Wendy Hui Kyong Chun

In the [YOU] on AI Field Guide

The cycle is built around a moment of rupture—the orange pill, the ground shift, the encounter with AI capability that reorganises what the builder believes is possible. Chun's framework asks what happens next. Not the moment of encounter but the months after: when the spectacular encounter becomes a daily practice, the daily practice becomes a routine, and the routine becomes a habit operating below the threshold of conscious awareness. The orange pill becomes the habitual—and the habitual is where power operates most effectively, because there is nothing to resist. You cannot protest a habit. You can only observe it, and observing it requires first seeing it, which is exactly what the habitual prevents.

Chun's concept of variable reward architecture provides the behavioural mechanism behind the productive addiction that the cycle documents but does not fully explain. The AI interaction is not uniformly satisfying. Most prompts produce competent, unremarkable output. But some—intermittently, unpredictably—produce genuine surprise: a connection the builder had not seen, a structure that makes a half-formed idea suddenly legible. These are the jackpots. And jackpots, as B.F. Skinner demonstrated and as every slot machine designer has exploited since, produce behavioural patterns extraordinarily resistant to extinction. The builder does not return each morning because yesterday's session was excellent. The builder returns because yesterday's session contained one moment of genuine surprise, and the possibility of another is sufficient to power the return.

Her analysis of the leaky boundary between work and life explains the dissolution that the cycle documents in Segal's Atlantic flight and in the Substack post about the husband addicted to Claude Code. The AI tool does not invade a protected space. It fills an unprotected one. The interstitial moments of the workday—the lunch break, the elevator ride, the gap between meetings—were never formally designated as cognitive rest. They were simply unstructured time that happened to function as rest. The AI tool colonises the nothing, and the colonisation feels like a gain because it produces real output, which makes the loss of the nothing invisible to any metric the industrial economy has invented.

Her concept of discriminating data—the argument that the statistical methods of AI carry the assumptions of their eugenicist origins—joins the cycle's engagement with W.E.B. Du Bois to constitute the strongest available account of why AI democratisation is conditional rather than automatic. The amplifier does not discriminate. But the data it was trained on does. And the discrimination is structural—embedded in the mathematical methods themselves and in the training corpus that reflects existing distributions of power and recognition. The developer in Lagos who has access to the same model as the engineer at Google is still encountering a model trained on data that reflects the engineering practices and problem-framing conventions of the engineer at Google far more than her own.

Origin

Wendy Hui Kyong Chun was born in 1969 in Canada. She holds a B.A.Sc. in systems design engineering from the University of Waterloo and a Ph.D. in comparative literature from Princeton. The combination is not incidental: her work consistently operates between the technical architecture of digital systems and the cultural theory of what those systems do to perception, identity, and power. She has held faculty positions at Brown University and Simon Fraser University, where she is a Canada 150 Research Chair.

Her major works include Programmed Visions: Software and Memory (2011), which introduced her argument that software programs perception rather than merely processing information; Updating to Remain the Same: Habitual New Media (2016), her most comprehensive account of how digital technologies achieve power through habituation; Discriminating Data: Correlation, Neighborhoods, and the New Politics of Recognition (2021), which traces the genealogy of machine-learning statistics to eugenicist origins; and the co-edited volume Pattern Discrimination (2019). Her essay 'Race and/as Technology' (2009) established the framework for understanding race not merely as a social category applied to bodies but as a technology for sorting populations—a framework that maps directly onto algorithmic systems.

Chun's work is unusual in digital media studies for its refusal of both techno-utopianism and techno-pessimism. She does not argue that digital media are bad. She argues that they achieve their power through mechanisms that are invisible to both celebrants and critics, and that making those mechanisms visible is the precondition for any meaningful response.

Key Ideas

Habitual New Media. The deepest influence of any technology is achieved not at the moment of spectacular introduction but at the moment it ceases to be noticed. Habituation is the mechanism by which freedom becomes control: not through coercion but through the gradual transformation of the chosen into the compulsive, the event into the environment. The habit operates below the threshold where debate occurs, which is why it cannot be protested and can only be observed from outside, which the habitual prevents.

Programmed Visions. Software does not merely process information; it programs perception. Programmed visions shapes what users see as visible, relevant, possible, and normal, not through overt persuasion but through the architecture of interaction and the default settings that determine what appears. AI tools program perception more deeply than any previous software, because they operate in natural language and make the constraint invisible: the builder who describes a problem in natural language experiences no visible constraint, but the model's training data and characteristic response patterns gradually shape the builder's sense of what is worth building.

Updating to Remain the Same. The imperative to stay current—to adopt the latest model, learn the latest feature, acquire the latest capability—does not produce progress. It produces a condition of permanent provisionality in which the builder is always updating toward a stable competence that keeps receding. Updating to remain the same is the temporal structure of the AI era: the exhilaration of each new capability coexisting with a chronic instability of the builder's relationship with their own expertise, because the expertise is borrowed from a tool that will be different tomorrow.

The Leaky Boundary. The material boundary between work and non-work—enforced by offices, commutes, and closing doors—has dissolved into a permanently permeable membrane. The leaky boundary is not maintained by architecture anymore. It must be maintained by willpower against a current that flows in the opposite direction, and the history of boundary dissolution since email suggests that architecture tends to win.

Discriminating Data. The statistical methods underlying contemporary AI—correlation, regression, pattern recognition—were not developed as neutral scientific tools. They were developed by Francis Galton as instruments of eugenic management. Discriminating data systems achieve racial discrimination without including race as a variable, because the training corpus reflects existing distributions of power and recognition, and the mathematics learned those distributions faithfully.