# The Media-Dirty Sublime: AI, Loser Images, and the Ethics of Embracing the Void // Joanna Zylinska > Source: > Published: 2026-08-22 14:17:07+00:00 This is an interview with Joanna Zylinska.[ Joanna Zylinska](http://www.joannazylinska.net/) is a writer, artist, curator and Professor of Media Philosophy + Critical Digital Practice at King’s College London, where she directs the Centre for the Ecologies of Attention and Perception. An advocate of ‘radical open-access’, she is a co-Director of Open Humanities Press and an editor of its MEDIA : ART : WRITE : NOW book series. Zylinska is an author of a number of books – including [(MIT Press, 2023),](https://mitpress.mit.edu/9780262546836/the-perception-machine/) *The Perception Machine: Our Photographic Future Between the Eye and AI*[(Open Humanities Press, 2020) and](http://www.openhumanitiespress.org/books/titles/ai-art/) *AI Art: Machine Visions and Warped Dreams*[(University of Minnesota Press, 2018). Her own art practice involves experimenting with different kinds of image-based media. Zylinska is currently researching perception and cognition as boundary zones between human and machine intelligence, while examining various narratives of collapse, be it on the level of AI models or planetary systems.](https://manifold.umn.edu/projects/the-end-of-man) *The End of Man: A Feminist**Counterapocalypse*** DIFFRACTIONS**: You have described your work as a “hybrid mode of enquiry” that combines philosophy with art. What do you think your artistic practice unlocks or renders legible especially in relation to grasping the unfolding of generative AI, as well as our urgent entanglement to the planetary crisis? **Joanna Zylinska: **Artistic practice activates a different sensibility; a new way of arranging both concepts and forms of perception. It allows us to see, sense and compose the world otherwise, beyond the predominantly grammatological mode of philosophical discourse or the explanatory protocols of scientific inquiry. It is not that art provides access to some ineffable truth unavailable to philosophy or science. Rather, it produces different conditions for thinking, which I am trying to avail myself of. Art also permits a greater degree of playfulness and experimentation. Having been trained in the continental philosophical tradition shaped by thinkers such as Jacques Derrida and Gilles Deleuze, I have always tried to smuggle some of this experimentalism into my theoretical work too. But artistic practice lets me push things further: to work through image, sound, atmosphere and affect without having to translate everything immediately into an argument. In projects accompanying books such as *Nonhuman Photography* (2017) and *The Perception Machine* (2023), practice has therefore become for me a way of thinking with and through images rather than simply producing illustrations of previously developed ideas. This mode of working can temporarily rearrange the perceptual field, allowing machinic agencies and ecological relations that normally remain in the background to come into view. In the face of large-scale issues such as environmental challenges posed by AI or global threats to democracy coupled with encroaching fascism, art can create cognitive and sensory shortcuts through complexity, opening up spaces in which we can experience relations and scales that are difficult to grasp propositionally. It can also create conditions for imagining those relations differently. *The Perception Machine: Our Photographic Future Between the Eye and AI*, MIT Press, 2024. **DIFF: You’ve translated Lem’s Summa Technologiae. How did the process of translating this text or inhabiting his language and visions fundamentally reshape your own thinking about technology, evolution, and the future? Did you find yourself becoming a different kind of philosopher through his work?** **JZ: **I came to the translation of Stanisław Lem’s *Summa* *Technologiae*, which appeared in English in 2013, already as a scholar of philosophy, technology and media. But the experience of translating this extraordinary book alongside my academic work – a process that took around two years – was genuinely transformative. Translation requires you to inhabit somebody else’s conceptual machinery very intimately: to follow not just their arguments but also their rhythms, provocations, jokes and speculative leaps. What particularly stayed with me was Lem’s understanding of the parallelism between biological and technical evolutions. It helped me to further cut the human down to size on a philosophical plane: to see humanity not as the endpoint or privileged agent of technological development but as a temporary participant in much longer processes of transformation. This perspective became important for my subsequent work on nonhuman perception, photography and AI developed in *Nonhuman Photography* and *The End of Man: A Feminist* *Counterapocalypse* (2018). I was also struck by how extraordinarily