The Introduction to this book started with astronomy and the capture of the light of the celestial night skies that became an object of calculation and comparison. Photographic glass plates detecting light turned into data about celestial objects; images triggered calculations that produced further (nonphotographic) images such as diagrams and measures from images (such as light values in comparative data tables). Thus it feels only natural to round out the book by returning to the sky, beyond human vision but constantly calculated, some of which even returns as an image. This is also a parallel story of a more recent date that provides a diffraction pattern of these two perspectives: different sites of practice, both about astronomy and images, more than one hundred years apart, yet both dealing with technical images and photography in ways that have a specific relation to data and institutions, measurement and invisual visibility. So, fast-forward 120 years to offer a fitting parallel development that concerns photography, data, and the planetary dimension of observation.

In 2019, a peculiar image was announced. Something made available for the wider public as a spectacle of a different kind and not merely for astronomers to analyze. Something that was even said, in some commentaries, to be a photograph, even if it could not possibly work like terrestrial photographs usually do, conditioned by sunlight. If photographs entered the world as a technology for capturing the solar spectrum of light that became quantified, compared, and managed, this one seemed to do the opposite: to capture the absence of light as it pictured a massive black hole at the center of the Messier 87 galaxy. The rim of energy around this entity that is 6.5 billion times bigger than our own Sun functioned as the proxy signal that black holes exist. As such, not exactly visible in the mundane sense of the term, but something that can be seen through measurement, parsed together as a composite image, fabricated based on a signal frequency captured by multiple radio dishes. Such an image lends itself to all sorts of philosophical theories concerning perception, absence, and other fantasies about the real and the imaginary, possible and impossible. But in this case, even the (seemingly) simple question of how this became possible is fascinating as an entry point to a different way of understanding not only the astronomical perception and its scientific instruments but also the function of photography. A more media-theoretical and technical account might simply refer to it as a large-scale Fast Fourier Transformation operation, where an image is determined in relation to the frequency response of the black hole. As Robert Pietrusko explains, an analogy for this procedure would be a sonogram of a sound wave, but one wherein parts of the signal spectra are missing.1 Operational images are not an aberration, but the standard case of images—and operations that produce images, alongside other entities such as datasets, models, and more.

The Event Horizon Telescope emerges not only as an astronomical science project but as a way to rethink images. This experiment with imaging is premised on a distributed high-tech scientific apparatus, training, and lessons from the history of science where aesthetic techniques have been mobilized for various ends. Harvard-based historian of science Peter Galison, involved in the Event Horizon and Black Hole Imaging project, explains that we are dealing with a peculiar perception, measurement, and registering of light that bends around the black hole: “Black holes are surrounded by orbiting gases that glow with heat. Still, the challenge of imaging is immense: not a pixel would exist without the combined efforts of telescope operators, instrument designers, theorists, data experts, and engineers, from early-career to senior scientists.”2

Therefore, part of the team’s work was to engage with the history of scientific imaging and notions of objectivity from drawing exercised in scientific atlases to photography and to the “trained judgment” that emerges as a particular skill of interpretation over the twentieth century.3 Such questions helped to guide the methods of how to form an image of something in traditional terms invisible but still potentially representable through a synchronization, analysis, and computational interpretation of massive digital datasets of signals (each telescope site of the project produced about 350 terabytes of data daily, stored on “high-performance helium-filled hard drives”4). As an exercise in massive-sized interferometry, coordinating and synchronizing with the help of atomic clocks, the received radio wavelength signals formed one operational backdrop to the complex process of constructing the relevant spectral signature of the black hole. At the same time, part of the problem persisted in how to compose, from those signals, an image of a thing that has never been verified in visual terms. Earlier theoretical models in physics exist, and now a measured spectral manifold too. But how might we build a plausible image that conforms to this space of expectations? In other words, there is no “ground truth” against which to compare the accuracy of the reconstruction. Moreover, as synthetic datasets are becoming more and more central to machine learning practices, there was an additional dilemma in relation to this comparative process of what signals indicate: What might they picture?

No wonder, then, that artist Rosa Menkman includes the black hole image in her catalog of “impossible images”: these are images that have pushed the boundaries of signal processing, and they linger on the horizon of visible/invisible, visual/invisual. The Event Horizon Project visualization sits in the category of “images of objects that can be only inferred, transcoded, or otherwise ‘perceived’ involving a significant signal/noise bounty” alongside other categories and examples of images that were once impossible (X-rays and others), that will become possible (Medipix particle imaging), or that remain fully speculative (an image of dark matter). As phantom shots, impossible for living humans, they occupy a central role in technical ways of seeing and knowing.5

A key technique examined in the Event Horizon Project, the CHIRP algorithm (Continuous High-resolution Image Reconstruction using Patch priors) engineered by Katie Bouman at MIT (again, we are back in Cambridge, Massachusetts), was presented as one solution to molding, interpreting, and preparing data into an image. It includes coordinating the various received signals and cleaning them from atmospheric data, as well as modeling the massive amount of input so that it helps the later process of building plausible models of what kind of an image could be composed in that image search space. Here, the image is algorithmically composed and premised on training sets in machine learning techniques.

