Abstract
For the past two decades, the camouflage patterns used on military uniforms have been computed by families of algorithms. The article argues that these computation methods fit within a genealogy of discourses on camouflage practices, which the text reconstructs along the axes of flatness and depth. Camouflage can be explained as an attempt at flattening information related to a target onto the underlying background, by producing general invariances within the environment and in accordance with the observer’s perceptive faculties. Carrying out this task, static camouflage responds to the aporia of using one pattern to disguise presence within multiple contexts. After analysing two discourses on camouflage in relation to the notion of flatness, the article considers a case of algorithmically generated static camouflage. Its disguising method abstracts information from both the environment in which an army is expected to fight and the observer’s perceptive capacities and then computes these threads of abstraction together in a pattern. The meshing function of computation thus flattens the principles embedded in the fabric, making them individually illegible. This changes the value that camouflage usually gives to target/background relations and to the observer’s perceptive faculties.
In general, generalization is to lie, to tell lies. (B.S. Johnson, The Unfortunates)
An unspeakable horror seized me. There was a darkness; then a dizzy, sickening sensation of sight that was not like seeing; I saw a Line that was no Line; Space that was not Space; I was myself, and not myself. When I could find voice, I shrieked aloud in agony, ‘Either this is madness or it is Hell.’ ‘It is neither’, calmly replied the voice of the Sphere, ‘it is Knowledge; it is Three Dimensions: open your eye once again and try to look steadily.’ I looked, and, behold, a new world! (Edwin A. Abbott, Flatland)
Introduction
In 2009, the US Army launched its Camouflage Improvement Effort Programme in order to replace the Universal Camouflage Pattern (UCP), regularly in service since 2004. UCP had proved inefficient in Afghanistan. When units on the ground raised serious concerns over the efficiency of the pattern (Whq-forum.de, n.d.), the Army was forced to reconsider its choice of uniform and replace UCP with an already existing interim pattern, UniCam, intended purely as a temporary solution (Whq-forum.de, n.d.). In 2010, the Army released a Source Sought notice, inviting private companies to submit a pattern family, consisting of three colour variations for woodland, arid and transitional territory, plus a blending pattern for Organizational Clothing and Individual Equipment (OCIE) (Wikipedia, n.d.). The notice was reposted on several news sources, all emphasizing the complexity of the problem at hand. Numerous Army experts, including Charles Ryerson (Cold Regions Research and Engineering Lab) and Lieutenant Colonel Timothy O’Neill, pointed out the multifarious parameters the submissions had to fulfil: colours, patterns and textures of natural elements in deployment territories dramatically change from one continent to another and from one season to another. ‘No camouflage can hide every soldier everywhere’, as O’Neill said (Whq-forum.de, n.d.). Therefore, design companies were invited to focus on invariances rather than differences. A sustainable camouflage pattern disguises human shapes not by mimicking natural elements, copying the components of a particular setting, but by charting and reproducing invariables within and across settings. Rather than looking like a part of the environment, the goal is to maintain intact the stream of ‘contextually plausible’ visual information. The implicit suggestion was to take inspiration from the more intimate organizational structures of nature. For instance, at the level of perceptual apparatus, non-conscious cognition processes a fractal form, recognizes it as familiar, in the sense of resembling spatial distributions frequently occurring in a natural environment, and quickly ignores it.
This stress on structural invariances shows a preference for generality. The abstraction of forms and hues has become the chosen strategy for confronting the aporia that every kind of static camouflage runs into: the question becomes, what trade-offs are required to produce one mimetic pattern that efficiently disguises shape and colour in various contexts? What are the general consistencies among particularities that can maximize a mimetic apparatus supposed to operate within the time-space of multiple particular settings? At a more conceptual level: how do we hide a ‘general’ type (or a generalization) within multiple ‘particulars’? As we will see, it is not simply a matter of which consistencies to choose but mostly of how to organize them in a fabric pattern and literally interweave them.
In the past 20 years, computation has been deployed to solve this problem. Algorithms establish the coordinates of the composing elements of camouflage patterns for military outfits. This turn to computation has contributed to rendering fractals and pixels – forms already amenable to machinic computation – prevalent features of the uniforms in use. All available data concerning the environment and the average weather of the war scenario in which certain operations will take place are combined with notions of pre-personal visual perception and existing machine-assisted recognition techniques to compute the optimal pattern. Well-known examples of algorithmically generated camouflage are the Canadian CADPAT, the US Marines’ MARPAT and the Italian Vegetato Mimetico. The first two feature pixelated designs, while the third utilizes fractal forms.
US4CES was the pattern that Hyperstealth Corporation, spearheaded by chief executive officer and designer Guy Cramer, submitted to the US Army Camouflage Improvement Effort Programme in 2013. It belongs to the later generation of pixel-based, algorithmically generated camouflage patterns. It is just one among many cases of digital computation applied to the production of mimicry effects, explicitly using notions of conscious and unconscious visual detection and recognition. I will employ this example with the purpose of elucidating the operative logic of the family of algorithms, but we could make similar considerations about less known examples for which, unlike for US4CES, no information on the structuring logic of the algorithmic family is available (Hyperstealth, n.d. c).
The following text argues that algorithmically generated camouflage is a recent interpretation of the practice of camouflage and, as such, partakes in a genealogy of discourses that themselves are constituted as the reverse of the history of ‘aiming at a target’. This genealogy unfolds along the dialectical axes of depth and flatness in relation to the specific methods of ‘aiming at a target’ that they hope to deceive.