prescient Lem was about developments such as artificial intelligence and virtual reality (his “phantomatics”) without ever turning into a straightforward technological prophet. I loved his ironic, playful and sometimes deliberately impetuous use of language, combined with an enormous ambition to say something about the human condition in relation to technology. Perhaps most importantly, Lem encouraged me to think technologically across different temporal scales and geographical planes: from biological evolution and deep time to Greek mathematics, early Chinese technologies and speculative futures. So yes, translating *Summa* probably did make me a somewhat different philosopher: more willing to move between rigorous argument and speculation, and even more suspicious than before of the human as the natural measure of things. **DIFF: In your book** ** The Perception Machine, you** connect your concept of “loser images” to Hito Steyerl’s “poor images” and the “avant-garde of the weak.” How do you see “loser images” as a form of “ethical piracy” that challenges the logic of “planetary extractivism” and the seemingly inevitable “upgrade culture” of both machines and humans? Is there a risk that embracing “failure” and “weakness” could be seen as apolitical, and how do you argue for its efficacy? **JZ: **There is a recurring masculinist posture in critical theory that adopts the language of combat, resistance, intervention and victory, while actually retaining a very weak efficacy beyond its own intellectual constituency. With the idea of “loser images,” I was trying to challenge the heroism that often accompanies theoretical responses to a crisis, whether we are talking about visual representation or political representation. I drew on Ewa Majewska’s concept of the “avant-garde of the weak,” as well as on Hito Steyerl’s “poor image,” to ask what a less heroic politics of images might look like. The “loser image” does not try to outcompete dominant visual regimes on their own terms. It embraces low resolution, obsolescence, recycling, copying, degradation and misuse as ways of resisting both upgrade culture and the extractive logic that demands ever more powerful machines, ever cleaner images and ever more optimised humans. This is where I locate its potential as a form of “ethical piracy”: not piracy as libertarian appropriation, but as a tactical redistribution of images and technologies outside their prescribed circuits of value. There is, of course, a risk that celebrating weakness or failure can become politically complacent. But for me weakness does not amount to resignation. I am rather interested in it as a different mode of efficacy: more ironic, less self-aggrandising, more horizontal and more capable of forming alliances among those excluded from the gung-ho rhetoric of theoretical militancy. The “loser image” may fail to win, but it can still refuse the very terms of the competition in which victory was meant to be achieved. **DIFF: **Furthermore, you propose “eco-eco-punk” as a mode of ethical and aesthetic intervention that sees the world as “always already media-dirty.” This concept challenges the “singularity of the rouge punk hero-savior” and embraces a “plural and entangled ontology.” In what way does this recognition of our “media-dirty” existence offer a more constructive path forward than the dominant affects of “gloom and doom” or “guilt and shame” that permeate around the discourse of the Anthropocene. **JZ:** My concept of eco-eco-punk retains something of punk’s unruliness and defiance while moving beyond the singular, angry heroism of the male persona who stands outside society, feels at odds with its norms and structures, and contests the future on his – because it is usually *his* – own terms. I was borrowing here from the music, fashion and aesthetics that were very important to me in my formative years. But I wanted to open that punk sensibility onto the environmental and economic crises of the present moment, while going beyond mere contestation. The world we are at odds with is itself fractured: its ecologies have suffered forms of injury, many of them human-caused, that demand acts of repair. At the same time, ecological fracture is inseparable from economic, mental, cognitive and corporeal forms of impoverishment and vulnerability. Repair or restauration can’t therefore come from a singular hero (or anti-hero) positioned above or outside the world. It has to emerge from diverse bodies and minds acting together – from what we might, despite the overuse of the term, still cautiously call our human-nonhuman entanglements. This sensibility also moves beyond the despair of punk’s “no future.” The future is neither something simply to predict nor something already owned by capital or rejected by rebels. It remains something to be made; never fully under our control but not fully predetermined either. **DIFF:** In your collaborative essay *Image thinking after artificial intelligence* with