying contemporary warfare?
Antoine Bousquet: When I originally wrote that article in 2019, I was concerned with assessing how far the label “digital” could do useful conceptual work in delineating a putative phenomenon of “digital war”. I was then rather dissatisfied with the casual use of the term, notably when opposed to the supposedly “analogue” as a way of distinguishing ICT-mediated activity from social life conducted outside those systems. That framing struck me as analytically imprecise, as the history of computing shows.
For most of its history, the automation of arithmetic and logic operations took the form of analogue computing devices from the Ancient Greek Antikythera mechanism to electro-mechanical fire control and navigation systems. Analogue computers work with continuous signals, exploiting direct physical analogues between the machine’s processes and whatever it is simulating. Digital computers, by contrast, encode and manipulate information as discrete symbolic values, which means that any continuous phenomenon must be sampled and quantised at the point of input. What is often forgotten is that the decisive shift to the digital domain actually came quite late. Pioneered in the 1940s, electronic digital computing only became fully dominant in the 1970s, and until then analogue systems were generally superior for real-time control and simulation tasks.
The development of semiconductor technology tipped the balance, making digital computing dramatically more efficient and easier to scale, while simultaneously magnifying analogue computing’s characteristic weaknesses, namely its sensitivity to noise and the difficulty of programming. Crucially, however, no “digital” system is purely digital. All of them depend on a prior conversion from the analogue domain, be they differences in the continuous intensities of electrical current within the chip itself or the various physical phenomena that can serve as inputs to computation. These are therefore already hybrid analogue/digital systems, and where the boundary between the two falls is always the outcome of specific engineering decisions.
The renewed appeal of analogue approaches today comes down mainly to concerns over energy efficiency. The extraordinarily high energy consumption of modern AI systems has become a serious constraint to their future development and general adoption. The contrast with the human brain, which performs comparable feats of pattern recognition and motor control at a tiny fraction of the energy consumption, has spurred research in analogue approaches such as neuromorphic computing. Rather than the circuit performing discrete symbolic operations that represent the computation as in all digital systems – modern neural networks included – neuromorphic architectures use artificial neurons in which the physics of the circuit itself is the computation. These architectures are particularly well-suited to specific applications such as sensor processing and robotic motor control but do not presently match general-purpose digital systems across the board. Most plausibly, what we are seeing is less a straightforward analogue revival than a renegotiation of where the analogue/digital boundary sits within hybrid systems.
For scholars of contemporary warfare, the most salient implications are concentrated in the domain of autonomous weapons and robotics. It is precisely here that the energy-efficiency and latency advantages of analogue and neuromorphic computing matter most. These properties enable capable autonomous systems that are smaller, faster, and far less dependent on external infrastructure. The future of swarming, involving millions, if not billions, of autonomous microbots, will almost certainly hinge on this technical revision of computing.
From a wider perspective, it is important to underline that the terms and significance of the boundary between analogue and digital domains remain contested, particularly with systems operating at thermal and quantum limits. This is particularly the case with our present understanding of the brain – there is no consensus on whether it is more usefully modelled as analogue, digital, a specific kind of hybrid, or something else entirely. As such, the digital neural networks powering the machine learning revolution are only very loosely related to the workings of an actual brain. Modern neural networks, trained on vast datasets and deep reinforcement learning techniques, are unquestionably impressive, but whatever path to intelligence they represent is clearly a deviation from biological brains that are incomparably more efficient in terms of the energy they consume and the number of data points they require when learning.
DIFF: Moreover, you caution against a “drone fetishism” and the tendency to treat contemporary technologies as radically novel. What other contemporary military technologies do you believe suffer from similar scholarly over-attention, and conversely, which understudied technologies deserve more focus?
AB: Here again, I was reacting to the way in which the object of the drone had become ubiquitous in my corner of academia in the second half of the 2010s, generating endless papers and conference presentations with evidently diminishing intellectual returns. There was even talk at the time of a new field of “drone studies,” which struck me as quite silly. This is rather emblematic of the general phenomenon of academic fads in which a particular object of study, concept, or theory becomes disproportionately popular, with the inevitable accompanying boom-and-bust cycle. This generally benefits the entrepreneurial academics who get ahead of the curve and reap abundant citations, less so the swathe of PhD students following in the slipstream and having to compete on a crowded job market after interest in the subject inevitably wanes.