Aiming and shooting at an enemy are actions dependent on the capacity to read a target’s position, orientation and shape, understanding not simply where a threat is located but also what it is, in which direction it is moving and where its weak points rest. Targeting is a matter of gaining a visual, and more generally speaking informational, vantage point to carry out deliberate decisions. Dazzle camouflage during World War I, for instance, served the British navy by delivering an ingenious method of disguising the shape of ships. Black and white stripes on the hull sides of large flagships diverged perspective focal points and made the direction of nautical crafts indeterminable (Scott-Samuel et al., 2011), until the widespread use of planes in warfare made the technique redundant. A better viewing position – and this, in particular, is a first case of the use of vertical scrutiny – rendered the camouflage strategy obsolete.
Camouflage pivots around the incapacity to differentiate shapes, colours, boundaries and directions. The mimicry of camouflage harnesses superficiality and works against depth – literally, as depth of perceptual field, for instance, or figuratively, as depth of knowledge – since depth is indispensable for the parsing of the elements that compose an image. In its various manifestations in time (from the first documented instances at the beginning of the 20th century onwards) and across milieux (animal or military), camouflage aspires towards flatness. A target’s shape and colour flatten on a backdrop, merging with prevalent visual information extracted from the surrounding environment. The deception of a predator’s perceptive faculties inhibits any deeper inspection of an image. The chances of detecting three-dimensional intruders drastically diminish with the shortening of exposition time. We can describe camouflage strategies as ensembles of techniques that try to suppress one visual dimension within the system of observation: depth.
In this article, perceived flatness counters depth rather than verticality – a term considered in previous stages of this work and then dismissed. The vertical dimension, with its bird’s-eye view and omni-comprehensive capacities, features significantly in recent scholarly work. Stephen Graham and Lucy Hewitt (2012) argue that there is enormous value in appreciating the specific qualities of verticality, as opposed to more consolidated horizontalist perspectives. Security apparatuses have undergone processes of material verticalization with surveillance from the sky. Urban spaces increasingly reflect unequal wealth distribution, marking differences between individuals living at street level and those who can access the top of skyscrapers (Graham and Hewitt, 2012). Figurative representations of space and city advertisement take into consideration the view from above (Graham and Hewitt, 2012). Yet, especially in discussions of surveillance and asymmetrical military activity, the notion of verticality is mostly a metaphor for omnipotent intervention. It captures aspects such as the asymmetry of power, access to information and technical advancement. Rather than actual verticality, the issue pertains to the capacity to find privileged observation points, which do not necessarily entail the view from above. The question of depth, instead, is consistent with that of the underlying general characters of perception that camouflage counters. The perception of depth is contingent upon the observer, its capacity to occupy better observational positions, which become, in turn, fixed elements for the system under study. Thinking in terms of volume (as suggested by Elden, 2013) is indispensable in the case of camouflage, not just for the need to consider all spatial dimensions, but because the dimension of depth helps in locating visual information in space and perceiving the presence of objects. Since this article deals with the existing relations between the capacity to see and the tactics nullifying this effort, depth constitutes a more convincing counterpoint to flatness than verticality.
To support this claim, we will first trace flatness in two different interpretations of camouflage. We will describe how hiding a target in a background, in these two cases, depends on the capacity to produce general invariances and consistencies and plays on the relations between target and perceived background. Along the way, we will appreciate a cybernetic notion of camouflage. On the one hand, this magnifies a structural similarity between camouflage and preemptive control (in the sense of bringing what is ‘outside’ inside thanks to topological malleability) (Massumi, 2007). This similarity does not simply contextualize our example within a historically prevalent mode of power: it also stresses a special affinity between camouflage, as the art of anticipating perception, and control, as the most characteristic manifestation of anticipation. On the other hand, this cybernetic reading helps a leap forward into a computational understanding of camouflage. At this conjuncture, we will place algorithmically generated mimetic patterns alongside their corresponding uniforms to finally evaluate what type of flatness these can afford.
In this latter case, camouflage will follow the way of abstraction or rather of threads of abstraction that match different tasks of human visual cognition and perception of pre-imagined war environments. To combine these threads, the family of algorithms reads them individually and then weaves them together in a computational mesh: the surface of the uniform. This fabric is woven from an abstraction 1 of the perceived environment, and, more significantly, it is the result of a combination of parameters that, once computed together, become indistinguishable. Camouflage equals unreadability. Indiscernible elements – in the sense that they are impossible to parse and hence understand – compose a localized image that can be exposed in plain sight, but rests illegible.
The argument we propose reflects on both computation and camouflage. Existing theories on camouflage have looked at practical interventions suggesting various tactics. 2 This article will propose a different method: it will observe camouflage at a more abstract level. It will mobilize the notion of flatness, in the sense of impairing the capacity for making distinctions and eliminating depth, to characterize a camouflage function common to different hiding practices. This foundation will help detect, in the last section, the type of camouflage that the computational principles of military equipment inspire and the kind of depth it may counter. Hence, this piece will not look closely at military strategies and their efficacy. It will rather take inspiration from one military application of camouflage to expand theoretical readings on mimesis and contribute, in the end, to the question of what does it mean to hide in the age of computation.
