Month: April 2024

A number of people have asked if my keynote from the symposium on “Questioning History in the Age of AI” at Berkeley would be recorded. Unfortunately, it was not, so I am sharing the unedited text of the talk here. A revised version will be included in a volume that will include papers by the participants of the symposium, and edited by the organizers Julia Irwin, Johan Fredrikzon, and David Bates. Thanks to them for the invitation to present this work!

AI and the Future of (Media) History

Shane Denson (UC Berkeley, April 11, 2024)

The reason why “Media” is in parentheses in the title of my talk is because I want to suggest that artificial intelligence has potentially transformative effects that mark the entry into a new era or epoch of media history, but that this might also mark a change in history more broadly. This has to do, as I will argue, with a shift in the mediation of time—specifically a shift toward futurity and microtemporality—that distinguishes AI and associated media technologies from the past-orientation of media that came before. Those earlier media were integral not only to the mediation but to the constitution of history; they participated, that is, not only in historiography, but also in the material and existential conditions of history itself. This is especially true for the modern world. With AI—and I should specify that I’ll be thinking mostly about contemporary forms of machine learning and generative AI, especially in the realm of images—the conditions of history and of media history both (and in direct proportion to one another) undergo a significant, if still uncertain, transformation. Or so I will argue in this paper, which is divided into four sections.

I. From the Genetic Function of History to the Generative Functions of AI

I’ll begin by turning to a text by German media theorist Lorenz Engell on questions of the mediation of history and the historicity of media. Engell, who holds the Chair of Media Philosophy at the Bauhaus University in Weimar, Germany, is probably best known for his work on television—though in a very different register than that of most Anglophone TV studies. In any case, the paper I am concerned with here, while it does touch on television, is something quite different; it is titled “Die genetische Funktion des Historischen in der Geschichte der Bildmedien,” or: “The genetic function of the historical in the history of visual media.” I particularly want to foreground the “genetic” or generative dimension at stake in the text, which I will be connecting with generative AI. And especially in this connection I will prefer here to translate Bildmedien more literally as “image media,” as this will help me to foreground questions about the generated images at issue in contemporary AI, and to ask how (or even whether) they are in fact images, and in what sense the models behind them can be seen as generative (or generative of what). In a sense that I will return to later, generative AI models like DALL-E, Midjourney, and Stable Diffusion are indeed image media or mediators of images, but they have very little to do with human vision, and it is therefore questionable whether we should consider them visual media at all. Most significantly, though, these questions of generativity and their relation to human sensation provoke the central question: how can these media be considered historical and/or transformative of historicity?

Engell’s paper, which was published in 2001, is interesting for a number of reasons. Among others, it is very conscious of and concerned with its own moment in history (and media history in particular) at the outset of the 21^st century. It attempts to provide a systematic account of media historicity that would apply broadly across epochs, while focusing more specifically on image media from the 19^th to the beginning of the 21^st centuries. And it does so in order to enable observers of the mediated world to take stock of a major upheaval taking place at that time of mass digitalization, with respect to a new regime of image media that threatened or promised to escape the conditions of historicity that Engell’s systematic account describes. In this respect, at least, the article might be compared, superficially, to Kittler’s apocalyptic end of media history foretold at the outset of Gramophone Film Typewriter. But the question of epochal change is approached from a very different angle, and Engell is openly critical of Kittler and what he refers to as the Berlin School. In the meantime, after almost a quarter century has passed since its publication, Engell’s article itself has become something of a historical document. First, and somewhat anecdotally, it is of historical importance for me personally, as it played a pivotal role when I was writing my dissertation, around 15 years ago or so; returning to it now, it carries a sort of self-reflexive or self-historicizing significance, allowing me to take stock of the history or development of my own thinking. But the text is historical in another, more public, and perhaps less positive sense as well: that of being caught in the past of its once-present moment of generation; by this, I don’t mean to dismiss its importance, and indeed continued importance, but a text from 2001 simply could not have anticipated contemporary developments in machine learning and AI, for example. (On the other hand, however, anyone familiar with Germany today will recognize its apparently dated concerns with digitalization as still very current, in an almost comical way: if you watch the Tagesschau or read German newspapers, you will still hear journalists and politicians talking almost daily about Germany’s ongoing “digitalization” efforts, sounding very much like the discussions that were going on back in the 1990s.) Whether historical or contemporary (or both), Engell’s paper provides a useful framework for taking stock of AI’s impact on the question of history and media history, as I hope to show in the following.

