PAUL FEIGELFELD: Let’s start with the problem of genealogy. I think it is necessary to stray away from a genealogical media lineage of machine learning or, even worse, so-called artificial intelligence. In academia and the art world, we still tend to separate histories of science, technology, and media from the so-called natural world and its messiness, from sociopolitical processes, and so on. We view technological developments as something suprahistorical that somehow evolved on its own. However, I think it’s essential we move toward something we could call a mediaspora, describing a diverse, dispersed, asynchronous, transdisciplinary, and transcultural hodgepodge of theories, practices, and ideas of thinking. I like the term ideas of thinking because it points to the many imaginaries and fantasies that surround the supposedly rationalistic term intelligence, a term we know little about and that to me always seems very misleading. The only historical statement I ever felt comfortable putting forward from a cultural and media-historical perspective – but one we could also discuss in relation to art and art history – is that intelligence is and has always been artificial. That is, the process of transforming reality into knowledge always happens through art and artifice. Provocatively speaking, we aren’t capable of “naturally” thinking or knowing anything at all without something outside of our brains; without a form of model, tool, media; without absorbing and then secreting some form of external representation as part of the process. Furthermore, we have to take into account the many untold and silenced non-Western components of this mediaspora: the mathematical imports from the Arab world, including the very etymology of our beloved algorithm itself; the impact the Chinese language and writing system had on 17th-century thinking about the relation between thought, language, symbolic systems, and generative potentials; and so many other threads that are still invisible and subcutaneous. If we start from within a more fragile, fragmented genealogy of these ideas of thinking, where does it get us in terms of thinking about the other questions you raised?

JOSELIT: The ordered procedures of the algo­rithm, even when many are operating in parallel to one another, are not inspired – or literally in-spired, full of breath or spirit, to bring us back to your invocation of nature – but rather follow the same sets of steps mechanically or by rote. I agree with you that intelligence is an art, in the broadest sense – perhaps the sense of ­inspiration – and there are many ways of defining such art or artifice. One of the things that strikes me about ­generative AI tools is that they are predictive, meaning that they marshal an archive to create a statistically plausible output. Western art has generally been understood as representative, suggesting that there is a meaning behind the work of art that must be decoded (much of art history and contemporary criticism is involved in such decoding efforts). What if, instead, as many artists have done, we think of art in terms of its capacity as a model for generating rather than predicting or representing meanings – as a form of intelligence whose significance is in front of the artwork instead of behind it? I think this was the major interest of Conceptual Art and its progeny – to suggest that art generates objects rather than consists of them. Conceptual art was a type of generative AI before there was a technology of that name. But there’s another dimension to your critique of genealogy that I’d like to take up. Diachronic forms of telling history, as in genealogies, seem to have lost their persuasiveness. I think this results in a kind of dissemination or synchronicity of different temporalities in the present. One crude way to think of this is that we now all live in our own temporalities generated by the devices that give rhythm to our daily lives. But in media art, one sees this kind of temporal stutter through the persistent use of loops, multiple screens and/or channels, and permutational narratives. It sometimes feels like one is locked in a permanent present, and this, it seems to me, leads to problems in telling history. For me, some of the best ways to understand new technologies are through the aesthetic procedures developed by artists in advance of those media. Stan Douglas’s long-standing method of creating permutational variations on a single scene – for instance, in his video works – seems to me almost a proleptic account of AI’s predictive model. It is much more compelling in my view than the kind of melting-screensaver formalism of an artist like Anadol.

FEIGELFELD: Couldn’t we view prediction as the condition for the possibility of intelligence, as its predominant motivation? In terms of survival, but also in terms of a species evolving, what one needs is to know the future to some extent. We evolve from pattern recognizers to pattern generators; we recognize tides, orbiting planets, rhythms in nature, etc.; and we generate new patterns in the form of models, writing – more and more complex – right toward art and science. It starts with prophecy at a time when the correlation between reality, model, and prediction is still haphazard and then moves on to prognosis. We realize that a finite set and sufficient com­binatoric rules can generate seemingly infinite meaning – permutation. In the 17th century, calculus brought us into the infinitesimal and let us glimpse into what true generation could be, once we’d have the computational capacity. And recently, we have progressed from prediction to production; we generate reality rather than just predict and represent it. I’d like to throw you a term to discuss in relation to the question of “in front of or behind the artwork” and maybe come back to the question of history later: prompt l’œil.