Aligning the algorithmic case with other examples of flattening elucidates the continuities and discontinuities that computation requires and affords. The hiding strategy, inspired by a certain reading of the case study, is no longer founded on either resembling the immediate background or harnessing common traits that we perceive in the environment, but on the qualities of computation itself. From hiding by producing similarities with one’s surrounding context, we move towards nesting in filaments of data seemingly impossible to sever. A mimetic uniform, resulting from the trade-offs of common elements underlying multiple particular settings, can still be detected within these same (or similar) particular settings. Yet, it can conceal – that is, make unreadable from any perspective, countering any form of depth – the generalities that the trade-off mechanism computes together in a knot difficult to untangle.
Brief considerations on camouflage and thought
Before diving into material interpretations of camouflage in relation to flatness, we shall spend some time with a possible conceptual articulation of camouflage, still utilizing flatness as a purely descriptive tool.
In their recent readings of Charles Peirce’s metaphysics, philosopher Reza Negarestani (2013) and mathematician Fernando Zalamea (2014) consider the interactions between continuity and contingency in the construction of a protean reality. For Negarestani, the interactions between mathematical continuity and contingency map onto the philosophical relations between universal and particular, with the articulation of increasingly sophisticated concepts. Such concepts, explains Negarestani, cause a rupture that requires a reorganization of the architecture of information. Once inserted in a generic space, a universal operates a double move of internal analysis and external synthesis with its surroundings. A general concept breaks the surface of a local situation and acts as a local horizon to push forward the production of novel constellations of thought. The newly formed junctures correspond to, first, new sets of coordinates for the moving horizon, which slowly constructs a map of conceptualization by interacting with different localities; second, the reconfiguration of the specificities, or parameters, of the locality; and, third, a channel of communication between universal and particular that the concept can productively manage.
The relation established between local and global passes through a universal concept, or a generality, here meant as a generative force: an expression of the power of thought. A universal is inserted in various particulars as a movable local horizon that, in its movement from particular to particular, (re)constructs the universal, in a version orthogonally related to the one with which we began.
In all this, Negarestani (2013) describes a notion of flatness: when global and local perfectly overlap and no new coordinates filter out, there is no further articulation or protean sophistication of thought. No rupture occurs. Further, the condition of flatness always presents itself during the initial interactions between global and local.
The case of static camouflage, with preset mimetic features hiding a shape within the context of multiple particulars, resonates with this set of images, especially when the mimetic features correspond to an initial generalization of environmental patterns. The goal is to materialize a generalization that can fall flat onto a vast number of particular settings – halting detection, recognition, and then also thought and speculation. The project is impossible and, using Peirce’s philosophical system, we can trivially demonstrate so. As we will see more in detail, camouflage patterns correspond to the abstraction from sets of similar particulars and are then reinserted within those. Therefore, the universal is not reconstructed as a local horizon moving across particulars but is a given: it is the outfit itself. That the general abstraction will merge with different local backgrounds, without odd edges standing out, can be but a hope or a matter of chance, unless the initial abstraction already comprehends (has already exhausted) all possible addresses for the movable local horizon. Such an event would be plausible only if the original sets of similar particulars, from which we abstract the general, were closed and completely under control. Clearly, this is not the case. In fact, the environments sampled for the production of camouflage patterns testify to a priori decisions over imagined theatres of war: where the US Army expects to be waging war dictates the colour variants of the commissioned outfits.
It is significant to redescribe camouflage, at an abstract level, as the (perhaps self-defeating) effort to flatten the general into particulars and so halt thinking. Ideas, notions, subliminal messages – to hide next to or inside each other – shall aim for flatness. Local and global horizons will look like the same, encrypting any newly synthesized coordinates in the perpetual production of self-similar connections.
Notes towards a genealogy of discourses on camouflage
From a more concrete perspective, any camouflage technique works against faculties of distinction, parsing, any reading of particularities and details that would otherwise stand out. Since the observer’s perception is a constitutive element of the target’s camouflage practice, mimicry becomes a problem of both time and space. Nevertheless, we preliminarily explore these two dimensions separately to show how informational flattening manifests itself and what kinds of configurations it produces. We will then recombine spatial and temporal axes to link camouflage and control strategies and lay the foundations of our analysis of algorithmically generated mimetic patterns.
Flatness as crushing to the ground: Data reduction
Roger Caillois associates natural phenomena of mimicry with psychotic and schizophrenic mental disorders, which hinder a subject’s capacities to determine its position in space. Quoting Lucien Cuénot, Caillois (1984: 25) dismisses evolutionary arguments that justify the selection of mimetic traits among invertebrates as either attack or defence mechanisms. Camouflage, for Caillois, is the actualization of a natural tendency towards imitation. The animate tends towards the inanimate in a singular renunciation, the ‘inertia of the élan vital’ (Caillois, 1984: 32). Like produces like, in virtue of an irresistible tempting force that the environment or space – at a more abstract level – exercises on animals and human subjects alike.
What is generalized are not recurrent elements from various environments, reproduced according to the perceptive capacities of a predator, but the immediate local setting. In the context of military camouflage, a correspondent strategy would be imitating heavy foliage, as in Ghillie suits. These are rarely effective, first, for woods are the only type of territory in which they can be deployed and, second, for each suit must replicate the exact kind of leaves and branches surrounding the potential target.