Let me start, then, with a rough translation of the paper’s opening, which puts media at the heart of the historical world:

“The world is not the problem, but rather the attempt to solve it. The price of every solution, however, is the generation [Generierung] of further problems. Problems, solutions, and new problems that exact new attempts at solutions continue to engender one other [zeugen einander fort]. Thus, the world, at least the world of modernity, stands under the demand of uninterrupted change, self-revision. It doesn’t stand still but works [or operates] continuously on itself. World only exists now as world in motion [bewegte Welt]. The world attempts thereby to solve some of its problems by means of—ever new—media. In particular, it works by means of media on the problem of solution itself, namely the always necessary self-change and self-movement [Selbstveränderung und Selbstbewegung] of the world. Self-change presupposes self-perception and self-relation. Making these possible and generating them is the purpose of media [Sinn der Medien]. In media, the moveable, changeable world formulates itself, observes itself, steers and regulates itself, criticizes, devises, works on and revises itself. Media are accordingly meta-solutions and meta-problems, solutions that provide the possibility of solutions and the need for solutions, which is to say, precisely: new problems. As a precondition for the attempt to solve the problem, they are therefore also the precondition for the world to be the world—this world—in the first place.”

“One of the strategies associated with the solution of problems, and simultaneously with the generation of problems, is that of history. It executes the attempt to grasp, in a methodical way, the movements and changes of the world. The “Formation of the Historical World” [a reference to Wilhelm Dilthey’s Der Aufbau der historischen Welt in den Geisteswissenschaften of 1910] is a comprehensive and successful attempt at a solution, even if it henceforth generates [erzeugt] problems of historical research, historical understanding, historiography, and historical criticism, can cross over into the forgetting of history, or finally even end up as a problem itself. The historical world, like every world, is fundamentally all-embracing and inescapable. It historicizes everything, including itself and all of its own functions. It knows nothing unhistorical. This goes for media as well. In a historical world, they occur exactly insofar as they have and can have a history. Not only media themselves, also the history of media must therefore, in a reversal of familiar and established perspectives, be conceived as solution, not as problem.” (33)

With the focus here on worldhood and its media-technical constitution, along with the reference to the hermeneuticist Dilthey, we might expect at this point for Engell’s claims and connections to be elaborated in a phenomenological vein. He might invoke Martin Heidegger, for example, whose infamous tool analysis in Being and Time showed the worldhood of the world itself to be at stake in a simple hammer and its availability for use or its breakage, which reveals an endless play of reference and dependence, or what amounts to a kind of self-moving circulation that is at once material, symbolic, and social. The hammer, in its relation to the totality of equipment, resources, needs, and projects that it mobilizes and from which it is inextricable, is both the solution to a problem and the generator of new problems (if all you have is a hammer, everything looks like a nail, as the saying goes). In its existential import, it participates in the world’s self-change and serves, as Engell says of media, as a precondition for the world to be the world in the first place. If this is true for a simple hand tool, how much more momentous must be the impact of modern media technologies? As a second possible route towards explicating the role of media in constituting the world and its historical development, we might look to Jean-Paul Sartre, who early in his career elaborated on the role of technologies in formatting the world as what he called a “hodological space,” or a non-Euclidean environment that is “furrowed with paths and highways,” filled with obstacles like locked doors, but also transformed topologically by tools that extend perception across space and time. The time that the young Sartre had in mind was of course phenomenological time, and what he called the “always future hollow” that our projects aim at by means of the tools and technologies at hand. Later in his career, the analysis would be significantly revised. What appeared to be an individualistic perspective in the early work is transformed in Sartre’s encounter with Marxism into a more robustly social articulation of industrial-technological worldhood. Serialized production was now seen to format space, subjectivity, and collectivity in a similarly serialized and alienated manner. The temporal dimension is also radically expanded, as the struggle to overcome this alienation takes on world-historical proportions. Technologies, including media (Sartre discusses radio in particular), can now be seen as integral to the articulation of epochal change. Finally, Bernard Stiegler expands critically on Heidegger’s treatment of technology, emphasizing the role of monuments and media, in their guise as tertiary retentions, as conditions of inheriting—and thus also producing—the historical world. He, too, worried about the formatting of experience, the standardization of time, and the threat to history posed by new media technologies.