The herd focus on a particular military technology comes at the expense of other less fashionable but arguably no less important objects. In the study of military technology, there is a long-established pattern of emphasising weapon systems over less showy technical innovations whose role is just as significant to the constitution and exercise of military power, such as canned food, penicillin, blood banking, technical standardisation, or any other aspect of the much neglected but crucial area of military logistics. I would gladly trade ten forgettable studies of drones for one serious investigation of any of these objects.
Moreover, we should never lose sight of the fact that military power is the result of a system and an exceedingly complex one with innumerable dependencies at that. Focusing on any individual component is therefore necessarily arbitrary. An arbitrary starting point is not a problem per se so long as our study serves to draw the meshwork of dependencies and genealogical relationships, thereby providing us with wider insights into the military system. If, on the other hand, we treat the object as causally determinative, as the singular driver of military change, we end up with a very impoverished, flattened account. This is what I could see as a tendency within the drone literature, prompting my somewhat provocative characterisation of drone fetishism.
In particular, I saw a propensity to reify the drone as a singular technology to which we could attribute radical rupture or novelty regarding the asymmetry of risk on the battlefield, the circumvention of sovereignty and international law, or loss of democratic accountability. These trends are very real, but to take the drone as a primary causation rather than merely as a symptom of longer-standing developments that manifest just as much in missile strikes and special force operations would be a mystifying inversion. Furthermore, the drone is not a single object but rather a concatenation of various technologies – flight, remote viewing, global geolocation, precision-guided munitions – that gets instantiated in a wide variety of technical systems encompassing MQ-9 Reapers, commercial quadcopters, and loitering munitions.
In my second book The Eye of War (2018), I tried to provide a genealogy of military perception by tracing the historical development of its operative functions – those of sensing, imaging, and mapping – and the different apparatuses through which they were realised. From this perspective, the drone appears as a materialisation of this logistics of military perception that has to be contextualised accordingly and apprehended in the technical detail of all its variants. And indeed, as drones have proliferated and their uses multiplied beyond the narrow framework of the War on Terror that popularised them, we have seen a retreat of the broad-brush arguments in favour of more nuanced, situated assessments.
Today, we are at risk of falling into a similar pitfall with the study of military AI. To be clear, the problem here is not so much that the current wave in AI is getting too much attention – we are clearly in the presence of a revolutionary technology that is developing so fast we are struggling to keep up – but that once again there is a tendency to lapse into thinking about AI as a singular technology we can talk about in general terms and submit to sweeping critiques. First of all, we need to be attentive to the specific uses to which AI is being put across military activities as varied as command and control, intelligence processing, target selection, autonomous weapons, logistics and human resources. Second, I think we have to engage seriously with the inner technical workings of these systems – whether computer vision, natural language processing, or autonomous planning and control – if we want to understand what these systems actually do and what their limits and failure points are.
DIFF: In your text, “War as Becoming”, you contend the human is itself an “indeterminate project of becoming.” If war and the human co-evolve, at what point does the phenomenon you are studying cease to be “war” as we might recognise it (politically motivated, organised violence between conscious adversaries) and become something else entirely (e.g., ecological management, algorithmic optimisation, cosmic entropy)? Is the term “war” even useful for our post-human condition?
AB: The answer to the question hinges largely on how we conceptualise war. If we foreground a continuity between expressions of hostility between animal groups in the natural world and the historical phenomenon known to us then clearly war can in principle outlast humanity. And when we look at the trajectory of technological autonomy in war, it may be tempting to echo Harun Farocki’s prophecy that “war will continue on well after humanity has disappeared from the face of machines.” The problem with this capacious conception of war is that it implies a dubious naturalisation of the phenomenon that strips from it any political character and agency.
