Peli Grietzer is a philosopher of literature who received his PhD from Harvard University in 2017. His dissertation used machine learning theory to study forms of ambient meaning, like moods and vibes, in modernist and avant-garde aesthetics. Part of it was published as ‘A Theory of Vibe’ in Glass Bead Journal in 2017.1 He’s currently writing a book called Big Mood: A Transcendental-Computational Essay on Art. He spoke with Effects editor Matt Rickard over Zoom back in March. The conversation has been lightly edited for length and clarity.
***
EFFECTS: What’s a vibe?
PELI: A vibe is an intuitive representation of a language that is suitable for the compressed representation of a field of phenomena bound by a structural affinity that expresses a shared generative process.
Say you have a field of phenomena, an assemblage of some billions and billions of possible objects or possible events, generated through different settings of the same type of mechanism, you can think of how all dogs are generated through different settings in the epigenetic space of possible dog DNA. When a shared originating process leads to a structural affinity—when I say ‘structural affinity’ I mean when two things are easy to compare and contrast, not necessarily when they are similar, but when it’s meaningful to talk about their similarities and differences—neural network processes and especially latent variable models, like autoencoders or GANs, generative adversarial networks, can easily learn a coding system that very compressedly represents each individual object in the multitude of objects and maps their relations to one another. So, for example, if you change the code in some linear way, it will make a face look angrier and angrier or younger and younger. And this code system is much better at representing phenomena from the assemblage or manifold of phenomena that it’s trained on than it is at representing phenomena from outside the training distribution. If you tried to represent an object outside its possibilities, it would take a guess at which vibe-coherent object it’s most similar to and distort it into that object. What we call a vibe is our intuitive sense of there being a compact language that can generate phenomena bound by a structural affinity, which traces back to some unity in the original production process.
EFFECTS: So—just to make sure I’m following—when we perceive a vibe, what we’re perceiving is a generative structure that the vibe-coherent objects have in common?
PELI: When things are vibe coherent, they’re commeasurable, in the sense that you can say that if one thing changes into another thing within the same vibe, you can describe the nature of the change, whereas if one thing changes into another thing across vibes, then you have to compare the vibes. When things are vibe coherent, you can compare them with one another directly.
The fact that vibe-coherent things are modeled by compressed representation is very important. Humans have very limited memory. When objects are vibe coherent, I can remember all these things at once, I can remember very large structures composed of these things, whereas if you bring in elements that are not vibe coherent, I won’t be able to remember everything because I can’t produce a compressed representation using the same language or coding schema.
The code language produced when you grasp something vibewise—when the process is in some sense analogous to an autoencoder or a GAN—is a kind of weak schema of the generative factors, or DNA, of the phenomena we’re trying to model. The goal when you use an autoencoder or GAN is to reverse engineer the factors of production of the phenomena that you’re modeling. It’s a compressed representation of a field of phenomena that has a loose generative structure; the better the representation, the better it approximates properties of the generative structure. Vibe-grasping a field of phenomena is a first step towards, a condition for, and in some sense an initial approximation of, the system that generates the field of phenomena.
Fredric Jameson, even though you can think of him as, like, the biggest enemy of vibes, says that there was a fall at some point after the era of Balzac where literature lost its ability to grasp the causal structure of social-material reality, and that, as we move into modernism and postmodernism, writers were content to substitute for that grasp of the causal structure of social-material reality sort of vague affinities or lists, like in Knausgård. He would say that what literature should do is help us grasp how different factors interact to produce the world of experience, and that it should do this in a way that is itself experientially graspable. Whatever sort of unity you find in a list isn’t going to get you very far. And I guess, sure, maybe grasping the vibe coherence of some phenomena isn’t enough for us to see the work as a product of historical-material forces either. But I think that grasping vibewise might still be a precondition for that. You have to grasp the vibe coherence of a field of phenomena before it’s viable to look for more finely structured causal explanations.
EFFECTS: I’m glad you brought up Jameson. For most people in the Jamesonian tradition, the interest in totality and cognitive mapping and so on is put in the service of critique. That’s not a language you use. What’s the relation between critique and vibe-grasping, or between the ‘structure’ disclosed through critique and the ‘structure’ disclosed through vibe-grasping?