We can rephrase the inclination towards imitation in terms of a learning method: a pragmatic acquisition of information that requires no further conceptual elaboration and, in eminently successful cases, flattens the individual onto its surroundings. Just as in some cases of psychotic and schizophrenic patients, space itself becomes a generalizing entity, producing a ‘depersonalization by assimilation’ (Caillois, 1984: 30). The potentially devastating effects of this ‘devouring’ force are quite literal – says Caillois – in the case of the Phyllium or leaf insect that imitates its natural environment, and especially the leaves it feeds from, instantiating tragic cases of mass cannibalism (Caillois, 1984: 25). Here, generalization corresponds to blank copying, and flattening to crushing to the ground. In regard to this, Caillois makes clear that mimesis and camouflage are essentially a luxury. Assimilation to space and loss of individuality via meshing with a background are expensive operations: anyone tempted to try should be able to afford their corollary risks. Much along the lines of George Bataille’s engagement with Marcel Mauss’s anthropological exploration of North American potlatch, expenditure is associated with antisocial behaviour and social unravelling. In this case, the redundancy of luxury cannot assume reversible social functions and, in fact, instigates constant antagonism (Bataille, 1985: 116–129). From an informational perspective, for instance, the imitation of data corresponds to redundancy and the horrifying danger of not being recognized, precisely when distinction is indispensable to avoid lethal data reduction.
We can map the metaphor of Caillois’s Phyllium, with all of its risks of misrecognition, onto military camouflage. Uncanny similarities among soldiers’ uniforms on a battlefield can make efficient patterns vulnerable to friendly fire. Within the broader context of the industrial war complex and its ramifications in economic investments, the little bunch of corporations that compete with each other over the budget allocated for the design of military patterns catalyses a psychotic function. It patterns particular national uniforms to the general environment of global conflict, magnifying the senselessness of war. 3 Recognition is crucial to soldiers almost as much as mimesis.
An evolutionary/cybernetic reading
If we disregard Caillois’s interpretation of mimesis and embrace more mainstream evolutionary approaches, dissimulation becomes mostly a defence or attack mechanism. The perceptual capacities of the observer are constitutive elements of mimetic traits, and, as soon as perception comes into play, the time factor joins the space factor to recompose the camouflage problem.
At the beginning of the 20th century, Abbott Thayer took inspiration from nature to produce the first experiments in military camouflage. His main references were to Alfred Wallace, who defined mimicry as a set of adaptive traits (Thayer, 1918). The virtues of colour-changing features, for instance, in the case of surprise attack or for deceiving a predator, have motivated their natural selection by a species. Therefore, every study of the coloration of an animal mimicking its background has to begin from the perspective of the predator to be deceived and its perceptual capacities (Merilaita and Stevens, 2011; Thayer, 1918; Troscianko et al., 2009). ‘Successful camouflage will be that which matches the statistics of the neurally filtered visual scene: the same distribution of luminance, colour, textures, edges and, where salient to the viewer, derived features such as shapes’ (Cuthill and Troscianko, 2011: 11). The capacities of the predator’s psychology and nervous system to filter information, both spectrally and spatially, the nature and degree of data reduction operated, and so on, are the main sources employed to explain the evolution of camouflage traits. From this perspective, a camouflaged animal is one whose visual appearance adaptively anticipates the selective evolution of the detective capacities of its predator. A target’s capacities to disguise and the correspondent predator’s capacities to detect are selective traits caught in protracted feedback loops that span evolutionary time.
A second type of feedback loop is at work on a much shorter timescale. Detection, recognition and identification are accumulative tasks and therefore happen over a time interval: a minor loop stretches between the accumulative work of a predator’s unconscious perception and cognition and the different mimetic capacities a prey displays to avoid background/target segmentation and shape identification. Noticeably, to remain hidden and avoid detection altogether, a target needs to remain under the threshold of awareness of the observer, not even interfering with non-conscious computational tasks, which operate at a much faster velocity than the conscious ones. 4 What manages to anticipate the intense neural processing of non-conscious perception will not be transmitted on to higher-level and slower-speed cognitive processes.
Troscianko et al. (2009) provide a systematic explanation of the tasks that target detection and recognition entail. A brief excursus can help elucidate the functioning of the shorter feedback loop and identify the objectives of the algorithmic subsets that compose the software of our case study.
Edge and motion detection and shape identification are the most significant tasks of target recognition. We can consider the first two as particular cases of the third and more sophisticated one, which consists of detection, recognition and mental reconstruction of a familiar form (Troscianko et al., 2009). Detection of an alien feature in the environment deals mostly with the question of location – where is the target? – while recognition and reconstruction address questions as to the nature of the object of observation. Although there is general agreement among evolutionary biologists that the tasks of determining presence and location of an object and the one of establishing its identity are carried out separately, their temporal relations and anatomical pathways are still unclear. As Cuthill and Troscianko (2011: 7) suggest, it seems unrealistic to treat the two macro tasks as strictly sequential – first detection and then recognition. It is more likely that incipient recognition feeds back into detection, originating recursive neural feedback loops within the stretch of time necessary for recognition. 5
Recognition of an object’s form requires matching three-dimensional shapes with the two-dimensional image available by either breaking the latter down into composing parts or considering it as a whole. This task is completed through what Troscianko et al. (2009) define as ‘shape recovery’, stressing that ‘the percept of three-dimensional shapes is not built from its elements. Instead, the three-dimensional shape percept is formed by the application of abstract shape properties, such as symmetry’ (Troscianko et al., 2009: 455). In other words, a form of abstraction is what completes the process of identification and optimizes its efficiency by simplifying and reducing the number of cognitive steps. Significantly, Troscianko et al.’s description of ‘shape recovery’ implies a pre-existing knowledge, if not of the shape of a target, at least of a number of expected invariances and patterns of symmetry. During the temporal interval required for recognition, the observer gradually becomes aware of the presence of a target and interprets its figure at a conscious level. Detection and recognition are the results of incremental operations that occur in a concerted manner.