For this reason, I incline towards thinking of war as a human activity with an inherently political and ethical dimension. This does not have to commit us to the provincialism of a cosseted essentialism so long as we ground ourselves in an open-ended process of becoming-human whose origin and terminus remain fundamentally indeterminate. Agency, interiority, and mutual recognition are variable, contingent constructs built over our history, not uniform essential traits of our species. In this regard, war as a contest of force between adversarial groups is not only a phenomenon whose historical development has demonstrably made it a powerful driver of societal and technological complexity – as important as this dimension is – but it has also fostered an intersubjective, intrinsically ethical relation between adversaries through which a violent exercise otherwise tending towards slaughter and annihilation finds its self-imposed limits and restrictions.
Violence is as much a mode of intercourse between distinct human societies as discourse, trade, or alliances and one that is not antithetical to recognition and collective identification across the battle lines. One of the major accomplishments of human civilisation has been overcoming the solipsism of early human groups, who typically viewed one another as fundamentally alien, giving rise to a notion of a common humanity whose members share in basic entitlements. In culturally bounding the permissible uses and expressions of collective violence, war has been one of the key sites of this accretion.
It is instructive that, across wide historical and geographical expanses, warrior cultures have always emphasised the honourable conduct that should govern combat, frequently paying almost as much regard to the valour and dignity of one’s enemies as that of one’s comrades. For the warrior, fighting prowess is an expression of agency that intrinsically has a moral character, however uncertainly and inconsistently it is realised on the battlefield. The norms and values enshrined within warrior cultures have provided the basis for all the contemporary legal frameworks devised to regulate and contain military violence by reference to a secular universalised conception of humanity. At the bare minimum, what distinguishes war from genocide or naked extermination is the mutual recognition of belligerents as symmetrical participants each entitled to persist beyond the current pass of arms. To put it another way, enmity is not the most odious relationship we can have with each other and, as we well know, yesterday’s enemy can become tomorrow’s friend.
The peril of technological war is its propensity towards an administrative-technological dissolution of enmity into targeting. Warfare is reduced to a process of technical optimisation and application of instrumental reason in which the means come to supplant the ends. The meaningful agency of human participants is correspondingly evacuated, reduced to performing narrow sub-routines within a wider assemblage. When the use of force devolves to an abstract, automated process of targeting in which the adversary becomes a mere algorithmic data point, intersubjective mediation and any true political relationship between belligerents is foreclosed. From this standpoint, it is questionable how much of the organised violence we see today is already something other than war. One could evoke here the chilling phrase “mowing the lawn” employed by the Israeli military, for instance. The point is not to romanticise what has always been the wicked reality of war but rather to underline that what may lie beyond war could be even worse.
DIFF: In “The Persistent Appeal of Chaoplexic Warfare: Towards an Autonomous S(War)M Machine?” you cite John R. Allen and Amir Husain (2017) envisioning “hyperwar” where “human decision-making is almost entirely absent” and the decision-action cycle collapses “to fractions of a second.” You also note the claim that AI could be “deployed at scales sufficient to essentially enable an infinite supply of tactical, operational, and strategic decision-making.” In the 2026 Iran War, the U.S. “reportedly” deployed autonomous drone swarms that adapted laterally to threats without human intervention, with the explicit goal of overwhelming Iranian air defences through saturation.
Simultaneously, Iranian forces relied on “waves of drones and missiles” in saturation attacks designed to “strain advanced, high-value U.S. and Israeli assets”. Given the operational tempo and scale of these autonomous engagements – where both sides seek to collapse the OODA loop and saturate the battlespace – do you see the 2026 Iran War or even the Ukraine War as a possible instantiation or candidate for “hyperwar”? Or does the persistent human strategic control (however attenuated) and the continued relevance of political objectives mean that this conflict still falls short of the “purely tactical mode of operability” (Guha 2010) you associate with the fully autonomous s(war)m machine?
AB: Whether or not it qualifies as “hyperwar,” what we have certainly seen in Operation Epic Fury is an unprecedented intensity in aerial bombardment in which 15,000 targets were struck by America and Israel in the first 14 days. While pre-existing target libraries can account for the opening spike of the campaign – around 2,000 strikes – the sustained tempo that followed speaks of a new compression of the kill chain. Through the fusing of satellite imagery, drone feeds, radar data and signals intelligence under machine learning architectures like the Palantir Maven system, targets are automatically being generated at speed, along with recommended weapons and strike packages. AI-assisted systems can now generate targets in minutes, compared with a matter of days when intelligence analysis was conducted manually and planning went through multiple bureaucratic layers as in the 1991 Gulf War.