PELI: A nice heuristic for a lot of post-Hegelian thought is that there isn’t a real distinction between means and ends. There’s a kind of Aristotelian organicism in which the thing that is the good life is also the weapon or the tool for achieving the good life. So in Jameson, cognitive mapping is a tool for creating understanding that can then serve in political or even militant struggle. But it’s also one of the grave harms of capitalism for a Lukácsian like Jameson that a kind of cognitive alienation from the material circumstances of our existence has driven a wedge between the world of experience and the realm of causes or laws. What I’m trying to say with this is that for Jameson as for Lukács—I want to be really careful here—one of the goods of a well-realized society, of a communist society, or what have you, is that it will not experience a failure of cognitive mapping. The realm of freedom and the realm of necessity will be well integrated. I’m taking the long way around to saying that neither Jameson nor anyone else is really interested in the experiential or philosophical sense of grasping the material dynamics of mind, world, society, and production merely as an instrument for the achievement of political goals. I’m tempted to say that for Lukács, Jameson, and others, the loss of successful cognitive mapping, while not the defining evil of non-communism, is nevertheless the defining evil of capitalism. It is the specific thing that other class societies did allow, even if they too had to go for other dialectical reasons.
In a way, my disagreement with contemporary Jamesonians is not directly about whether grasping vibes contributes to class struggle, but about what kind of cognitive relationship to causal-material structure is involved in or facilitated by vibe grasping. A Jamesonian would have some skepticism about my position that grasping a vibe is in and of itself a way of at least partially grasping underlying causal structure in the world. To the Jamesonian grasping a vibe can turn out to be preliminary to a more structure kind of grasping, or can turn out to be a dead-end, a kind of placebo cognition.
The Romantics had this distinction between imagination and fancy. Fancy is what looks superficially like a work of imagination. It surprises and delights, like when you combine one animal with another animal. Fancy enthralls with its newness. But fancy doesn’t end up leading anywhere in the deeper project of world-making, and it’s that deeper project of world-making that imagination has the potential for.
I think the Jamesonian would regard a lot of our current preoccupation with grasping vibes as being a kind of fancy. So, like, a high modernist work like The Waste Land or something (by contrast with something like Balzac) gives us vibe but doesn't enable deeper cognitive mapping of causal-material reality. And while high modernist literature at least reflects on this failure to move on from vibe to deep cognitive mapping, late modernist and postmodernist 'list-making' literature can’t even admit that there's something more than vibe to aspire to. That's the Jamesonian complaint as I understand it.
EFFECTS: It sounds like you’re agnostic about whether grasping vibes could be or should be a tool of political analysis. Is that fair?
PELI: I think there’s a bit of a question for me about what kind of political analysis we’re talking about. Are we talking about what’s going to be useful for drafting policy ideas? What’s going to be meaningful for running black bloc anarchist actions? Is critique a spiritual activity that embodies the same values that lead us to strive for material changes, or is it an activity that’s directly helpful to our ability to enact those changes?
What I mean is that I do think that our ability to aesthetically work out what comes from what world of thought is indispensable for us, even for the practical goal of wanting to be able to do crude inductive reasoning or to develop the aesthetic faculty for grouping things into kinds. (When I say ‘kind’ I mean a world of thought and action, a world of possibility, a world of what can be intelligently thought about relative to what.) But I have a little bit of doubt about whether grasping the vibe of New York as a city means you’ll be able to better analyze the financial underpinnings of American imperialism. I don’t know, maybe? I’m not saying it’s absurd. I just don’t know.
Even if I think there’s economic and social-material factors that explain cultural phenomena, which I’m sympathetic to, I don’t know that they sort of richly explain them in such a way that the explanatory structure is something that’s lived in the fine details of the experience of cultural phenomena. Or that the descriptions of fine details of cultural phenomena are going to lead us to the correct analysis of economic phenomena that maybe causally explain cultural phenomena. But, again, I don’t want to be too overconfident about this stuff: I read a lot of poetry and poetic theory; I read a lot of Marxist theory and Marxist philosophy; I have some doubt that we can maintain intimacy in thought all the way through whatever causal lines from the aesthetic, to the experiential, to the cultural, to the social-material, to the economic, to the technological.