On the opposite front, to become less visible, animals deploy various techniques to counter at least one of the three cognitive tasks just summarized. The mimicry strategies providing most of the inspiration for military camouflage belong to the categories of background matching and disruptive coloration.
Camouflage entails anticipation of the work of vision, which is constantly learning how to read new parcels of information in an image. We can localize a temporal gap at the levels of both co-evolutionary adaptation and neural response. What the observer’s visual apparatus will likely perceive in the future, thanks to the traits that adaptive evolution selects over generations, is hidden in the present. Similarly, what the neural system will recognize after extended focused exposure cannot be seen at first sight. We do not hide just in space, but also in time. In the case of military camouflage, a successful uniform immerses the soldier in a moving time-space, always a step ahead and a moment before being recognized, closer to identifying the enemy and further from being located. The uniform is designed to embed a suitable replica of the invariant features of the surrounding environment, one that matches the ‘perceived environment’, anticipating the operations of differentiation between environment and target.
Now, if, instead of contrasting Caillois’s interpretation with this second evolutionary/cybernetic argument, we integrate the two together, the operation of flattening onto an artificial pre-personal natural guise – depersonalization in a generalizing space – gains a novel psychological dimension. A new thread of intimate correspondences, passing through non-conscious, pre-personal perception, tightens the time-space of the war theatre. The perceptual system of the observer and the features of the observed are now inseparable, in the sense that they can no longer be parsed. In a way similar to that of the previous discourse, flattening operates as the suppression of one visual dimension within a frame of observation; however, in this case, the flattening operates within a time-space perceptual continuum. It might be thought that the suppressed dimension of depth, here, would highlight the discrepancies between the perception of a particular context and that of the generalizing features of the camouflage pattern, but the two are flattened onto each other, at least temporarily, thanks precisely to the underlying perceptive gap.
This reading of camouflage pivots on two classic aspects of second-wave cybernetics: not only are the perceptual faculties of the observer constitutive elements of the perceived object but, specifically, the faculties are those of non-conscious perception. Cybernetic experiments conducted between the 1950s and the 1960s introduced a new idea of vision. Milestone investigations, recently indexed by Orit Halpern (2014a), such as that on the vision of frogs by Lettvin et al. (1959) or Béla Julesz’s experiment (1960) on depth perception and form recognition, refined the visual apparatus. The eye became an autonomous machine for abstracting the world (Halpern, 2014a: 62).
Tracing in history the connections that go from perceptual circuits, to structure of cognition, all the way to governance, Halpern (2014a: 207) re-reads these experiments to highlight two findings. First, instead of transmitting simple copies of its object, the eye provides to the brain information that is already highly interpreted and organized. Lettvin et al. (1959: 257), for example, observed the behaviour of a frog’s eye and the data it transmits to the brain and argued that the type of perception occurring at the level of visual apparatus is already a form of cognition. Some decisions have already been taken, without the intervention of the brain, outside the realm of conscious representation. Put in the perspective of a war scenario, visual apparatuses, machinic and human, co-constitute each other’s appearance via feedback loops that involve non-conscious autonomous perception. Second, vision emerges autopoietically from within a system. The test subjects of Julesz’s experiment could see depth in abstract computer-generated patterns that are not available in nature. Vision was thought of as disjointed from an external referent and simply emerging from within a specific system of reference.
In their symmetries, the two findings speak of a vision that not only transmits but also organizes, or rather produces, meaningful images. If vision emerges autopoietically from within a system, then its relation with abstraction gains another layer of sophistication. Vision abstracts information without the help of conscious cognition – as in the experiment of the frog’s eye – in the sense that vision emerges from abstract forms: it produces information out of abstraction – as in Julesz’s experiment. Therefore, on the one hand, vision moves and organizes space-times using processes of abstraction; on the other hand, if the system anticipates visual inputs, it can induce the type of image produced. This constitutes an efficacious way to manipulate perception, but it also marks the production of a computational model of vision and of the corresponding mechanisms preventing detection and recognition. The numerous cognitive tasks usually at work when focusing on a target can be predicted, modelled and computed via algorithms. This leap into computation, with its specific processes, produces new affordances beyond a basic replication of vision.
Using the interpretation provided so far, we can easily map control, as the prevalent mode of power, onto camouflage, as the art of anticipating perception. For control, here, we shall mean the regime of the post-2001 ‘war on terror’, described by Brian Massumi (2007), with his differentiation between prevention, deterrence and preemption. In camouflage and control, we face similar configurations.
In the evolutionary/cybernetic reading of camouflage, the observer becomes an integral part of the system to the extent of losing its nominal function: its presence modulates the spatio-temporal reality of the system and is modulated in return. The result is a self-generating continuum, which encompasses prey, predator and medium, altering conventional definitions of boundaries and effectively obliterating both observer and medium. The continuity of the surface, in the form of the correspondences that flatten the distinctions between ‘perceived environment’ and target, sustains itself on a gap between first-sight data acquisition and further learning, non-conscious visual perception and conscious cognition. Such a gap is analogous to the one indispensable for control to preempt future events, without altering the topological structure of reality.