The main bottleneck today is at the level of human verification and approval, with many concerned that this is being rushed through with insufficient assessment by the momentum of the machine-accelerated process. Widely-circulated reports indicate that similar systems have been used by the Israeli Defence Force in Gaza, generating 100 targets a day when a mere 50 a year were produced under previous methods. Such a volume of output evidently cannot be verified by human analysts who end up having to trust algorithmic recommendations. While the cause seems to have been one of outdated human labelling rather than algorithmic error, the death of 120 children at the Minab school in Iran is indicative of the kind of disasters that are liable to occur at an accelerated tempo, with minimal vetting. One also wonders what kind of war we will get when, instead of the current asymmetric contests, we have two sides both operating at a machinic tempo with a corresponding pressure upon them to go full throttle.
We are also seeing how the compression of the kill chain is enabling a strategy of decapitation that has killed a swathe of Iranian officials, including the Ayatollah Khamenei and top military and intelligence leaders. Through the ability to strike mobile targets on the very latest intelligence, the Americans and Israelis have succeeded in executing targeted assassinations with an unprecedented efficiency. And yet, for all this tactical success, the US has very little to show for it at this time of speaking in April 2026. All its strategic objectives to topple the regime and end its nuclear ambitions have been frustrated and the Iranian takeover of the Hormuz Straits has placed the US in an objectively worse situation than before the war. The Iranian leadership has hunkered down and is unlikely to be dislodged by further air attack alone. This conflict is turning out to be a chastening lesson in the dangers of tactics and operational brio outpacing sound strategy. You have to say the United States has been here before, but somehow never learns this lesson.
DIFF: Your concept, borrowed from Reza Negarestani’s “hyper-camouflage,” is a powerful analysis of the interplay among individuals and groups to continuously evade the totalising gaze of modern militaries. As we look to 2026, we see new evasive strategies emerging – from “digital vanish” techniques that scrub one’s online presence to “algorithmic camouflage” designed to fool AI pattern recognition. How do these new forms of “anti-perception” alter the power dynamic between the watcher and the watched, and do they ultimately push conflict deeper into the civilian sphere, as you have previously suggested?
AB: I employed the idea of “hyper-camouflage” to name the ever more radical dissimulation that manifestly responds to the totalisation of military perception, the expanding ability to align perception and destruction, to see everything and destroy whatever we can see. Beyond the technical means available to evade perception, the more fundamental and unsettling response participates in the dissolution of the boundaries between military and civilian spheres, a voiding of the distinction between war and peace. In the context in which both Negarestani and I were originally writing, we were chiefly thinking of the War on Terror where the conventional military dominance of the West was clearly a spur to terrorist practices that mandate the deep infiltration of the enemy’s civilian society and indiscriminate targeting of its populations, in turn causing their state apparatuses to securitise their societies ever further.
Today, the focus has shifted to great-power contest that engenders a more symmetrical but no less corrosive version of the same logic. The infiltration of adversarial civilian infrastructure – through cyber attacks, influence campaigns, the penetration of supply chains, or the sabotage of critical systems – has become a primary register of competition between states that turns everything down to individual neurocognition into an attack surface.
Foregrounding this contemporary context is important in assessing the techniques of “anti-perception” that individuals might mobilise against state surveillance. To the extent that the civilian information infrastructure is being recruited as a crucial part of the sensory and cognitive apparatus of military competition, the evasion and corruption of that apparatus risks becoming simultaneously a civilian and military act, whether intended as such or not. As to how the power balance between watcher and watched is presently shifting, this is an empirical question that cannot be resolved in the abstract. What we can say is that the asymmetry of resources required is severe.