Bouman used a machine-learning algorithm to identify visual patterns that tend to recur in 64-pixel patches of real-world images, and she used those features to further refine her algorithm’s image reconstructions. In separate experiments, she extracted patches from astronomical images and from snapshots of terrestrial scenes, but the choice of training data had little effect on the final reconstructions.6

The six observatories around the globe (in Chile, Hawaii, Mexico, Arizona, Spain, and the south pole) that were part of the registering of signals are a significant upgrade from the glass plates sent from Peru to Harvard. Furthermore, instead of only (human) computer analysis of annotation and comparison, computational methods helped to synchronize the perspectives from the various rotating data-recording places of observatories in order to get a rich set of signal-detecting angles, so to speak. Of course, it was not merely studying automated algorithms that enabled us to see phantom views. As Galison outlines, the project’s focus on what kind of a data process leads to a reliable image was based on training four teams with different methods to work with a sort of reverse engineering of visual datasets back to the test image:

First, all four teams received test data sets created from unknown images. Some were astronomical objects, others were simulations, one was even an image of Frosty the Snowman. Running blind, each team had to reconstruct the original test image. When all teams could successfully pass the tests, then—and only then—did we tackle real data. Still in isolation, the teams began peering, in awe, at the first real black hole images. A month later (July 2018), the four teams gathered to compare their catch, pixel by pixel.7

From early expert reading of photographic images to such later teamwork that consisted of developing algorithms, methods, and more, at least one thing remained similar: the ability to stitch together input from around the planet in order to enhance the accuracy of the analysis. Pictures, to say the least, are processed and fabricated in complex operations on the ground of the Earth, in institutions, in and through such operational apparatuses that are more than just one machine. Distributed operational aesthetics indeed. In the case of Event Horizon, argues Benjamin Bratton, the whole planet is turned into a massive camera, the planet becoming an integrated sensing surface that enables data transmission and synthesis.8 However, to be more accurate, the actual operations consisted of much more hard work in the synchronization, fine-tuning, shipping of massive datasets on physical hard drives (delivered by human couriers on ships and planes), with the center of operations (still) in Cambridge, where data was parsed together in the interferometric operation. But as an operation of integration of scales, we can still refer to “new economies and forms of knowledge,” as Orit Halpern puts it, that are at play. In these operations of scale, surface, and imaging, she continues, particularly acute questions of “how we currently govern and manage life on Earth through computation”9 emerge.

These high-tech images are made in teams, they are constructed out of signals and other data, and they are modeled (with machine learning techniques of optimization and search) to fit preexisting ideas of what an image might be like. Such images are born thanks to Bayesian heuristics (among many other layers already mentioned). Even if operational, they are thus often pictorial, even in contexts of advanced big science (including for reasons having to do with gaining public support and attracting funding, etc.).10 Moreover, images—and this applies to not only astronomical imaging but also contemporary data culture—are constantly formatted and calibrated in relation to their technical platforms of use as well as what is expected of a picture; all of this shifts the primary focus from what is being represented to their operational preparation.11 And there is an interesting feedback loop between specialist practices and our contemporary theoretical work that asks: What even is an image, especially in the light of things seen and unseen, calculated and synthesized, measured and rendered.12 Remember: tell me what spectrum of electromagnetic radiation you are focusing on, and I will tell what your discipline is.

I have tried to avoid such juxtapositions that early photographic images are more naive ways of picturing and that now, with digital (big) data platforms, we are forced to shift from the pictorial to the nonrepresentational. Historians of photography would be quick to point out how inadequate this sort of a linear narrative is and how early on particular analytical qualities of imaging were at least as, or even more, central than just depicting 3D things on 2D flat surfaces. Instead, seeing, representing, and sensing has taken many twists and turns in ways that build up to the complex layers that make up the media archaeology of operational images.13 We could, thus, take Elizabeth Eastlake’s mid-nineteenth-century words as a good guideline for such investigations: photography is “neither letter, message, nor picture.”14 I have a personal fondness for this short quip. I would go as far as to claim that it is more interesting than many things said about images and photographs within the (commercial) industry discourse of AI and image datasets. Technical images are not just pictures, messages, or communication. They demand a revised vocabulary to address them. Hence: operations.

The concepts, approaches, and ideas that have been discussed in this book, from photographic glass plates circa the 1890s and fin de siècle and then to the computational images around the 2020s, inform the end result of not only what but also where operational images come to mean: grids, platforms, (training) datasets, AI models, urban remote sensing, histories of photogrammetry, and measurement alongside contemporary examples of art and design that practice invisual culture. Operational images consist of layers upon layers, recursive chains of techniques, turtles upon turtles of technologies, practices, people, training, legal and financial instruments, and many other aspects—operations that hit the ground. This approach has meant building on Farocki’s original work and diverting from it to investigate the concept as a force of its own; we can refer to this as conceptual persuasion, to echo Brian Massumi’s terminology mentioned in the introduction.