So, as I was saying we might expect Engell to develop his argument in one of these phenomenological directions. Well, he doesn’t. Instead, he follows Niklas Luhmann in developing a systems-theoretical account of medial self-historicization. But I think there is room, and need, in fact, for a dialogue between these perspectives, so I’ll be returning later to some of the threads suggested here. As I’ll argue, the phenomenological perspective is essential—both because of what it can and what it cannot do—in terms of coming to terms with the transformative effects of AI.

So back to Engell. As we have seen, history for him depends on a system’s self-perception and self-relation. Accordingly, it is not enough simply for something to happen, or to change; rather, history is constituted in acts of self-historicization, when a system takes stock, by way of media, of its own development, marking its transformation or evolution from a past state to a qualitatively different present. Media history, in particular, makes this process apparent; it can self-reflexively reveal “the historical world itself as a reflexive one” (34), giving insight into “the genesis and the transformation of the conditions of possibility of history” (34). Accordingly, we are dealing here with quasi-transcendental or “meta-historical” conditions, in Reinhart Koselleck’s sense, though expanded beyond the purview of conceptual history and focused on forms of mediation more generally. I won’t go into all of the finer details of Engell’s argument, but it should be apparent that the kinds of transformations at stake here cannot be cyclical or strictly repeatable, as this would not generate the kind of difference measured temporally and medially in the historical sphere. And so the argument concerns specifically modern history, also in alignment with Koselleck. The causes of these non-cyclical changes are describable in a variety of ways—either external or internal to the system, or eigendynamisch and morphogenetic, risky and unpredictable, as Engell elaborates. Engell goes on to trace how these various explanations of change influence and shape a number of media-historiographical approaches, including some that are only implicit in various strands of media theory. Most important, though, is the claim that history is generated by the system itself when it has developed media that allow it to access its past “out of sequence” (45)—that is, when it has developed a kind of random-access memory that allows for juxtapositions that foreground significant transformation. The developments or “events are obviously handled and marked in every case as ‘past,’ for example through dating” (45), but the relation or self-relation to the past always occurs in the present, and thus relies on “records, traces, symbolizations, traditions [Überlieferungen], archival materials, etc., or even just projections” (45). This is a theory, then, of the mediation of what Koselleck refers to as the “present past” (or gegenwärtige Vergangenheit) that is at stake in history.

And this mediation is especially complex when the history at issue is that of a medium or medial constellation. “The medium in question begins therefore at a certain stage in its evolution to make self-observations and to produce self-descriptions; it also realizes thereby self-symbolizations and self-distancing; it gains access to its own, past developmental states and at the same time acquires thereby the ability to put itself in a relation of alterity [Fremdverhältnis]. This situation presents itself for instance when established, functioning media are forced into self-historicization by media change, namely when other media constitute themselves as ‘new’ (and thus as evolutionarily and historically relatively unconditioned) in contrast with ‘old,’ existing media, which are thus required to historicize themselves” (47). In this relational field, the new medium tends to appear in a spectacular form, where “material performance and phatic use of the medium” (48) self-reflexively take center stage over and above any specific “content.” Here we can think of early film in its guise as a “cinema of attractions,” in Tom Gunning’s phrase, when the new medium appealed to viewers through what Neil Harris has called an “operational aesthetic,” or a dual form of address that splits viewers’ attention between the surface-level images and the underlying operations of the apparatus (whereby the illusionistic or representational powers of the medium stand in tension with it “material performance and phatic use”). The Lumière brothers are said to have projected a still image of their famous train before cranking it into motion by hand—thus enacting a powerful demonstration of the Cinématographe as re-animating dead photographic traces before the audience’s eyes in an act of historicization that marked one visual medium, photography, as inert while announcing the new medium, cinema, as a dynamic agent of history itself. And, of course, the documentary powers of the cinema would immediately be set to work in service of further historicizing the world more generally (think of the footage of the rubble in the wake of the great 1906 SF earthquake). Of course, novelty cannot, by definition, be sustained forever, and so the medium turned, as is common in a second stage of medial development, to the example of other media such as literature and the theater; the cinema became more narrative and stagey, before working out the more or less inconspicuous model of its third, classical phase, as embodied by Hollywood and its “invisible style” of continuity editing. Later, in a fourth, post-classical phase, self-reflexivity, self-ironization, and self-historicizing pastiche come to the fore, as cinema is crowded by and responds to newer media.