EFFECTS: When you were describing how neural networks work as a model for vibe-grasping, you said something like ‘we make an inference to a kind of compressed language, which points to a structural affinity, which itself points back to a common generative process.’ Is that right?
PELI: Yeah.
EFFECTS: Because you might think that a certain kind of Marxist would respond, ‘oh, right, the common generative process: that’s called capitalism and it holds for every phenomenon. You can’t name a phenomenon for which it doesn’t hold.’ And I guess I sort of agree with you that this response seems fine, as far as it goes, but it doesn’t shed much light on the differences between phenomena, differences that matter a great deal to our experience of the world, differences that we often want to understand on their own terms.
PELI: Yeah, I’m with you. It’s very interesting to me that vibes seem to be local. Can there be a sort of ascent? Say you grasp all the vibes. Can you then grasp a vibe of the vibes? And then the vibe of the vibe of the vibes?
Hegel does a lot of this in the Philosophy of History—here are the particular characteristics of these particular states and kingdoms, and here they combine into the spirit of the epoch, and so on—though I have to admit it is an open question to me as to how this is supposed to work. I would say I don’t have a well worked out concept of the relationship between vibes and more conceptually structured Geisty totalities of the kind that Hegel discusses. I’m sure that you need to have some vibewise grasp in order to get going with more conceptual understandings of Geisty structures; what I’m not sure about is how much benefit for that conceptual grasping can be found by intensifying and intensifying your vibe-grasp. It’s a little like saying, ‘being able to see and recognize physical objects is very important for doing physics’. But how far does this go? Is being a great painter going to make you better at physics just because it intensifies your intuitive understanding of physical time and space?
To try and be sort of precise: I’m pretty confident that vibewise grasping is a precondition for a more conceptual grasping of Geisty phenomena, grasping of the kind that Adorno or Hegel are famous for performing in impressive ways. I’m not sure how much of a focus on more and more vivid and detailed vibewise grasping can claim to contribute to this more conceptual grasping of the Geisty things in question. Just as I think that we're never sure how much Adorno’s or Jameson’s stylistic analysis of some culture under capitalism is going to help a more concrete- or action-oriented economicsy analysis. Again, some amount may be important, but is it the case that the more you grasp the Geist of capitalism the more you grasp the concrete material forms of political action? Unclear.
Walter Benjamin is an interesting case. Benjamin I think represents an intermediary point between conceptual and aesthetic engagement with material conditions and their structural underpinnings.
EFFECTS: That reminds me of a passage in “A Theory of Vibe” that sounds a lot like Benjamin’s notion of ‘constellation’. At one point, you write that ‘it is logically possible to share a trained autoencoder’s formula directly by listing the substrate of a neural network bit by bit, but it’s a pretty bad idea to try, because the computations are too complex’. And then you say that ‘one mathematical fact about neural nets that neural-netty creatures like us can easily use, however, is the practical identity between the trained autoencoder and its canon. If grasping a loose worldly vibe has the form of a trained autoencoder, we should expect to share our vibe insight by intersubjectively constructing an appropriate set of idealized phenomena.’ This follows from an earlier moment in which you cite a blog post that catalogues vibes by simply naming the set of objects and places that share them. There’s no meta-aesthetic proposition that’s verifiable with respect to these objects. It’s just about situating an object vis-à-vis comparable objects. Is that right?
PELI: Yeah, that’s perfect. Autoencoders are lossy. They don’t create perfect reconstructions; they create approximations. What I call ‘the canon’ in my dissertation (and now call instead ‘the image’ in my book) is close to what literary scholars would call ‘the imaginary’. The idea is that we have the actual objects that the autoencoder is trained on, and then we have the set of possible outputs of the trained encoder. The outputs are always going to be idealized or simplified even with regard to objects in the training set. That’s because we design autoencoders to be incapable of memorizing the objects in the training set, such that it must learn a structural principle rather than make perfect reconstructions. What’s interesting is that the ‘canon’ or ‘image’ of the autoencoder will have a much denser or purer form of the vibe than is found in the training set because things in the image only vary with respect to the autoencoder’s structural principle, what it learned for recreating its objects. They can’t vary in any sort of noisy way. So things are only going to differ from one another along respects that have been structurally modeled by the autoencoder as meaningful. So there are no ‘meaningless differences’ between objects in the canon/image, whereas there are ‘meaningless differences’ between objects in the training distribution. (I’m skimming over some technical niceties here.)