At the beginning of this book, I hinted that I am more focused on operations than images. And yet, these two cannot be positioned as opposites either. Instead, they are codetermining instances: operations that build up possibilities of an image; images that have operational capacity and meaning in different institutional situations. I have, though, wanted to map an operational approach to images and investigate what the term “operational images,” after Farocki, has come to mean. This has been somewhat less an exegesis of its multiple uses (even if I have tried to include many references to earlier work and applications, including also the past ten to fifteen years of artistic practices) than a scoping for potential; What can it do? How can it function? How can the concept itself operate?

In an interview with cinema theorist Thomas Elsaesser, Farocki had confirmed that considering cinema as part of “the history of other techniques and technologies of surveillance, measuring, calculating, automation”15 is an apt characterization. Elsaesser followed up on this line of thought in his theoretical work, not least in the already mentioned notions of S/M operations of cinema, including science and medicine, surveillance and military (and one could add: security and money, statistics and machine learning, stonks and metaverse, and others).16 Beyond cinema, operational images have been discussed in this book as those instances that are about images and “other techniques” which in some cases also problematize what is visible in the first place. An ontological example such as black holes is in this sense intriguing and influential, but many sites of social visibility and power (of capture, targeting, control) on the earth’s surface would be too. Farocki, writing about the U.S. carceral system, echoes Deleuze’s point about societies of control: “With the increase in electronic control structures, everyday life will become just as hard to portray and to dramatise as everyday work already is.”17 Another point about operational aesthetics raised throughout this book: that which pushes much of visual culture to the sphere of invisual operations that mobilize different techniques of observation does not necessarily look like anything special. It is in the intermedial relations, the techniques of imaging and observation, and their operative stance to contemporary formations of territory, extraction, finance, and the production of data (and more), where some of the structured dynamics of the terms “operational aesthetics” and “operational images” emerge.

In many ways, I have tried to keep in mind Farocki’s audiovisual method of soft montage when writing the various entry points to elaborate the potential scope of operational images. This has not meant that Farocki’s words or works are to be taken as some magical key that unravels the core truth about the issue—such a logic of the secret and solution is far from what conceptual work should do in this instance. But as a method, it facilitates how material can be structured in nonlinear yet informatively articulated ways. It demonstrates one mode of thought that maps concepts, historical sources, and contemporary discussions into an operational loop. As instances of “multi-dimensional thinking”18 as well as “loop-like series,”19 the soft montage of images forms a series where relations of combination stand out, but so do the gaps in between:

Farocki himself discusses his poetics under a different heading, not metaphor but montage. His montage takes two forms. One is as a sort of meta-commentary, traversing especially the early films like a steady murmur, repeating the need to “separate and join.” The other type of montage is embedded in the movement of the thought, as its structuring dynamic, but verbalised, if at all, only as the cut, the gap and what becomes visible “in-between.” Where film theorists speak of segmentation, Farocki (or his characters) discuss the difficulty of thinking things together at one level, while at another, making distinctions and keeping things apart. Only when the two levels are aligned, are the preconditions of new knowledge present: making connections on the basis of having taken something apart is thus where the rhetoric of metaphor meets the technique of filmic montage.20

I am not making a film, though; this is, after all, a book. But the method is still useful in answering the question: What forces compose the concept of the operational image as a particularly apt investigatory tool of contemporary culture, and how are operational images part of a broader landscape of epistemic, military, aesthetic, and data-driven institutions? The answer varies based on context, the connection between contexts, and sometimes what falls in-between. The loop and the recursive serialization that this book presented suggests that operational images can be mapped into a specific historical period of media theory and technological culture (not least since the 1980s to early 2000s as the period when a lot of influential theory from Flusser, Virilio, Kittler, and Baudrillard took shape), but it also resonates with the broader context of “operations” as they are featured in contemporary theory, such as “operative ontologies” (chapter 1). In addition, questions of data and invisuality (chapter 2), measurement and territory (chapter 3), operational aesthetics across contemporary institutions of science and data (chapter 4), and in urban and nonurban contexts of lidar and machine sensing (chapter 5) form the essential answers to the arguments and hypotheses posed at the beginning of the book: that the “operational image” is a term that helps to understand the hinge between data and (in)visuality, space and image, and temporality and image. In doing so, it becomes a methodological tool for investigations that proceed via a series of cross-disciplinary leaps: media theory in combination with art studies, in conjunction with architecture and critical data studies. And perhaps this is needless to state, but I will do so anyway: the particular loops and series in this book are not meant as the final word on the topic, but as (hopefully) helpful guidelines that deepen the often-mentioned point: we live, work, function, and imagine in the midst of such images that do not primarily represent but operate. We are, persistently, mapping techniques of reaching the phantomlike invisible that increasingly transforms into the platform-like invisual. Nothing less is the task of analyzing mediations in and beyond media studies proper, from visual to invisual cultures.