And while I would want to complicate the seeming linearity and neatness of Engell’s account of medial self-historicization, I think we can recognize each of these four stages or moments across a wide variety of media. But does this remain true for AI and other algorithmic media? Certainly, their novelty is introduced in typically spectacular fashion, and computational media seem to have perfected the “demo” mode as a means of hyping innovations and onboarding consumers. Meanwhile, as critical voices seek to deflate the hype, we already see AI models entering into an imitative mode, aping the photographic and cinematic media that provided the training data in the first place, or being used to produce Shakespearean sonnets. Who’s to say that generative AI won’t find a classical and unobtrusive mode, only to shift gears at some point (back) into a baroque self-reflexivity? But even if such a developmental trajectory is plausible, would history be mediated in the same way?

II. Cinematic Pasts and Post-Cinematic Futures

It’s important to note that what Engell refers to under the concept of “self-perception,” the system’s mediated self-observation of its past, is aligned with a particular conception of media: namely, the broadly “cinematic” conception that, for thinkers like Stiegler encompasses a variety of industrial-technical modes of recording, from photography to phonography to film and television. These are retentional and mnemotechnical media, media of “tertiary retention,” in Stiegler’s terminology, the very purpose of which is to record experience and make it available for playback at a later time. With such media at hand, the so-called self-perception or self-relation to the past becomes a relatively trivial matter. From a phenomenological perspective, however, these topological deformations of temporal flux pose a number of challenges, including for the production of the future—as Stiegler famously worries in the wake of realtime media, which threaten to collapse the difference between primary, secondary, and tertiary retentions and thus standardize or stifle protentional or future-oriented desire. 

Interestingly, Engell has very little to say about futurity, though certainly it was every bit as much at stake as past and present in those spectacular demonstrations of cinematic novelty and innovation. Take, for example, Georges Méliès’s Voyage dans la lune from 1902—an adaptation of a science-fiction story by Jules Verne, but not quite a science-fiction film, as that cinematic genre wouldn’t solidify for another half-century. The futuristic vision of the narrative serves here, as in most films of the time, to foreground the novelty of the medium. Compare, now, the film’s reappearance at the opening of the 1956 adaptation of another Jules Verne story, Around the World in 80 Days. Here Méliès’s film appears more or less explicitly as a historical document in a textbook case of the cinema’s self-historicization. As Engell’s model would predict, film had been forced into a defensive position by a newer medium, television, which was drawing the older medium’s audience away. Cinema responds, in this case, by demonstrating its ability to adapt, to update itself, and the demonstration is executed in a spectacular medial form: the grainy black and white images of Méliès’s film make way for lush color footage, and we can imagine red velvet curtains opening up from the boxy Academy ratio framing to a glorious widescreen. In this transition, Méliès’s primitive cannon-launched rocket is contrasted with a state-of-the-art NASA blast-off, thus aligning cinema itself with a cutting-edge exploratory vehicle.