Confounding computer vision or pattern recognition requires a sophisticated understanding of the specific model architecture being evaded, its training data, its loss functions, its known failure modes. Much of this knowledge is proprietary and concentrated in a limited number of state and corporate actors. However, it is impossible at this present juncture of rapid developments in artificial intelligence to know how these will ultimately redistribute power. We have barely begun to contend with the implications of an agentic age in which billions of autonomous agents will be embedded in every layer of social and institutional life. We do not know how centralising and decentralising tendencies will play out with AI, whether individuals will be able to independently run powerful open-source models or whether the concentration of compute resources will confer decisive advantages to the most powerful actors and institutions. The outcome of these conflicting tendencies will have a major bearing on the sociotechnical orders we come to inhabit in the coming decades.
DIFF: You note that the logistics of military perception underpinning the use of armed force rely on an entire sociotechnical assembly (sensing, imaging, mapping). Where does data poisoning – introducing corrupted or mislabeled data into the training sets that underpin these models – fit into a praxis of counter-conduct? If the martial gaze is “trained” on a specific corpus of labeled imagery, does the act of polluting that corpus (for instance, flooding the internet with images designed to break future AI training) constitute a new form of collective “becoming-imperceptible”? And how does that differ politically from the individual act of wearing a stealth hoodie?
AB: All the forms of counter-conduct designed to evade, disrupt, confuse, or distract the martial gaze fall under the category I refer to as “hiding.” Within that category, we can distinguish various exercises in concealment (the blending of a target signature into its surroundings, such as through camouflage or electromagnetic stealth), decoy (the counterfeiting of a target signature to purposefully draw the gaze), and bedazzlement (where enemy perception is variously blinded or overwhelmed to prevent any discrimination). All of these operate as anticipations of perception, leveraging what is known about its sociotechnical logistics to exploit its vulnerabilities and blind spots. The example of the stealth hoodie or dazzle make-up and hairstyling devised by the artist Adam Harvey illustrates that point well, with their designs evolved through both a general understanding of infrared cameras and facial recognition algorithms and a process of empirical validation.
Data poisoning that seeks to corrupt the training data used to develop AI systems so as to compromise their integrity, reliability, or accuracy fits into the general praxis of counter-conduct and perceptual anticipation outlined above. This practice does, however, reflect the emergence of a new attack surface resulting from new layers of automation within the logistics of perception performing complex modes of pattern recognition to identify target signatures. As we have gathered from reports on the recent operation in Venezuela and Iran, LLMs have now been integrated into already established computer vision and pattern-of-life algorithms so as to provide analysis of text-based intelligence. Corruption or manipulation of any of these layers could give rise to any of the practices of concealment, decoy, and bedazzlement, particularly if they install backdoors that can be triggered by transmitting a predetermined signal.
At this point, the technical feasibility of such data poisoning is well established even if there is as of yet no public evidence of state-sponsored data poisoning of a specific defence AI system. Absence of evidence is not evidence of absence, however, and a well-constructed backdoor within a black-box neural network may well be undetectable by conventional assessment. Commercial companies like Anthropic and OpenAI, who are now providing their models to the military and are far less secure than national security organisations like the NSA, have to be considered especially vulnerable targets for state infiltration. The means available to non-state actors – such as poisoning the vast repository of the internet – are much less likely to be effective, with curated private datasets being increasingly privileged. More broadly, the question of power in the twenty-first century will likely not come down solely to who builds and operates the most advanced AI systems but to who can control and shape the data environments that will support these systems.
DIFF: Looking ahead, what is the next horizon for your research? For instance, do you see your future work continuing to explore the specific sociotechnical regimes you’ve been analysing – perhaps extending into the domains of AI, quantum computing, or climate change as new forces shaping war and political thought – or do you envision a more fundamental shift in your approach or subject matter?
AB: I am in the process of completing a monograph provisionally entitled Shadow of the Bomb that seeks to recover the intellectual crisis that the advent of nuclear weapons provoked among scientists, strategists, philosophers, and political theorists across the first decades of the Cold War. This project largely brackets out the sociotechnical aspect to focus on the Bomb as an event in thought, one which unsettled Enlightenment ideas about the role of reason in human history as refracted through various intellectual fields. My central contention for the project is that this intellectual crisis was never conclusively resolved during the Cold War and merely forgotten after its conclusion and the fear of nuclear war receded. Now that nuclear anxiety is again resurgent, it is vital that we return to these original intellectual encounters, understand the impasses they led to, and perhaps trace a way out of them.