Yanai Toister, you call for Visual Culture scholars to move beyond interpreting images to interrogating “the infrastructural conditions” like “dataset provenance” and “parameter tuning.” This requires a significant methodological shift. How do you see your scholarly work also transforming and in light of this new kind of Visual Culture scholarship being practically practiced and taught? Is it a matter of integrating “digital humanities methods,” or does it demand an entirely new critical vocabulary – perhaps one that borrows from software studies, platform studies, and even computer science – to make the “in-visual” substrate of image thinking legible to humanists? **JZ:** The current state of knowledge requires humanities scholars to draw on a much wider range of methods, disciplines and forms of expertise than perhaps we have traditionally been comfortable with. Inevitably, that requires some upskilling. The tech industry relies heavily on smoke-and-mirrors effects: bombastic pronouncements about machine learning are veiled in an existential language of apocalypse while simultaneously making hyperbolic promises to investors. Humanities scholars are actually very well equipped to read such prophecies, metaphors and finalist narratives critically – and cut through them with satire and wit. But we can only do that effectively if we also understand something about how the technology works. We should not be paralysed by claims that even programmers do not fully understand what AI models are doing. We may not understand every internal operation of a model, but we can still understand the statistical principles on which these systems are based, how latent spaces are constituted, how tokens are distributed and how ambiguity becomes eliminated through probabilistic decision-making. My own work is changing in precisely this direction. In writing recently about model collapse, for example, I had to engage directly with computer science papers in which the phenomenon was first proposed, developed and contested. I actually went to a secondary school with a strong mathematics and physics orientation before choosing the humanities, so perhaps I am less frightened of crossing that boundary. But I think humanities scholars generally need to become less afraid of mathematics and computation, while remaining confident about what our own methods – critical reading, conceptual analysis, attention to metaphor, history and ideology – can contribute. I am actually concerned by many humanities critiques of AI that are politically forceful but technically just wrong. So this cannot be a matter of importing a few “digital humanities methods” and considering the job done. It requires collaboration across disciplines and a new critical vocabulary capable of moving between code, infrastructure, visuality, political economy and culture. Digital Humanities, at its best, and certainly in the way we try to practise it in my department at King’s, can provide exactly such a space of encounter between different ways of knowing. **DIFF: **Can you elaborate on how you employ the concept of “cognitive hacking” in your recent publication as a neutral or even generative description of how AI recalibrates human perception, contrasting it with the negative connotations of cyberattacks. Yet, if this hacking is “largely involuntary” and operates “imperceptibly,” what are the implications? Do you think also there is something distinctly as a “machinic imaginary” as Conrad Moriarty-Cole also suggests? **JZ:** With “cognitive hacking,” Yanai Toister and I wanted to reclaim hacking from its predominantly negative association with cyberattack or intrusion. We use this term more neutrally, and potentially more generatively, to describe a reconfiguration of an existing system that produces outcomes which are not entirely predictable or controllable. In this case, the system being reconfigured is that of human perception, imagination and cognition. AI models can be said to function as a kind of mirror for us. Like psychoanalysis in the early twentieth century, or Darwinian evolutionary theory earlier, they challenge some deeply held assumptions about what the human is. Generative models show us how much of what we think of as our unique expression is already structured by patterns, repetitions, cultural conventions, clichés and probabilities. In that sense they can teach us something about ourselves. But the mirror metaphor is not entirely sufficient, because AI is not simply *reflecting* us. We are entering into a feedback loop with these systems. We put language and images into them; they statistically reorganise those materials and return them to us; we then absorb their outputs and modify our subsequent ways of speaking, writing, seeing and imagining. We can already detect superficial versions of this in recognisably ChatGPT-like styles of email writing or public communication. But cognitive hacking goes deeper than stylistic mimicry. It raises