This is what I call the ‘art-like property’ of autoencoders. You have a material world, you have a way of seeing developed by the autoencoder in order to understand the material world, and you have the imaginary world that is basically the interpretation of the material world in accordance with a way of seeing. You can think of the imaginary world as a set of samples from the image of the autoencoder; it directly expresses the worldview learned by the autoencoder. There’s an interesting property which is that if I’m trying to train a fresh new autoencoder to be similar to the autoencoder that I already trained, I can make a shortcut by training from the image of the previous autoencoder rather than the training set. This suggests that you can learn the vibe much quicker from the image than you can from the training set. It’s like the difference between the Kafkan and the Kafkaesque. The output of the autoencoder is Kafkan and the set that the autoencoder trained on is Kafkaesque. Then the space of possible events and states that can take place in Kafka’s stories and novels are Kafkan, whereas things in the world can’t be Kafkan, but they can be Kafkaesque; they can be approximated by the kinds of structures one can build using the Kafkan schema. So exposure to the Kafkan can teach you the ability to grasp the Kafkaesque in the real world. (I use Kafka because one has a strong sense that everything was shaped by an economic, terse, underlying schema that very strongly restricts what kind of shapes can appear in the Kafka universe.)
The model is obviously a toy. It assumes a lot of weird things about the process of creating the work of art—person gets influenced by lifeworld, person thinks lifeworld into work—and this is not a fine-grained picture of what artists do or what literature gives us. But I do think it’s an approximation of a way that things sometimes work and what we sometimes get from literature, and I think that it demonstrates basic principles that we can then modify and reorganize to create formal analogies or allegories of much more complex, rich, and robust stories about what the creation of literature is and what the consumption of literature is. I like having a point at which my work verges on something like concrete computational cognitive science or cognitive theory. It’s not like the goal is ‘concrete computational cognitive theory that we could eventually test.’ It’s more that I want to make hermeneutic and phenomenological use of these mathematical or mathematical-empirical models, and I want that use to be something other than merely metaphorical, allegorical, or metaphysical. My way to prove that there’s more than just metaphor or metaphysics here is to try to produce this sort of basic story about artistic production, consumption, and its cognitive functions.
Sometimes I look at Badiou or something. Obviously he knows much much more mathematics than I do—
EFFECTS: Does he?
PELI: It’s interesting. It’s not like Being and Event is mathematically nonsensical. Apparently he made some major, although extremely technical errors, which he then acknowledged after a mathematician wrote to him. Then a later book, which uses a lot of category theory, was positively reviewed by logicians and mathematicians. Sort of like ‘oh this is a very nice and creative textbook that explains concepts in category theory in very interesting ways and suggests very fun philosophical analogies, though of course I’m not a philosopher and so I can’t really judge.’ So the thing with Badiou is that he’s not a professional mathematician, but he’s a serious amateur, and after messing up in his first book, in his second book he’s up to par.
But still, for me, Badiou doesn’t go far enough. Even if you want to tell this very rich mathematical story, I feel like I need to at least anchor what I’m doing in an account of how these kinds of processes could be causally involved in producing certain aspects of aesthetic and artistic experience as we know it.
To be a little more concrete, I’d say that I’m interested in the interdependence of practices, patterns, and mechanisms. One way to look at human behavior is through practice, that is to say whether and to what extent the things that we do are done in accordance with rules or norms. Another way to look at human behavior is through pattern. When I speak about pattern, I mean our ability to represent something as a whole rather than just a list of parts—a whole as opposed to just one thing after another. The question is, ‘what masses of stuff can we successfully treat as objects and relations and structures?’ So we can look at a body of behaviors as a practice, as responding to a certain rule in a game of rules, or we can look at it as a pattern, where, in order to apply the rules, we need some sort of basic cognitive tractability, and pattern refers to either scientific or transcendental constraints on when we can form practices and norms and rules.