These examples would seem to vindicate Engell’s relative lack of attention to futurity, as what’s at stake in both cases is the currency or up-to-date-ness of the medium. However, another way of looking at things is that these spectacles are designed to widen the gap between what Koselleck calls the “space of experience,” or the realm of the “present past” that I mentioned before, and the “horizon of expectation,” or the “presentified future” (gegenwärtigte Zukunft). That is, the demonstration of novelty is precisely geared to show that past experience, retained and recalled in the present, is inadequate for judging the future—a future which the medium is itself imagining and at least in part engineering. Koselleck’s terms, Erfahrungsraum and Erwartungshorizont are clearly modeled on the phenomenological concepts of retention and protention, but they are expanded to a societal and world-historical scale. The unhinging of the futural horizon of expectation from the space of past experience, which for Koselleck defines modernity and its mode of history, necessitates speculation, as opposed to certainty, about the future. In the examples I have given here, this is associated, at least aspirationally, with a sense of excitement about the future—whether the future of the medium, or the futuristic Space Age dawning outside the theater. 

Against this idea of the gap between experience and expectation, Anders Schinkel has argued that it is simply not possible to separate the two. Expectations necessarily arise on the basis of experience, which itself is acquired through the prospective filter of expectation that runs ahead of us to intercept the flow of time. This, I think, is consistent with phenomenological accounts of temporality, including those of Husserl, Heidegger, Sartre, and others. On the other hand, Koselleck’s diagnosis of a gap—that is, his assertion that past experience seems less and less a good predictor or regulator of what to expect in the future—seems to be genuinely illuminating with respect to relatively recent historical disorientations, so perhaps it’s simply a matter of scaling from the individual-phenomenological to the societal-historical processing of time; perhaps, that is, the latter does not operate quite like the former. Be that as it may, Schinkel proposes imagination as a third “category of history,” alongside Koselleck’s experience and expectation, suggesting that, while there can be no disjoining of those categories, it is through imagination that the precise relation between experience and expectation can vary historically. Personally, I am inclined to believe that this is more a restatement rather than an alternative to Koselleck’s model, but the idea of “imagination” as mediating between retentional experience and protentional expectation is useful—and it happens to accord in important respects with Heidegger’s phenomenological reading of Kant’s transcendental imagination as it appears in the first edition of the Critique of Pure Reason.

To return to Méliès’s film and its later recycling, we could say that it is precisely an imaginative relation to history that is at stake, an activation of the imagination as the realm of images that either are no more or are not yet, that are not actual in the present but remain as traces of the past or projections of the future. For Kant, “Imagination is the faculty of representing in intuition an object that is not itself present” (165), and this would seem to accord broadly with Koselleck’s presentified pasts and presentified futures. If history, or modern history, results from the tensions between these two poles or modes of presentification, then we can say that it turns precisely on tensions in the imaginative domain. In his phenomenological study of the imagination, Edward Casey associates imagination with possibility, multiplicity, and variation: “In the present context such multiplicity assumes the specific form of variability, that is, the mind’s freedom to vary itself indefinitely and without end.” For Don Ihde, there is a natural affinity between this power of variability and the multistable phenomena that he examines: from simple Necker cubes and duck-rabbits to the reversible relations we have with mediating technologies that allow us either to look through or to look at them. And it is precisely this multistability that’s at stake in the operational aesthetic: Méliès invites us to look through the screen, entering into the fictional world it mediates for us, but he also thereby foregrounds the spectacular operation of the medium, which draws our attention back out to the machinery responsible for producing the images. For the spectator, these imaginative tensions—tensions between looking through or looking at the medium—mediate and open onto the larger historical and media-historical questions broached by Koselleck and Engell. That is, the imagination might be seen as the mediator between the individual-phenomenological, the societal-historical, or the scalable-systemic perspectives that are in play here.