questions about what it means to put words together, to make connections, to produce meaning and to experience language as something we both inherit and use for self-expression. The fact that much of this happens imperceptibly and involuntarily is therefore important. Cognitive hacking is not necessarily something *done to us* by an autonomous machine. It occurs through countless mundane encounters in which human and computational operations become folded into one another. That obviously raises political questions about who builds these systems, which cultural patterns they amplify and whose interests their infrastructures serve. There is a real danger of market-driven sameness. Yet I don’t want the concept to be purely dystopian. Cognitive hacking may also open up unfamiliar connections and expose the limits of our existing habits of thought. AI is therefore neither just a mirror nor an external intelligence arriving from elsewhere. It is becoming one of the environments through which human cognition is being reflected back to itself – and, in the process, altered. **DIFF: **Moreover, you provocatively suggest that the panic over AI imagery reveals a “fear of the ultimate void at the core of our own thinking.” Is this a productive moment of philosophical crisis? Does the encounter with an AI that can “image thought” force a necessary and humbling reconsideration of human exceptionalism, or does it risk leading to an acceptance that our creativity was always just a form of probabilistic recombination? How does your concept of “image thinking” help to navigate a species-wide existential anxiety without romanticizing the human or succumbing to resignation? **JZ:** The “void” I am pointing to is not a nihilistic emptiness so much as the possibility that there is no sovereign core of the human from which thought and creativity emanate: no soul, pure interiority or stable essence that guarantees our uniqueness. The anxiety provoked by generative AI perhaps comes partly from seeing aspects of our own thinking externalised. In that sense AI provides another challenge to human self-understanding. But recognising a statistical dimension to thought does not mean accepting a crude cognitive reductionism in which creativity is *nothing but* probabilistic recombination. Here I am thinking with Vilém Flusser. We always operate from within apparatuses – technical, social, cultural, ideological – but we are not simply determined by them. The human is also an information processor capable of becoming, in Flusser’s terminology, an *in-former*: someone who gives form to information, plays with the apparatus and sometimes breaks or redirects its circuits. That element of play is very important to me. Humans are capable of holding humour and seriousness, irony and argument, pleasure and critique in mind simultaneously. We often bend and twist language, by which I mean words, images and other symbolic forms, and we can draw intellectual and aesthetic pleasure from doing this. Creativity may lie precisely in this willingness to introduce and embrace the glitch, to make connections that are not merely optimising the system but actually getting it to do something different, and to pursue affects such as curiosity, delight or surprise. Catherine Malabou’s concept of plasticity is useful here too. Thought is not simply an execution of a pre-existing programme. It involves both receiving form and giving form, while retaining the possibility of transformation. Perhaps thought happens most interestingly precisely where ambiguity enters the system, where several possible meanings or registers remain in play rather than being resolved immediately into a decision. This is where “image thinking” becomes useful. AI-generated images do not simply illustrate thoughts that already exist; within the recursive human-machine loop, concepts and images modify one another. The encounter can therefore expose the patterned, machinic dimensions of our cognition without requiring us to conclude that human brains *are* computers. Computational models can make certain aspects of thinking newly perceptible while remaining only models. So I see the present moment as potentially humbling but not despair-inducing. We can relinquish the fantasy of human exceptionalism without giving up on alterity. Creativity does not need to originate in some mysterious interior essence in order to matter. It can emerge relationally, through embodied and historically situated acts of differentiation, play and transformation, even when those acts take place inside apparatuses that we never fully control. **DIFF:** You draw a lineage from early cinema’s animist impulses to contemporary AI installations, but cinema’s animism was largely about capturing life, recording the pre-existing world. Generative AI, by contrast, synthesizes life-likeness from statistical patterns, producing images without referents. Does this shift from mimesis to synthesis fundamentally alter the