Possibilist Speculation and Spotify Personas

Now that I’m finished with The Sonic Episteme I’m working on some currently very structureless project on the aesthetics of post-probabilist neoliberalisms. So this is me working through some ideas related to that project.

At the level of math, derivatives supplement probabilities. But as Louise Amoore argues, even this math isn’t quite enough on its own. When it comes to deciding which risk correlations need to be acted upon and which don’t, derivative calculus can’t separate signal from noise. This requires qualitative information. According to Amoore, at various points in the process, from programmers writing the code to the CBP agents reading the risk profiles that software outputs, people rely on aesthetic judgments to bridge the gap between where the math ends and the minimum threshold of information one needs to make a justifiable decision about, say, whether to detain someone at the border. Aesthetics are how we “incorporate the very unknowability and profound uncertaintity of the future into imminent decision” (9) — we have no mathematical or empirical knowledge of how things will play out, but when we rely on aesthetic conventions and ideals, we’re basically appealing to senses calibrated by established systems of domination to guide us. This guarantees that whatever decision we make, the future will be one that reproduces if not amplifies existing relations of domination and oppression.

Amoore identifies two components or steps to possibilist calculus: (1) the proliferation and arraying of possibilities, and then (2) the decision about what to do–which of these possibilities deserve our attention because they don’t feel right?

The arraying of proliferating possibilities is an attempt to catch what Gaussian normal curves can’t. The problem with probabilistic thinking is that it doesn’t help you prepare for improbable but devastating events like terrorist attacks. These are singular events for which there is no past dataset to use to establish a regular rate of recurrence. Possibilist speculation tries to figure out what isn’t strictly likely to occur, but which may occur. In the same way that probabilist distributions turn randomness from a bug into a feature by making it regularizable and predictable, possibilist speculation takes the bugs in probabilist math and transforms them into features: “the degrees of doubt always already present within mathematical probability multiply and take flight as imaginable, if not strictly calculable, possibilities” (10). Key here is the multiplication and proliferation of these imaginable futures, which Amoore frequently describes as an “array.”

Whereas probability distributions put everyone or thing on a single, ontologically flat continuities (i.e., every data point is included on the same spectrum), possibilist arrays are mobile, iterative fields of proliferating, constantly revised gradations. “Arraying” involves “the fractionation of ever-more finite categories of life–degrees of safe and dangerous, vulnerable and durable, mobile and restricted, identifiable and unidentified, verifiable and unverified, and so on” (12). This fractionation slices and dices the same material in whatever ways it can, cutting and re-cutting groups, individuals, and even sub-individual parts into ever finer-grained and novelly parsed slices. Instead of single continuity, possibilist calculus produces multiple, proliferating, constantly revised gradations; through these constantly proliferating and revised gradations we generate more and more and more constituents of the array.

Probability curves organize data points by finding equilibrium; normalization basically functions like a kind of equalization, finding the most normal instance and then bringing everything within relative proportion to that norm. With possibilist calculus, “the emphasis…ceases to be one of the balance of probability of future threat and occupies instead the horizon of actionable decisions” (Amoore 58; emphasis mine). So, the organizing principle shifts from proportion or balance to horizon. We might think of horizon as a vanishing point beyond which we cannot perceive. As in 2D art, horizon-as-vanishing-point is also an orientation point, the zero-point in reference to which everything else is situated. This orientation point also implicitly situates the spectator: we adopt the view provided by the image’s framing as though it was our own, unmediated perception of a situation we were physically located in. Here horizon in the artistic sense bleeds into “horizon” in the phenomenological sense as a subject’s experiential and perceptual locatedness. As Linda Alcoff explains, as we interact with the world, we develop a set of implicit knowledges on the basis of which we judge all new info as rational or irrational, legible or illegible. This constitutes what Alcoff calls our “implicit horizon.” This horizon is the back door through which possibilist speculation brings aesthetics and introduces them to complex mathematical and computing systems. Horizon is the key technique that bridges the calculative part of possibilist speculation (the generation of possibilities and the arraying of them to create a horizon) with the aesthetic or affective part (horizon as phenomenological ground for subjects). Horizon organizes data or information in ways that orient perception and, importantly, is oriented to a particular kind of perceiver. Arrayed horizons are legible to us when they fit the interpretive horizons we bring with us. For example, the possibilist logics that Customs and Border Protection use to try to figure who among the many people trying to enter the country is a terrorist only make sense to people who bring with them the sense that terrorism is something non-white and non-Westerners do, not something that, say, white nationalists do. Because we are ultimately making sense of these arrayed horizons by comparing them to our own interpretive horizons (this is the second step, which I discuss just below), arrayed horizons have to be made compatible with the interpretive horizons of the people they’re designed to speak to. And since all of our interpretive horizons are shaped by interacting with a world organized to foster patriarchal racial capitalism, this means that arrayed horizons are also designed to be compatible with and foster it.