Massumi defines preemption as a speculative mode, epistemologically and ontologically distinct from prevention and deterrence. The latter two operate against already identified threats, for which we can objectively name the causes, and, for both, ‘uncertainty is a function of a lack of information’ (Massumi, 2007). Control, instead, has no precise target. Indeed, it should not have one, but rather an open fork of possibilities with unpredictable variables (Massumi, 2007). With a threat that has yet to emerge, the quasi-empirical calculation that leads to decisionmaking cannot prevent the fact, as there is no fact to prevent (or clear perspective fact to deter), so it preempts what has not yet emerged. That is, it avoids the unknown threat by generating the causes of the wanted effects. As Luciana Parisi (2013: xvi) explains in her reading of Massumi, ‘the effects of the unknown have become the causal motor by which control is unconditionally exercised’. As Parisi explains in a previous article, the self-perpetrating dynamic of control inscribes potential futures into the present in the form of probability, putting to work the topological connections between present and future and effectuating a correspondence between what we expect and what has to happen (Parisi, 2012). 6 The topological continuum allows the present decision to shoot off a connection with an efficient cause that inhabits the indeterminate future. And as long as what is outside, in the future, is made amenable to and brought into the present conversation, via means of propaganda for instance, the topology of the surface is preserved.
Any anticipatory intervention on the surface should not break the continuity of the overarching present mesh, here meant also in the sense of a mesh that preserves the status of a constant present. 7 Therefore, such preemptive interventions – the dragging of an indeterminate future potential inside the realm of probability as cause of the present action – can occur, quite literally, in the form of a ‘constant nervous anticipation’ (Halpern, 2014b: 232), a rapid movement that intervenes in the time-space gap between autonomous vision and conscious thinking. The continuum generated at the level of vision appears unaltered if the motion of manipulation happens at a scale commensurable to the gap between the eye and the brain. Figuratively, this is the same gap in which we can still hide.
The primary difference between the two configurations, the one of camouflage and the one of control, is in the resilience of the two hiatuses. Each gap hosts an intervention, and the hiatus remains open as long as the intervention bears its effects. In camouflage, the observer tends to learn and gradually catch up with the once-indistinguishable, flattened image. Consuming the advantage of the gap, the observer eventually appreciates a depth dimension and, hence, the discrepancy between target and environment. To preserve the gap, a target needs to change its appearance, something that happens at an evolutionary scale but is not always possible at the moment of detection. The topology of control, instead, is based on a more dynamic intervention hinging on the possibility for iteration. In static camouflage, an animal can usually ‘bring into’ its appearance salient elements of the ‘perceived environment’ only once. In control, the operation is constantly repeated.
In the military sector, static camouflage needs to rely not simply on the observer’s perceptive faculties but also on the generality of certain environmental elements and of the way in which those are perceived. Nonetheless, past a certain threshold, the observer’s perception will eventually detect the target. 8 So, how does the static algorithmically generated camouflage of our example intervene in the space of a war scenario? It aspires to participate in the logic of control that Massumi describes, but its intervention is limited. If we think in terms of temporality, it moves in two directions. Algorithmically generated patterns anticipate the battlefields in which the US Army will fight, envisioning desert and woodland territories, and deploying computation with its modelling and anticipatory capacities. Yet, these are modes of anticipation rather than preemption. The empirical ground is still to be tested. At the same time, computational design mobilizes environmental invariances, pre-individual natural guises. The algorithmic family abstracts regular patterns of distribution, producing contextually plausible visual information. The Massumian movement of ‘bringing’ future possibilities ‘into’ a present discourse, in this case, takes the form of an abstraction of frequent organizational structures that pertain to general nature. The strategy is tapping into our perception of regularities, cutting across common human faculties for pattern recognition. Patterns, here, are expected to remain under an observer’s threshold of awareness. The problem is that the threshold of awareness moves. 9 Therefore, we can see that algorithmically generated camouflage uniforms are aligned with static natural camouflage rather than with Massumian control. The patterns are static rather than interactive. They are assembled on the basis of already-known psychological notions, and the mimetic uniform is just one step ahead of the evolutionary stage those embody. They mostly perform a deterring function. An observer has significant chances of eventually detecting and recognizing the target, exhausting the perceptual gap.
The limit, the halting mechanism, is the physical limit of the uniform itself: a banal piece of fabric rather than uncomputability and omega numbers. The mechanism of arrest, owing to restricted iterability, diverts attention towards the distribution and concretization of the pattern within the uniform itself. We have already discussed how the camouflage uniform cannot sustainably merge with multiple particular backgrounds, standing out with its odd contours and eventually breaking the mesh of vision. But, what happens inside the uniform? How are the different tasks of recognition and detection prevented? Can we still find an element of flatness? And towards what type of depth dimension is it directed?
US4CES: Flatness as computational weaving
Algorithmically generated camouflage patterns borrow from nature the mimicry strategies of background matching and disruptive coloration to impair the perceptive faculties of ambient and focal vision. Ambient vision is predominantly responsible for detection – the observer successfully distinguishes target and environment and identifies the presence of a potential target – while focal vision is responsible for recognition – the observer recognizes the shape of a three-dimensional object in the field of vision as that of the target. Although disruptive coloration and background matching lead to different types of concealment according to different principles – for instance, background matching mostly aims at the avoidance of detection – disruptive patterns improve when combined with background-matching strategies, such as imitation of colours, textures, density and the distribution of markings in the surrounding environment (Stevens and Merilaita, 2009; Wilkinson and Sherratt, 2008).