The third way of looking at human behavior is mechanism, which is a way of answering why there are patterns, why we are dealing with a mass of phenomena that’s amenable to our cognition and that’s tractable to our concepts—or, to be more precise, that’s tractable to our cognitive preconditions for building concepts. Now when you start talking about mechanism, that’s when Kant would start talking about God or God-that’s-not-quite-God, his wise and benevolent but not omnibenevolent or omniscient architect of the universe, whereas we would talk instead about some combination of evolutionary and information-theoretic and mathematical processes or forces of structuring the phenomenon, in this case human behavior, such that the mathematical constraints or structures or force account for the fact that there are patterns, and that patterns exist such that the world is tractable to our cognitive capacities, and the world is tractable to our cognitive capacities such that we can institute norms or play language games or whatever.
I really think that’s the unique thing I’m trying to do: to bridge the hermeneutic and phenomenological uses of mathematics and computer science in the humanities and philosophy and the more computational cognitive science of literature. My heart is in the phenomenological and the hermeneutical, but then my conscience, my epistemological conscience, is with computational cognitive science. In a way my whole interest is in the grounding of the phenomenological and hermeneutic in the causal-material, and I think I’m trying to do it by translating an aesthetic and literary-theoretic and phenomenological vocabulary, going back and forth between that and the vocabulary of mathematical processes and properties that are cheap to materially implement, in the sense that nothing very fancy would have to go on for these things to be materially implemented. The kind of structure that if you’re just given a bunch of neurons and some kind of feedback mechanism performing like a natural selection function like autoencoders or compression schemas these things easily emerge.
EFFECTS: Since you mentioned that this kind of cognitive-theoretic grounding is what distinguishes your work from other kinds of vibe talk, I wonder what you make of the recent journalistic interest in vibes. Why is everyone interested in vibes now? What are we missing when we don’t think about the concept in cognitive-theoretic terms?
PELI: One way to think about the work that high theory does for artists, which is a little different than what it does for philosophers, is that it’s a way to allow for certain complex aesthetic/affective experiences to make a claim on material reality. Like ‘oh this thing that I’m feeling or you’re feeling or we’re feeling is a complex aesthetic structure, not merely how we’re personally constituted but something in the world itself,’ and I think that vibe talk is almost functioning as a way to get there without high theory. Or a way of doing something that’s a little less than high theory. When you’re talking about a vibe, you’re making the minimal differentiation or objectivization of a complex of perceptions and emotions and affects and ideas. You’re not committing to the nature of the mechanism that it’s grasping. You’re not backing into any particular theoretical or philosophical tradition.
Maybe we could put another term before practice, pattern, and mechanism, which is experience. With high theory, you start by doing some complicated hermeneutics on a feeling and then say that what we’re feeling is the world historical force of capitalism or the being of language or something and then further connect it to the material or metaphysical. And I think that, with vibe talk, it’s more like there’s some affective emotional aesthetic complex of perception and then instead of starting to do hermeneutics on it you’re making a minimal claim that this comes by way of some sort of pattern and that the pattern is not just in me but out there. There’s a reason I’m grasping this as a vibe and it’s because it has the kind of unity that allows me to grasp it as a vibe. I can’t just grasp any collection of objects as a vibe; they have to be hospitable to vibe grasping.
It’s a more sort of empiricist rather than idealist or materialist path into Geisty phenomena and their social-material infrastructure. I guess it’s inevitable that this would be attractive to a sort of businessy or practical sector of the discourse. There’s part of me that’s just like ‘have to take that L’. Some of the people doing it are in this artist, consultant, trend-forecaster business and saying things like ‘this is the market you should be getting into’.
EFFECTS: The infamous vibe shift article2 is, as you say, in part about trend forecasting, and I can imagine that you wouldn’t want trend forecasting to be the horizon of a theory of vibe. But I wonder if the very idea of forecasting vibes offers some proof of your sense that vibes are cognitively or epistemically tractable. Like one way we know they’re real is that you can predict them.