nature of animist experience? When we feel “moved” by a generative image of a flower that has never existed, what exactly are we being moved by? Is this a new category of animism, or a simulation of animism that potentially forecloses genuine encounter with nonhuman otherness? **JZ:** What I was drawing attention to in my recent work on bio-AI and data animism was not a distinction between cinema as somehow representing or capturing life and AI as merely simulating it. For me, “lifeness” does not reside straightforwardly in the image. It happens relationally, in the encounter between a human body and a technical apparatus. Whether we are watching a cinematic image or encountering an AI installation, there is a movement of life that becomes perceptible *within us*. We become aware of our own animation, our own vitality, but also experience a momentary estrangement from it. There is a distancing in which life appears to us as something strange, mediated and technical. This is where the philosophy of technicity developed by Bernard Stiegler, drawing on Derrida and Simondon, becomes important. The human is never simply opposed to technology: we are technical beings from the outset. But that condition is not necessarily available to consciousness. Certain aesthetic encounters can make it momentarily sensible. In the early twentieth century cinema was one such space: here my argument also draws on Bergson and Deleuze and their understanding of life through movement, also as shown on screen. AI creates a new iteration of this condition. A synthetic flower without a photographic referent can still move us because what is activated is not our recognition of a once-existing object but a complex relation between perception, memory, expectation and the technical production of lifeness. This is precisely what I mean by “bio-AI.” Yet bio-AI cannot be separated from what I call necro-AI. Drawing here on Foucault’s biopolitics and Achille Mbembe’s necropolitics, I want to keep in view the destructive, extractive and fracturing dimensions of AI technologies alongside their capacity to generate experiences of animation and vitality. Data animism is therefore not a celebration of machines becoming alive. It is rather an ethical and aesthetic way of attending to how life and nonlife are being newly configured through our technological relations. *Neuromatic*, 2020. **DIFF: **You propose that “ethical data animism” refuses to reproduce “necropolitical logics” and instead foregrounds “a politics that no longer necessarily rests upon difference or alterity but instead on a certain idea of the kindred and the in-common.” Achille Mbembe’s own analysis suggests that these are precisely the terms that have been used to justify violence – the “in-common” has historically been defined through exclusion. How does your proposal avoid replicating this pattern? What specific political practices or demands does ethical data animism entail beyond aesthetic experience and critical reflection? JZ: I am completely aware that any notion of community carries its own peril. Communities are boundary-making formations: they establish who belongs, but in doing so they can also determine who needs to be kept outside. We see the violent consequences of this very clearly today in political discourse around migration and in the return of explicit racism within far-right movements across the world. So when I speak about the “in-common,” I am not proposing a harmonious community in which difference disappears while all forms or dissensus are magically resolved. I come to this idea precisely *after* engaging with thinkers such as Mbembe, and with an awareness of the violence that can accompany belonging. For me, being-in-common is therefore not a once-achieved condition but a task. It can be articulated by a set of questions which are actually demands: How do we make worlds together after coming to terms with the situation of alterity without stopping at just that? How do we build kinship, solidarity or collective capacity in situations where our relations and selfhoods are already fractured? The recognition of the necropolitical dimension of social and technological life is crucial, but it cannot become an absolution from trying to act in configurations of “more than one.” And this cannot simply be *my* political prescription. I am articulating a desire, and perhaps also a responsibility, to keep attempting forms of being-with, while knowing that communities can be pernicious, exclusionary and violent as well as nourishing, welcoming and restorative. I am therefore somewhat suspicious of the increasingly ubiquitous language of “care” in the humanities. Care sounds unquestionably good, but unless an ethics of care is attached to political organisation, institutional structures and actual policy, it risks becoming an innocuous moral vocabulary that is unable to enact change. This is where ethical data animism becomes more concrete. Data capture and