Disruptive coloration works according to a set of five subprinciples. First articulated by Thayer (1918) and recently reviewed by Stevens and Merilaita (2009), these principles 10 provide inspiration for the military industry. In background matching, different hypotheses concerning optimized colour/pattern relations have been formulated over the course of the 20th century. Thayer first suggested that animal coloration averages samples of backgrounds where the target is more likely to be seen by its predator. Contra Thayer, Endler sustained that to optimize background matching, a target should resemble a random sample of its usual habitat: the match is with the coloration of a patch type or the compromise among a limited number of patch types, as Merilaita and Stevens (2011) have proposed. In the realms of both biology and military applications, camouflage strategies negotiate between specialization and compromise.
To circumvent the aporia of having to hide a human body in multiple locations with a limited set of capacities remotely determined, recent samples of algorithmically generated camouflage choose the way of abstraction. Rather than simply copying the background or the perceived background, with a depersonalizing generalization, algorithmic subsets originate pattern components by abstracting the colour and prevalent shading of an environment and by tracing backward the operations of abstraction carried out by human and machine vision. Generalization is not meant in the sense of horizontally imitating the surrounding setting, producing a guise that can be imposed on a background. Rather, it traces rules apparently organizing natural shapes. As already seen in the previous section, the strategy is not flawless and it delays recognition rather than permanently preempting it. Still, fractals are just one of the abstracting principles strategically interknitted, first, in the software family and, then, in the uniform. If static camouflage cannot interrupt recognition, its attempt to do so produces, precisely in the discrete unit of the uniform, a self-contained moveable abstraction of the perceived war scenario. In this process, the distribution of the pattern via computation deserves some attention.
The case study for this text, US4CES, is displayed online (Hyperstealth, n.d. c) alongside useful insights into the logic of its algorithmic family. The design is described as being the result of the concerted effort of macro-pattern algorithms, micro-pattern algorithms, colour enhancement and camouflage colour algorithms that aim at tricking target/background segmentation and object recognition. In terms of camouflage strategies, all algorithms of the family draw on the subprinciples of both background matching and disruptive coloration, following the notions on perception mentioned above.
The macro-pattern algorithm aims at delaying and, when possible, avoiding detection. It is responsible for the production of the broader pattern of the outfit and distributes repetitive shapes across the uniform according to overarching fractal logics, the distinctive feature of US4CES. Among its subsets, we find a symmetry-disruption algorithm and a symmetry-axis algorithm that breaks mirror distribution along all symmetry axes. A movement-concealment algorithm aims at disguising the human shape when the user is standing still or in movement. The algorithmic subsets target the working of ambient vision, while they have little to no impact on focal vision. For this reason, previous static camouflage gear overlooked the need for a macro pattern of distribution and deployed just non-coordinated random disruption and background colour matching. The design principles of US4CES, instead, suggest that to hide a shape in an intricate environment it should be irregular, but not too irregular.
To achieve the right balance between invariances and discontinuities and also to deter focal vision, recurrent distribution patterns are combined with visual noise. A micro pattern complements the macro pattern. The two are computed together according to a flattening of temporality – something significant for our argument. 11 In the words of programmer Guy Cramer, the micro pattern aims at preventing recognition once the observer has detected a potential target. 12 The computing logic, therefore, differs from both previous random and semi-random patterns and natural perception, which we know operates cumulatively over temporal exposure and does not require a linear sequence from detection to recognition. US4CES conceives the two as separate problems to target with different tools that are simultaneously embedded in the outfit at the time of computation.
Finally, in order to determine the colour schemes for the three patterns required by the US Army, a camouflage colour algorithm breaks down the primary colours into percentages of presence in each sample environment, and establishes the hues to simulate shadows and contrast. 13
The family of algorithms works through available data about the environment and, using the logic of human vision target detection and recognition, tries to preempt it. Given that each military uniform is a limited space, the subsets have to compute according to trade-off mechanisms, similar to the ones appreciated in evolutionary paths, where the evolution of a certain feature is the result of a number of concurrent selective forces. The objective is not to produce the most symmetry-breaking pattern or the one that integrates the most fractal-like designs, but to achieve the best possible camouflage by averaging any preemptive visual input and assimilating to space by approximating its organizational patterns and colours. The techno-genesis of the family of algorithms is constituted by the ‘convergence of [its] structures into a structural unity’ giving to the software its ‘specific identity’, using Simondon’s (1980: 15–16) words, moving towards a unitary concretized organism that then leads to the actualization of a specific global pattern. Clearly, the structures converging here expand well beyond a coded version of human psychology. They branch off into the logics of the military industrial complex and contemporary warfare strategies, as well as into the capacities of computation.
There are startling differences between this model of camouflage and the pragmatic learning and imitation we described above. If we look at the individual uniforms, their interactions with the observer’s perception are similar in being eventually self-defeating – an observer can figure out the individual target’s silhouette. But, what interests us is the novel kind of flattening that this model of camouflage introduces. To appreciate it, we turn to the pattern itself. The software developed by Guy Cramer does not simply perform the operations of abstraction of data or implement the usual rules required to cover the interval between legible code and printable pattern. In order to solve the problem of negotiation-specialization, computation is deployed to produce, or rather weave into the outfit itself, an autonomous abstraction of the environment.
This abstraction of the perceived environment is by no means monolithic and interknits the heterogeneous threads of other abstractions. These encompass: generalization of colour and shadows, prefiguration of the types of environment where a certain army is expected to intervene, and all the other principles contributing to micro- and macro-algorithm subsets. These threads become parameters for algorithms that negotiate their own relations in the space of the outfit and materialize a unified concept of what the perceived environment supposedly is. Then the newly produced ‘concept’ is let loose. Released into multiple environments, it finally explicates three fundamental functions: it enacts a structured repository of general information on the environment and an observer’s perception of it; it behaves as part of the environment itself; and it works as a connector, bridging different elements of the environment without altering the amount of information cognized by an observer.