PELI: With all these AIs that produce images or texts in a particular style, we’ve received a lot of evidence recently that what we call vibes or styles are very natural mathematical structures. Processes that get fed the kind of data we’re interacting with tend to converge into structures that are close to the kind of structures I’m working with. Twitter gives even more direct observational experience of what happens when you get a mass of text and different subcultures constantly juxtaposed. It makes concrete how these vibey organizing principles hold things together.
I think of my project as being close in spirit to Reza Negarestani. He wrote an avant-garde horror novel called Cyclonopedia that’s a big deal, but he’s also a very gifted amateur technical philosopher, and he wrote this book called Intelligence and Spirit, which is all about the prospect for a scientifically materialist Hegelianism in terms of ‘can we think of Rationality in terms that explain how we build mechanical cognitive processes even if Rationality is not necessarily definable in mechanical terms but must be the kind of thing that shapes and is shaped by mechanical processes?’ So there’s stuff that’s going in some sections of what’s called post-continental philosophy, people who used to be associated with speculative realism but no longer are, sometimes called neorationalist like Reza, Peter Wolfendale, Ray Brassier, who are all philosophers interested in negotiating between practice, pattern, and its causal-material infrastructure. I’m especially interested in how aesthetics and experience enter into it, which is something that interests the rest of them less. I want to say that there’s an important horizon to bringing some empiricism or empiricist tendencies into the discussion of Geisty entities. I don’t think it’s only interesting to me and marketing people.
EFFECTS: Last question. You’ve talked a little bit about the autoencoders as a kind of analogy that might be useful both for producers and consumers of art. It helps us interpret things, but it also interfaces at some level with the making of things. What kind of art or literature would you want to see such that it felt as though it confirmed or elaborated your ideas?
PELI: If it can inform a way of thinking about art that’s potentially useful to artistic practice and criticism, beyond just being philosophically interesting, it’s that it helps bridge aestheticism and strong cultural materialism, that everyday ‘well art is always about something in the lifeworld’ point of view. It helps bridge between the realms of art as pure formal intensity that we just experience as patterned or structured fields of coherence and the desire for art to speak to life or speak to world-historical conditions. Relatedly, I think it can also help bridge or smooth out our understanding of mimetic works of art and our understanding of forms like installation where there’s just a bunch of objects that are interesting. It can build a continuum between them in as much as we can think of them both as different strategies for presenting a collection of objects or moments that fit into a compression schema such that one can learn the compression schema for them and apply it more widely as a worldview.
It’s not that all aesthetic coherence in a collection of artifacts will have a functionality as a lossy interpretive schema for the world at large, but it does say that it easily can be so, that it’s a plausible or natural application of any field of aesthetic coherence. Any field of aesthetic coherence can be converted into an interpretive schema for the world at large. Whether it was designed to do so or whether it will have something very interesting for us if we use it that way is a further question, but there is always a story to be told about how aesthetic coherence has the potential to turn into a pair of glasses. It defines a very natural aesthetic coherence-to-interpretive schema pipeline: the idea that every aesthetically coherent assemblage is a compressed representation and interpretive schema for itself and therefore potentially for things that are even a little bit like it or even remotely like it.
It’s definitely become part of how I think of works of art now. First, I absorb the capacity of the work to hold together, and then I start to think about which things in the world approximately hold together in the same way, and as I find them I can start looking at them with the same attunement that I developed in order to grasp the coherence of the work of art. The slogan for this is: you can learn a way of seeing by apprehending the set of objects that this way of seeing sees the best.
My friends who are in those sorts of practices moving between the representational and the non-representational have found this way of thinking really useful, like people who do experimental film and also installation, or people who do experimental dance and choreography which sometimes veers towards performance art and sometimes towards more theater and blends them. They’re thinking about different modalities and techniques that create a field of consistency that naturally repurposes itself for interpreting things outside its strict boundaries. In this sense, vibe is a unified theory of mimesis for things that are directly mimetic and for things that aren’t at all.