III. Algorithmic Temporalities and Artificial Schematisms

If the mediating function of the imagination goes some way toward reconciling the scalar shifts between phenomenological and historical temporalities, it seems less certain that it can accommodate the contemporary shifts in media towards microtemporality, futurity, and what I have theorized more broadly as phenomenological discorrelation, or the fact that contemporary algorithmic media operate outside of what Husserl called “the fundamental correlation between noesis and noema.” That is, computational imaging systems, including generative AI, operate at scales and speeds that categorically elude human perception. The subsymbolic nature of contemporary machine-learning algorithms means that their operations are not just incidentally black-boxed, but that they are categorically immune to perceptual or cognitive capture. As such, they do not serve the same retentional purposes as the cinematic media considered above; they do not offer objectal traces of the past in the same way—even if they take such traces en masse as their training data. Instead, they are generative in the sense that they are aligned more with the protentional pole of temporal experience. If self-historicization depended on self-perception, which it was the responsibility of retentional media to provide, what happens now to the imaginative interplay of presentified pasts and futures? 

Attempting to come to terms with these temporal shifts in my book Discorrelated Images, I considered Yuk Hui’s interesting provocation that what’s at stake in computational media, and smart systems in particular, is a shift from Stiegler’s “tertiary retention” to “tertiary protention,” or an exteriorized form of anticipation or expectation. Importantly, in the present context, Hui frames his argument in terms of imagination, arguing that predictive technologies come to shape our imaginations in their protentional dimensions. In my earlier engagement with Hui’s argument, I criticized what I took to be an equivocation between the subjective operations of imagining a future (for example, I am imagining drinking a beer after the conference right now) and the categorically asubjective modeling of the future by algorithms. I suggested that even human protentions can be either referential (as when I intend an image of that beer) or subreferential (in which case they concern the undetermined future of the just-about-to-come, that must be presupposed in order that I can continue speaking this sentence into an open future). These latter, subreferential protentions do not rise to the level of consciousness in the form of concrete expectations, and they are therefore not objectal in nature; rather, they underwrite the pre-perceptual and unthematized flux of temporal experience. Husserl writes: “[e]very primordially constitutive process is animated by protentions which voidly [leer] constitute and intercept [auffangen] what is coming, as such, in order to bring it to fulfillment” (1964, 76). If we want to ascribe a protentional dimension to algorithmic systems, I argued, it will have to be on the model of these empty protentional openings onto the future, not the determinate and referential expectations I might call to mind in imagination. I stand by this argument, but I do want to reconsider the role of imagination in our interface with AI and other algorithmic media, as I am now convinced that, once these corrections are made, Hui’s invocation of imagination might help us to make sense of the historical and media-historical implications of AI.

The problem, plainly, is that by shrinking the temporal circuit of mediation to the microtemporal feedback between the just-past and the still empty future just-about-to-come, algorithmic media seem to escape the broad historical horizons marked out by more referential-retentional media. (Engell already gestures towards such a problem with digital images back in 2001.) Algorithmic media therefore align themselves with a more foundational and pre-subjective level of self-modifying temporalization, such as that described by Kant under the heading of self-affection. This would seem to undercut imagination as a “faculty of representing in intuition an object that is not itself present” (165), and this for the simple reason that the non-referential nature of embodied and computational retentions and protentions cannot accommodate such fully-formed objectal representations. Importantly, though, Kant identifies self-affection with the synthesis of the productive imagination, which mediates between intuition and understanding and provides the schematized conditions of possibility for the referential experience of objects. It is beyond the scope of this paper to work out all the details of how this works—or even what it implies for conceptions of artificial intelligence—but this at least offers what I take to be the promising beginnings of a scalar model of imagination that would mediate between pre-subjective, subjective, and social levels of temporal experience. 

