N. Katherine Hayles

In the [YOU] on AI Field Guide

The cycle that began with [YOU] on AI asks what it would mean to take the orange pill—to see the machine clearly, without the narcotic of hype or the paralysis of fear. Hayles is the theorist who explains why seeing clearly is so difficult: the AI text seduces with its characteristic fluency, the decorrelation of fluency from authority is a material property of the substrate, and the builder who does not attend to the substrate will be shaped by it without knowing. Her concept of material literacy—the ability to read the substrate through the text, to hear the training corpus in the patterns of expression—is the practical discipline the cycle recommends under the name of discernment.

The ascending friction thesis that runs through the cycle acquires new depth through Hayles’s framework. When AI eliminates lower cognitive tasks and relocates difficulty to a higher floor, it is not merely changing what work is hard. It is restructuring the material substrate of subjectivity itself: the identity that knew itself through syntactic struggle must now reconstitute at the level of architectural judgment, and the reconstitution is a change not in the content of the self but in the structure of the self. The twelve-year-old who asks her mother “What am I for?” in one of the cycle’s most arresting passages is already operating in the territory Hayles maps: the question arises from the experience of having stakes in the world, of being mortal and situated and finite, conditions that AI makes visible precisely by threatening to render them economically irrelevant.

Hayles’s SIRAL model—the five cognitive capacities of sensing, interpreting, responding, anticipating, and learning—provides the sharpest vocabulary for what the cycle calls the amplifier. AI exhibits all five capacities, often at extraordinary levels of sophistication, and yet it exhibits none of them experientially. The model does not know it is thinking. The silent middle that the cycle identifies as the most honest position in the AI discourse is, in Hayles’s terms, the population that has achieved material literacy—that can hear the substrate speaking through the text and has not yet found a cultural script adequate to the full complexity of what they hear.

Origin

Hayles was trained in chemistry before turning to literature, and the double formation—scientific rigor, humanistic breadth—is the signature of her career. She came to the history of cybernetics through literary theory, tracing the seam where information theory and narrative intersect. Her central discovery was the one Claude Shannon had not intended: that the brilliant abstraction which defined information as pattern independent of substrate had been generalized, far beyond its engineering domain, into a metaphysical claim that would haunt the culture for decades. If information is substrate-independent, the mind can be uploaded, the brain is a computer, and AI processing is equivalent to human thought. Each proposition follows from the disembodiment thesis. Each is wrong for the same reason: information is never disembodied.

Her 1999 masterwork traced the history of this error from the Macy Conferences of 1946–53—where Norbert Wiener, John von Neumann, and Warren McCulloch dismantled the assumption of a unified self by showing that feedback mechanisms operate identically across biological and mechanical systems—through the cognitive science revolution that absorbed cybernetics into a model of the mind-as-computer. Each time the humanist subject was threatened, she showed, it reabsorbed the challenge without relinquishing its foundational assumptions. AI has made that reabsorption impossible. In a 2025 lecture she declared her “astonishment” at large language models while immediately insisting on what she called “a minority view”: these systems create “billions and billions of correlations from human-authored texts” and “form networks between these correlations, and make inferences as a result”—a description carefully calibrated to acknowledge genuine cognitive output while refusing to attribute consciousness or understanding.

In Unthought (2017) and Bacteria to AI (2025) she extended the framework into the present. The concept of the cognitive nonconscious—the vast domain of technical information processing that operates without awareness and shapes the environment of conscious thought—gave her the vocabulary for AI that earlier work had reached toward. And in her 2025 interview with Jesse Damiani, she stated the conclusion directly: “It’s a disastrous mistake to think that human agency is the only agency working in what I call cognitive assemblages—that is, collectivities that include artificial intelligence as well as nonhuman intelligence. The first step along this journey is to recognize that agency is distributed, cognition is distributed.”

Key Ideas

Re-embodiment. Information is never substrate-independent. A poem read by candlelight in a leather-bound volume is not the same cognitive event as the same poem on a smartphone in a subway. The materiality of the reading experience is constitutive of meaning. AI text emerges from a specific substrate—billions of parameters, an optimization function trained on human evaluation—and that substrate shapes every output in identifiable ways. The characteristic fluency of AI prose, the smoothness that seduces and conceals, is not a stylistic choice. It is a material property of a system optimized to produce text that sounds like thinking without the embodied process that constitutes thinking. Distributed cognition researchers like Edwin Hutchins had shown that thinking extends into tools; Hayles insists that the specific properties of those tools color the thinking.

The flickering signifier. In print, the signifier stays put—the ink fixes to the page, the durability underwrites authority, citation and attribution are possible because the passage can be located and verified. Digital text flickers: continuously refreshed, rendered from underlying data structures, subject to alteration at every level of the stack without trace. AI text is the ultimate expression of this condition: it does not merely flicker between states of a pre-existing text but flickers into existence, token by token, through stochastic prediction. The same prompt submitted twice produces different text. The output is radically contingent—one realization of a probability distribution that could have selected differently. The builder’s craft in the AI age is therefore not the traditional craft of production but the craft of selection: evaluating, discriminating, choosing among the machine’s productions with the taste and judgment that only embodied, experiential cognition can supply.

The cognitive assemblage. Human-AI collaboration is not a person using a tool. It is a cognitive assemblage—a collectivity in which cognition flows between biological neural networks and silicon architectures in patterns that neither component could sustain alone. Three levels of collaboration are possible: editorial (the machine helps the human say what she already knows), structural (the machine reveals patterns implicit in the human’s thinking), and generative (the machine makes a connection neither partner predicted, changing the direction of the work). The third level—the emergent insight that belongs to neither component—is what the liberal humanist framework cannot accommodate, because it cannot be attributed to any single origin. It arises from the interaction. The assemblage thinks it.

The cognitive nonconscious. Most cognition occurs in the dark: the brain parses syntax, regulates posture, constructs the perceptual world from raw sensory data, all without awareness. Hayles extended this concept to encompass the technical systems that process information alongside us—the algorithms that filter searches, the networks that route communications. AI is the most sophisticated expression of the cognitive nonconscious yet achieved. It exhibits sensing, interpreting, responding, anticipating, and learning at levels that blur the functional boundary between nonconscious and conscious cognition—not by achieving consciousness but by achieving a degree of functional sophistication that makes the distinction difficult to maintain in practice. The key question is not whether AI is conscious but how the partnership between human conscious cognition and AI nonconscious cognition is organized, and whether conscious thought retains governance.

Material literacy. The corrective Hayles offers is practical as well as theoretical: the capacity to read the substrate through the text, to hear the training corpus in the patterns of expression, to feel the optimization function in the trajectory of the argument. The builder who possesses material literacy does not evaluate AI output as though it were a printed text. She evaluates it as what it is: the output of a stochastic generation process that produces fluency without discrimination and confidence without knowledge. She reads for the characteristic errors of the substrate—the confident interpolation that fills gaps with statistically plausible but factually incorrect content—and she maintains, against the seduction of fluency, the cognitive authority that only embodied, experiential cognition can supply.