As already pointed out, perception is accumulative in time. US4CES and similar algorithmically generated camouflage patterns do not position visual inputs, countering identification from different distances and at different times of exposure, at the time when they become needed. Instead, the software preinscribes all such inputs in the military gear at the moment of production. They are flattened onto the surface and arranged in the limited space of the uniform according to an efficient disposition that carefully reflects mimetic principles. All threads of abstraction are locally articulated on the pattern according to the parameters of a specific distribution. Streams of information are compressed in a limited space and quite literally folded into it on the bases of strategic positioning.
Computation makes it indispensable that each cognitive task, working on the recomposition of environment and target, is first considered individually, as a single thread of abstraction. In a second moment, the production of the outfit computes all the threads together, each infecting the others. This orchestrated display of abstracted information on a flat surface inverts the current logic of clear data visualization and the supposed capacity to hold a visual grasp on large quantities of individual data in real, or almost-real, time. It is precisely the meshing or weaving function of computation that flattens otherwise cumulative visual inputs, their different temporal intervals and spatial gaps, on a singular surface. They become all co-present, in the sense of co-implicated, thanks to careful interknitting. Sadie Plant’s (1997: 12) description of interactive and hyperactive media suits this case well: ‘The yarn is neither metaphorical nor literal, but quite simply material, a gathering of threads which twist and turn … shuttles and looms, cotton and silk, canvas and paper.’ This newly formed thought of the environment, once the algorithmic family has woven it together, can no longer be undone: each thread meshes with the others. The movement goes from human vision, as a set of interrelated detection and recognition functions, to the machinic coding of those functions as individual tasks, to their recomposition in a fabric pattern. By the third stage, we can no longer run a precise parsing operator: pulling a thread would mean undoing the mesh entirely.
This type of static camouflage gear is not qualitatively different from non-algorithmically produced types, when it comes to protecting individual soldiers on a battlefield. But what interests us is the novel kind of flatness that computation may introduce. This involves not flattening a target onto the environment but reciprocally flattening the threads of the parameters composing the target itself. Therefore, rather than discussing US4CES outfits as pioneering a new model of soldiers’ protection, we will look at their meshed threads and how these, once tangled up together in computation, become hidden because they are individually unreadable.
Beyond its capacity to counter detection on a battlefield, the computed mesh of the outfits inspires a way to camouflage in other kinds of hostile environments. Contemporary forms of targeting (in military and civilian contexts) are characterized by the collection of personal data and surveillance – drones are a military manifestation of this pervasive regime. Most security interventions are based on the capacity to discern data and establish responsibilities. Therefore, hiding entails hiding information or, at least, organizing information in such a way that it cannot be expounded with clearcut precision. Flattening information, by computationally weaving threads of abstraction in an indistinguishable mesh, may counter precisely these modes of surveillance and the rhetoric of precision they rely upon.
In the previous discourses, the invisibility of a target corresponded to the incapacity to read or distinguish its features and isolate one of them. Invisibility relied on the flattening of discernible elements into a background. In this latter case, instead, being able to see that ‘something is there’ and being able to discern its internal features are different problems. Before, the inability to detect a target relied on the inability to make distinctions among colours, shapes and directions of target and background. Now, even detecting the presence of an intruder standing out from a context does not guarantee that, eventually, we will be able to read, track and finally name its composing elements, the threads of abstraction woven in the outfit. The flattening occurs in the individual uniform. It pertains to the composing threads, rather than the relation between outfit and environment. The relation target/background becomes secondary, in comparison with the previous cases. The reproduction of general traits consistent with the environment is significant here, but mostly because the traits end up composing one autonomous abstraction of such environment. Finally, the observer’s perception and capacity to avoid the flattening effect seem to lose importance. Indeed, if weaving hides individual components in the computation process, what type of perspective can restore depth and make them legible again? This mode of flattening may counter a mode of recognition that dislodges the definition of depth to the point where it no longer depends on perspective.
We ask, then, what is the mode of recognition in question? The contemporary regime of data collection pivots around a supposed capacity to atomize data and recombine it with other inputs. The regulation of more or less sophisticated tasks, from traffic flow to consumer choices, depends on the constant breaking down and rewiring of data into usable – that is, readable – narratives. Gaining a better perspective to recognize objects or read sets of data often equals to isolating and recombining them. Naturalized positive notions of data access mask the operations of mining, crafting and designing (or beautifying, using Halpern’s paradigm) that data undergo while entering the visual dimension of our perceptual field. Among the many features that data crafting capitalizes upon are atomization and spatialization. We seem to be able to decompose data into ever-smaller particles to be reorganized in the required types of visualization. In this, the capacity to discern at smaller and smaller scales and parse increasingly refined – that is, precise or personal – sets of information is crucial to security, and especially to the rhetoric to which it is attached. In the example of algorithmically produced military outfits, instead, the possibility of unwiring and rewiring input threads is limited: the data are the wiring design itself.
Footnotes
Acknowledgements
I would like to thank Celia Lury and Olga Goriunova for their patience and constant encouragement, and Louise Amoore and Vohla Piotukh for organizing the workshop for which I wrote the first version of this article.
Funding
The doctoral research upon which this article draws was supported by a Chancellor’s Scholarship from the University of Warwick.