Moreover, by aligning the human and computational processing of time in this nonreferential and subsymbolic space, where AI’s latent space and indeterminate, pre-imagistic schematism converge, we get a better picture of how these processes might work on human imagination. Kant’s schemata are abstract and perspectiveless; they are indeterminate but determinable stereotypes that allow for the recognition and naming of objects and images. For example, Kant writes: “The concept ‘dog’ signifies a rule according to which my imagination can delineate the figure of a four-footed animal in a general manner, without limitation to any single determinate figure such as experience, or any possible image that I can represent in concreto, actually presents” (182-183). Such schemata are not seen but are the conditions of seeing; they are indeterminate with respect to point of view, but they enable recognition and impose a point of view. In an important sense, schematism determines subjectivity itself by “placing” the subject with respect to a schematized (or stereotyped) object. These are the pervasive micro-interpellations by which the world calls to us, and we turn to it, and in turning accept a subjective stance or persona. The alignment suggested above between the pre-subjective dimensions of human imagination and their computationally modeled counterparts should by now be quite concerning: if this model makes sense at all, then it seems we have hit on a mechanism by which so-called algorithmic bias can operate directly on our pre-personal interface with the world, pre-formatting or at least influencing the shape that our imaginations come to take in their more subjective and social-historical dimensions. There is a feedback loop here whereby socially constructed norms and stereotypes are injected into the preperceptual level of our realtime interface with algorithms, which in turn influence our subjective apprehensions of the world and our social interactions. Because these stereotyping forces are at work, microtemporally and protentionally, in filtering our lowest-level encounters with the world, they are able to get ahead of our subjective defenses and exert an unpredictable influence on our historical becoming.

IV. Imagining (Media) History with AI

Of course, what I’ve just said might all sound like science fiction or baseless speculation. So in order to bring things back down to earth, so to speak, I want to end by considering some artworks that shed light on how, concretely, our imaginations are at stake in AI as Bildmedien or image media, and how these media participate in and complicate the ongoing processes of self-historicization. 

A 2024 show at Gagosian Beverly Hills of works by Bennett Miller, generated with OpenAI’s diffusion-based text-to-image model DALL-E, is instructive in this context. Miller, an acclaimed film director nominated multiple times for Academy Awards, has produced a literally post-cinematic project in the wake of making an unreleased (perhaps never-to-be-released) documentary about AI and Silicon Valley’s efforts to innovate and capitalize in the field. In lieu of those cinematic images, which for legal or other reasons are inaccessible to us, Miller offers the viewer fourteen still images that seem to return, in an instance of what Richard Grusin calls “post-cinematic atavism,” to pre-cinematic forms of photography. The blurry sepia-toned images, printed in large-scale square formats (33.75” x 33.75” and 56” x 56”), are vaguely surreal—but in an understated mode very much at odds with the more spectacle-oriented psychedelic or otherworldly images that dominate AI’s marketing wing. One image depicts what seems to be a young girl with downcast (perhaps closed) eyes, standing in front of a buffalo who has apparently snuck up close behind her. In another, an empty chair occupies an empty stage, facing an audience of empty chairs. Other images, all of which are untitled, are less determinate. What looks like a human figure seems to be diving, but it is unclear where this indefinite body is diving (or falling) to or from. Everything is clouded in a haze or mist. In another image, I imagine I see another body falling, maybe into the water, but then I wonder if the background is not a rocky cliff, and the body perhaps not a body at all but a shadow cast by chance upon the rock. The hazy monochrome images challenge perception, as they oscillate between different thematic and formal configurations, between conflicting arrangements of figure and ground. The interplay serves less to emphasize the novelty of AI and more to bring it back into conversation with the many anonymous photographic images from which these synthetic ones have been generated. 

Above all, the encounter with the large-scale prints foregrounds the role of imagination in animating them—they are inert, and yet they are constantly transforming, always just eluding the determination of perception. Interestingly, this play of the imagination is significantly diminished when the encounter is mediated by a camera or screen (as here) rather than apprehended in the flesh. Standing in the gallery, my mind cycles through possibilities, but when I take my iPhone in hand to produce a reminder of my experience, I find the image considerably sharper, less blurry, more determinate—and thus less lively. Perhaps the AI in my phone’s camera has tried to compensate for the blurriness that Miller—no doubt with great effort—has induced his AI to generate. All of this is, after all, at odds with the operational principles of diffusion models, which start with pure noise and work recursively to un-blur an image, to make something crisp and definite for a human to see. Turning back to the print on the wall, I find that I see much more, in fact, but far less determinately. My embodied consciousness goes beyond the sensibly given, filling in and trying out pre-conceptual hypotheses (if such a thing can be said to exist), cycling through different ways of seeing. In such an encounter, I feel that my productive imagination has been liberated, if only briefly, from the stifling correlations and typifications upon which AI and its more mainstream products depend. Against a kind of computational “deathlife,” as Vivian Sobchack has put it, I discover in this indeterminate and paradoxical play of in/visuality the richer animacy of a post-cinematic imagination.