JOSELIT: Prediction is certainly one mode of intelligence, but what if we set beside or against prediction the urge or desire for utopia, for that which cannot be predicted, which has certainly motivated much of modern art? Equally in politics, someone like Hannah Arendt would define the promise of politics as producing something new in a space of encounter and dialogue. For that matter, can a dialogue be entirely predicted? At the end of the day, could Chat GPT have been predicted? Is prediction a moment of execution after a moment or moments of invention? And of course, as many thinkers about bias in machine learning have pointed out, the kind of ­prediction one can make is based on the data one has scraped, which can lead to all kinds of harmful distortions. But to switch sides on myself, I do think the interest in science fiction, and especially in Octavia Butler in the United States, and what Saidiya Hartman has influentially called “critical fabulation” practiced by herself and other major African American thinkers, does correlate to an epistemology of prediction – but what is predicted in this work is founded on a different set of histories, political investments, literary techniques, and attitudes toward futurity. This leads me to your provocative neologism, prompt l’œil. I’d love for you to expand on this term, but here are some preliminary responses. First, I think that ­inevitably the genre or art of the prompt will continue to develop as a site of creativity (as I think the constitution of databases and archives will as well). But I want to put some pressure on l’œil because something that strikes me about the current state of machine learning – and correct me if I’m wrong – is that images must be translated into text on various registers before they can be generated as visual outputs. The prompt then is a very discursive form of visuality that places significant limits on what can be produced. This perpetual translation into text inhibits the spontaneity and invention of visual form.

FEIGELFELD: Maybe prediction was the wrong road to go down to begin with, and we should focus on what you talked about as generation and I as production. I like the term production more because it stresses the aspects of labor and power, which play such a huge role in this. Maybe it has long been much easier to simply produce realities than it is to predict them and (re)act accordingly – if you have the power over the means of ­production. Which brings us to the power of the prompt. I don’t think it’s us who are prompting, and I am doubtful there is much creativity in it, as long as the large language models are built by those who are building them now. Of course, art and artists have critically and creatively misappropriated the means of production and media time and again over millennia. However, the agency of the medium, its own artistic potential and autonomy, has significantly increased, perhaps even crossed a critical threshold. So maybe LLMs are preemptively prompting humans rather than the other way around, ushering us into incessant data production and paying us in rather generic AI slop, be it image, text, or sound. It really seems to me that quite often what we are getting back from our technologies these days – whether the supposed connection of social media or marvelously bad AI art – are surface effects, fascist-capitalist ­camouflage only thinly disguising a planetary infrastructure of computation that is predominantly occupied with extraction, expansion, exclusion, and totalitarianism. The inspiration for prompt l’œil is trompe l’œil after all, which is all about deceiving the eye, the gaze, the very idea of what is real and what is art and what is behind the canvas or curtain. I know I come across as a doomsayer here, and it has been said by others elsewhere before, but I do think we cannot and must not think about AI and art without stressing this. As for the eye, l’œil, you’re right, it’s ultimately all text – or rather, symbolic representations – and we’re not the first to realize this. Vilém Flusser, Friedrich Kittler, and many others in the late ’80s and the ’90s during the Iconic Turn have stressed that under the technological condition of mimetic Turing machines, the Real and the Imaginary – to drag Jacques Lacan into this – are subjugated under the order of the Symbolic. It’s all text, and all text is code, all code is numbers, all numbers are ultimately physical states of electricity in transistors. But this incredibly reductionist fantasy has to be counteracted with something, as you say, right? It can’t be turtles all the way down. I love that you bring up the concept of “critical fabulation” in this context. Do you think that through critical fabulation, speculative fiction, alternative futurings, and other techniques like these we might be able to reshape our technological condition? I do think that art has not only the potential but also the responsibility to shape technologies. But I am pessimistic as to the extent the tech industry has shifted its sights onto exactly that risk and is usurping creative production while turning creatives into the stress testers and bug hunters of their systems.