(2) As Amoore emphasizes, the point of generating all these arrayed possibilities is to help us decide what to do when we can’t probabilistically predict the future. By comparing possibilities arrayed on a horizon to our own interpretive horizons, we can easily and efficiently find the threats or problems that escape probabilistic calculus. Such comparisons are efficient because they don’t need to pay attention to anything that feels relatively normal, nor do they need to seek out abnormalities. They are looking for “singularities against the grain” (30), the Heideggerian broken hammer that only rises to the level of our attention because something’s not right. Here, “grain” describes the lines of orientation worn into us as we experience the world, the threads that compose our interpretive horizon. Possibilist speculation seeks out instances that clash with the grain of our interpretive horizons, things that disrupt our implicit knowledges’ ability to read and make sense of things. “The politics of possibility invites attentiveness to the affective judgments of something that ‘feels right’ or ‘looks out of place’”(25). Princess Leia’s bad feeling about the cave the Millennium Falcon just landed in is an example of such judgment: a cave on an asteroid shouldn’t be so moist, spongy, and humid. We rely on these bad feelings to “decide the exceptions that make it possible to act even in the face of radical uncertainty” (Amoore 9). Instead of using statistical normalization to cut the line between population and exception, possibilist calculus relies on our bad feelings to render individual cases exception. Thus, it doesn’t matter what an individual has actually done or is doing–only what those in positions of power feel like they could do. Thus, as Amoore explains, “the mode of possibility…does not govern by the deductive proving or disproving of scientific and statistical data but by the inductive incorporation of suspicion, imagination, and preemption” (10). Math and computing provide a continuous feed of arrayed possibilities, and aesthetics/interpretive horizons parse that vast and variable body of  information into signal and noise. For example, in data analytics, users’ sense of aesthetic form guides their interpretation of datasets: relationships among datapoints that follow formal relationships we have become habituated to perceive are the first ones that rise to the level of our perception. As Amoore explains, “aesthetic judgments and inferred correlations…are invited within calculation. Aesthetic judgments on the pleasing quality of particular forms, though they did eschew strict statistical distribution, nonetheless inferred correlations among apparently disparate objects” (139). One “advantage” of this technique is that our aesthetic norms are thoroughly shaped by and saturated with white supremacist capitalist patriarchal logics, so the probabilist appeal to aesthetic judgment is one way to ensure that these new modes of calculation produce the same old relations of inequality.

Spotify’s “personas” tool is one example of how central aesthetic judgment is to possibilist speculation. Like possibilist calculus in general, Spotify personas were invented in response to the limitations of probabilist speculation, which studies norms across a single population and can’t sufficiently disaggregate and segment that population. As they explain, “designing for a mass, generalised audience isn’t likely to end up pleasing ‘everyone’. So in 2017, our team was challenged to create a better understanding of existing and potential listeners. We wanted to agree on how to differentiate the needs of these listeners and the problems our products could solve for them.” Personas are arrayed possibilities that let Spotify employees make programming decisions based on potential futures.

Personas are rooted in sliced and diced data analytics, but they supplement that math with aesthetics. Though persona designers began with robust audience research, “we were determined to put a face to our listeners.” To do this, they literally made characters that “represent” each persona: “we arbitrarily picked genders, names and appearances that matched the range of people we interviewed…we reduced the representation of personas to keywords, colours, symbols and energy levels reflecting their enthusiasm for music. This exercise helped us navigate through the variations of poses, facial features, clothing and visual styles we created.” Clothes, styles, colors, genders, names are all things users interpret on the basis of aesthetic norms they’ve already internalized and become habituated to. And it’s these aesthetics that, Spotify explains, form the basis of “believable human traits and flaws help create empathy with problems and needs” — that is, these aesthetics help Spotify employees identify what potential problems and needs listeners may have.

As you can see from these illustrations of various personas, we’re supposed to read into their identities, their sartorial styles, and the colors used to represent them. And this is yet another way that pre-existing biases get imported into neoliberal modes of speculation.