If Miller’s works utilize AI and its interface with imagination in order to complicate linear and progress-oriented self-historicizations of image media, American Artist’s video installation 2015, along with the accompanying app 1956/2054 suggest an alternative outcome. Here, schematization serves to solidify social stereotypes and to reiterate historical patterns over time. The video shows a police dashcam overlaid with a fictional heads-up display in which predictive algorithms dictate the driver’s route on the highway and through a predominantly Black and Hispanic neighborhood in Brooklyn. We see a schematic map, as well as animated navigational instructions, crime statistics, and other data, including what looks like a visualization of neural nets cycling through noise in order to isolate and determine actionable informational patterns. Though we never really see anything happen on the streets, we will occasionally see a pop-up message announcing “Crime Deterred.” The events, if they even exist, are left to viewers’ imaginations. The video, which has been shown at the Queens Museum and the Museum of Modern Art in New York (where I saw it in January of this year), takes its title, 2015, from the year when the New York Police Department began implementing predictive policing systems. Of course, the heads-up display we see here is an imaginative interface, not what police actually see. Installed across from low bleachers, where viewers watch as a group, the video enacts a strange interplay between perspectives (both individual and collective), between identifications (with the police officer subject or the barely seen and criminalized objects of their attention), and between imaginations (spectacular or dramatic imaginations of the crimes supposedly deterred, deflated by the fact that their supposed deterrence means that there is nothing to see here). In this way, the predictive formatting of schematized vision is made both the theme and the medium of the piece. 

Furthermore, low-level operations are shown to be in explicit dialogue with societal and high-level historical developments. 2015 is of obvious historical significance; the adoption of predictive policing follows in the wake of police killings of several unarmed Black men, including the July 17, 2014 killing of Eric Garner, which gave rise to nationwide protests and contributed to galvanizing the still young Black Lives Matter movement. This historical nexus is further historicized by the video’s exhibition alongside the app 1956/2054. The title of this piece refers to the year Philip K. Dick wrote his science-fiction novella “The Minority Report,” 1956, and to its futuristic setting in the year 2054. The story was made famous, of course, in Steven Spielberg’s 2002 film adaptation starring Tom Cruise as a detective in the Precrime division, where he uses futuristic interfaces to interpret the visions of three mutant “precogs” who can see the future; on this basis he works to arrest suspects before they can actually commit any crimes. (Interestingly, this film appears immediately in the wake of 9/11 and just prior to the protentionally-coded preemptive war launched against Iraq.) Against the spectacular display interfaces featured in Spielberg’s movie, American Artist’s works offer mundane, even boring interfaces: the video is anticlimactic and the app is just a smartphone app like any other, offering news and facts about predictive policing, notes, and other significant but unspectacular information. Wedged between the “past future” of 1956 and the “present future” of 2054, 2015 suggests both a conventional form of self-historicization, along the lines of Engell’s retentional-referential model, as well as one that has been transformed by predictive technologies, which lead the mundane, violent present seamlessly from the violence of a racialized past toward more of the same in the future. While the visualization thus retains a power of imaginative historicization, its encounter with prediction in the form of strict, cyclical repetition calls into question the trajectory, or the possibility, of meaningful development in an age of AI. But precisely by setting these multistable alternatives into motion, the work activates the imagination as the necessary mediator between prepersonal operations, subjective phenomenologies, and collective histories. It calls on us to imagine what is not present, either because it cannot be seen or because there is nothing to see in the first place, and to confront the predictive policing of the historical imagination itself. These are the stakes of the schematism today, which in the uneasy encounter between automation and autonomy will determine the future course of (media) history.