JOSELIT: It’s interesting to me that artists have historically absorbed the technological breakthroughs – revolutions even – of photography, film, and television/video with much more success than the internet and AI (at least so far). Maybe there’s something about ever-greater determination in these formats that makes them less amenable. In the early days of the internet, a DIY ethos still seemed possible, but that is certainly no longer the case. In your terms, these modes of production are hard to take over, whereas cameras, both still and moving, eventually became cheap and ubiquitous. The internet is everywhere, too – but its technological apparatus is not accessible to end users in the same way. Artists often break technologies instead of using them as intended; it might be harder to break the internet because it is so famously distributed. I’m not one for predicting the future, but given what seems to be a reorganization (correction?) of the art market and under conditions of widespread populism and censorship in the arts (at least in the United States), perhaps we’re at an inflection point in how art will function. But no one can predict the future – not even AI! Nevertheless, I’ll hazard a few thoughts. From what I’ve heard from artists in New York, the market for sculpture or experimental media is very limited right now – ­painting is the predominant commodity. So it seems doubtful to me that generative AI will become some kind of market sensation. But what does interest me more than the big, spectacular projects (such as Anadol’s) that have caught art-world attention are those artists, like Trevor Paglen, who are exploring how training sets are developed and what they might look like in advance of their processing through predictive algorithms. The means by which our lives are being regulated technologically (including how our information is scraped from us and archived by corporations) are largely hidden. By making such procedures visible, artists like Paglen shine light on apparatuses whose power lies in their invisibility. For a long time, people spoke of the spectacle – now I think power lies in its implosion and veiling. Invisibility is power.

FEIGELFELD: Speaking of power and how it manifests itself in the digital realm, it may be worth adding that the prompt also has a dictatorial aspect to it. I think an inherent problem of digital media is its command-control architecture, which inhibits actual collaboration. And while it seems we have reached a point of absolute and total wish fulfillment, and our machines will generate everything we didn’t even know we wished for, the political dictators and shadow dictators in the tech and finance world pretty much just determine what’s what while we still discuss what’s art and what’s not. I don’t know how to wrap this up really. It seems we are just getting started, and we haven’t even entered (or left?) the latent space. It seems clear to me that we need some form of infrastructural critique of art and an artistic critique of infrastructures, but it still begs the question of who cares. We can critically prompt these systems all we want, but it won’t change them. Do we build our own, or is it too late for that? I’m not even sure whether we are at a bifurcation or already long past it, in terms of a choice. An underlying problem, perhaps the most fundamental one, is the idea and ideology of the market that governs both the art world and technology, or even reality as a whole: a seemingly inescapable and ultimate post-political and post-historical logic that, we could even argue, found its first total manifestation in the invention of art as a commodity or tradeable entity around which a largely unregulated, mystical system of valorization was built. In this respect, Satoshi Nakamoto’s bitcoin white paper makes perfect sense as Conceptual Art, and generative AI ultimately aims at nothing less than a completely and totally self-regulated economic system of free-floating value creation. But self-regulation has been a ruse at least since ideas of cybernetics and ecology hallucinated a “natural system” in perfect equilibrium, harmony, and homeostasis. The regulation is done by dictators and CEOs, and while large parts of the art market contribute to producing the cultural camouflage, there are few people who critically engage with that system on an artistic and theoretical level. There needs to be more work like this, and in order for it to be developed, a cultural and educational shift is necessary, as well as sustainable critique. We are witnessing what – in the context of machine learning – is called a model collapse, or what we could see as a form of ­epistemological incest: Self-referentiality becomes a huge issue when all is generated and trained on other generated data; it ultimately leads to degeneration. This is true for research ecosystems that draw funding on the basis of citation indices, so the preprint databases are secretly flooded with fictitious papers, citing made-up sources by made-up authors, just to spike the numbers. Universities produce generated term papers and generated corrections and reports. This kind of decoupling of epistemic and artistic production and critique is an interesting thing to look at, but it cannot become ­ubiquitous reality. We need critical participation and care, and for those to be more than good-looking words in curatorial statements in galleries. We need to counteract “intelligentrification.”

JOSELIT: One thing that art can do is to slow down technological procedures and explore or ­reorient their affective charge. Generative AI has a very particular out-of-focus, flowing, morphing ­aesthetic. It wants to lull us to sleep, while we really should wake up.