I have already published one article on “nuclear existentialism” in which I examine the post-war philosophical reception of nuclear weapons. Drawing on the legacy of Nietzsche and Heidegger, thinkers like Albert Camus, Jean-Paul Sartre, Karl Jaspers, and Günther Anders grapple with the nihilistic implications of apocalyptic weaponry that could realise a “being-towards-species-death” that would negate any completion of human history. Paradoxically, they ended up conceiving the advent of the Bomb as the most meaningful event in human history, the occasion on which humanity would either perish or ascend to a higher state of consciousness. Yet their hopes that the fear of the Bomb would provoke the necessary existential awakening have clearly not come to pass. And so the problematique they confronted still remains very much our own.
While this project has occupied most of my attention for the past few years, I have been following closely the developments in artificial intelligence since the public launch of ChatGPT in 2022, and I have several forthcoming publications on the military and geopolitical implications of AI. This will be the focus of my research for the foreseeable future with the sociotechnical dimension returning as a major strand. I don’t think the historical genealogy I usually incline towards will be a prominent feature, however. Things are moving so rapidly in this field right now that staying on top of current developments and making sense of them seems already a daunting task in and of itself.
I am also very interested in exploring the ways in which AI models and agentic systems can be leveraged usefully for research purposes. I understand the apprehension that many colleagues have about the impact of AI on academia and the university writ large; the challenges coming at us are enormous and our institutions are slow-moving at the best of times. Yet it is also clear to me that this technology is here to stay and will unquestionably overhaul knowledge production in ways we can only barely conceive of today. And for all the pitfalls of deskilling and cognitive offloading we must absolutely be mindful of, there are also some tremendously exciting opportunities. My time recently spent interacting with coding agents has been particularly eye-opening. Anyone can now spin up a computer programme without knowing how to read a single line of code, a radical democratisation of software that is surely to be welcomed. It is ultimately up to us to seize this moment for ourselves – or, as Deleuze put it in his “Postscript on the Societies of Control”, “there is no need to fear or hope, but only to look for new weapons.”
REFERENCES
Allen, J. R., & Husain, A. (2017). On hyperwar. Proceedings of the U.S. Naval Institute, 143(7). https://www.usni.org/magazines/proceedings/2017/july/hyperwar
Bousquet, A. (2018). The eye of war: Military perception from the telescope to the drone. University of Minnesota Press.
Bousquet, A. (2020). Researching “Digital War”: Terminological snares, conceptual pitfalls, and methodological hazards. Digital War, 1 (1), 159–163. https://doi.org/10.1057/s42984-020-00018-7.
Bousquet, A. J. (2022). The scientific way of warfare: Order and chaos on the battlefields of modernity (2nd ed.). Hurst & Company.
Bousquet, A. (2023). The persistent appeal of chaoplexic warfare: Towards an autonomous s(war)m machine? In A. Gruszczak & S. Kaempf (Eds.), Routledge handbook of the future of warfare (pp. 97–108). Routledge. https://doi.org/10.4324/9781003299011
Bousquet, A. (2024). Nuclear existentialism: On the philosophical response to life and death under the bomb. Review of International Studies, 1–19. doi:10.1017/S0260210524000512
Bousquet, A. (2024). Becoming (Im)Perceptible: From Scopic Regimes to the Martial Gaze. Drone Aesthetics: War, Culture, Ecology.
Bousquet, A., Grove, J., & Shah, N.N. (2020). Becoming war: Towards a martial empiricism. Security Dialogue, 51, 118–199.
Deleuze, G. (1992). Postscript on the Societies of Control. October, 59, 3–7. http://www.jstor.org/stable/778828
Farocki, H. (2004). Phantom Images. Public, (29). Retrieved from https://public.journals.yorku.ca/index.php/public/article/view/30354
Guha, M. (2010). Reimagining war in the 21st century: From Clausewitz to network-centric warfare. Routledge.
Negarestani, R. (2008). Cyclonopedia: Complicity with anonymous materials. re.press.