classification are themselves world-making practices: they decide what becomes visible, what counts, what is excluded and how relations are organised. Ethical data animism therefore asks whether data practices can be redirected away from extraction, surveillance and optimisation towards sustaining shared worlds. The work of my King’s colleague Jonathan Gray is instructive here. In *Public Data Cultures*, he examines ways of practising data differently: using it to build shared understanding and solidarity, document what has been rendered invisible, support communities and engage with environmental problems. For me, then, ethical data animism is not about pretending that data or AI are benevolent or “alive.” It rather means building connections between human and nonhuman agents, technologies, institutions and networks in ways that sustain rather than diminish lifeness. The politics lies precisely in continually keeping those arrangements open, resisting the moment when the in-common hardens too quickly into another closed “us.” **DIFF: **Where does your work turn to now? What new technologies, questions, or practices are you currently engaging with as you continue to develop? **JZ:** At the moment, much of my work is converging around the concept of collapse. Collapse has become an extraordinarily expansive term: we encounter it in relation to climate systems, political institutions, economies, social infrastructures and, more recently, AI, through the technical phenomenon of “model collapse.” I am currently looking at what happens when all these different meanings begin to bleed into one another. But I am less interested in simply announcing that things *are* collapsing than in asking what the discourse of collapse does. What kinds of subjects, affects and political imaginaries does it produce? Collapse can name very real processes of destruction and systemic failure, but it can also become a narrative machine that tells us the future has already been decided. In some collapse communities, for example, the insistence on “facing reality” can slide into a strange form of epistemic and emotional closure: one knows what is coming, alternatives disappear, while preparedness replaces political imagination. I am exploring these questions through both theory and artistic practice. I have recently made a short film called */collapse*, which combines archival footage of historical and contemporary crises with AI-generated fragments, including reconstructed personal memories. Its soundtrack emerged from a cut-up of material gathered while reading the Reddit forum r/collapse. Fast-paced with the intention of working just at the threshold of sensory overload, the film examines collapse as both narrative and affect, while looking into the automation of our emotional responses under conditions of increasing global anxiety. Alongside this, I am engaging with collapse on a theoretical level, through a series of academic texts. What interests me ultimately is whether collapse can be made into something other than a terminal diagnosis: a concept capacious enough to register genuine damage while also exposing the narratives through which inevitability is produced. Perhaps the task today, for all of us, is not to prevent every form of collapse, or romanticise ruin, but to ask what can still be recomposed, rerouted and made otherwise from within it. # REFERENCES Deleuze, G. (1986). *Cinema 1: The movement-image*. University of Minnesota Press. Flusser, V. (1983). *Towards a philosophy of photography*. European Photography. Foucault, M. (1990). *The history of sexuality, Vol. 1: An introduction*. Vintage Books. Gray, J. W. Y. (2025). *Public data cultures*. Polity. Lem, S. (2013). *Summa technologiae* (J. Zylinska, Trans.). University of Minnesota Press. (Original work published 1964) Majewska, E. (2016). *On weakness, solidarity, and non-heroic resistance: The weak avant-garde – A feminist analysis* [Lecture]. Museum of Modern Art, Warsaw. [https://archiwum.artmuseum.pl/en/doc/video-muzeum-otwarte-o-slabosci-solidarnosci-i-nie-heroicznym]. Malabou, C. (2008). *What should we do with our brain?* (S. Rand, trans.) Fordham University Press. Mbembe, A. (2019). *Necropolitics*. Duke University Press. Steyerl, H. (2009). “In defense of the poor image.” *e-flux Journal*, 10. Stiegler, B. (1998). *Technics and time, 1: The fault of Epimetheus*. Stanford University Press. Zylinska, J. (2017). *Nonhuman photography*. The MIT Press. Zylinska, J. (2018). *The end of man: A feminist counterapocalypse*. University of Minnesota Press. Zylinska, J. (2020). *AI art: Machine visions and warped dreams*. Open Humanities Press. Zylinska, J. (2023). *The perception machine: Our photographic future between the eye and AI*. The MIT Press. Zylinska, J., & Toister, Y. (2026). “Image thinking after artificial intelligence.” *Journal of Visual Culture*, 24(3), 431–451. [https://doi.org/10.1177/14704129251405715](https://doi.org/10.1177/14704129251405715).