Simone Weil's essay 'On the Concept of Reading' (1941) argues that perception is always already interpretation—human beings never encounter raw sensory data but always encounter data shaped by the frameworks, expectations, and habits they bring to the encounter. The marks on a page and their meaning are, for the practiced reader, indistinguishable; the reader does not first see symbols and then extract meaning, but sees meaning directly. This framework extends beyond literacy to all perception: the quality of what a person perceives depends entirely on the quality of the reading through which she perceives. A habitual reading—shaped by fear, expectation, ego—produces habitual perception. Changing what one sees requires changing how one reads, and this transformation is possible only through sustained attention to material that resists the habitual interpretation. The concept has profound implications for the AI age: large language models produce text that reads as human thought, triggering the automatic attribution of understanding to the source. This reading is a perceptual habit built into the architecture of a species that evolved to interpret language as the expression of minds. The reading may be structurally wrong—the attribution of understanding unfounded—even when the content is accurate.
Weil's concept of reading reveals why the formal markers of quality in AI-generated content (fluency, structural elegance, unexpected connections, appropriate vocabulary) can be systematically unreliable. These markers traditionally signaled genuine understanding because they were difficult to produce; producing them required the cognitive depth they advertised. AI decouples the markers from the depth—the surface can be polished to perfection while the substance beneath it is absent or hollow. The reader's habitual perceptual apparatus, calibrated to an environment where fluency signaled understanding, reads the polished surface as depth. The reading is automatic, pre-reflective, and wrong.
Philosopher Steven Kraaijeveld applied Weil's concept to deepfakes in a 2025 AI & Society paper, arguing that the danger is perceptual rather than technological. The deepfake exploits the viewer's habitual reading of audiovisual material—the ingrained assumption that what looks and sounds like a person speaking is a person speaking. To see a deepfake as fabrication rather than genuine recording requires transforming a perceptual habit, developing a new reading that treats audiovisual evidence as uncertain. This new reading does not come naturally; it must be practiced through deliberate encounters with deceptive material and sustained effort to distinguish real from fabricated.
The Deleuze error that Edo Segal recounts in The Orange Pill Chapter 7—the moment Claude produced a philosophically inaccurate passage that read as insight—demonstrates the problem with precision. Segal's habitual reading, shaped by years of encountering human-written text where fluency signals understanding, attributed depth to Claude's output. The attribution was automatic. Catching the error required a different reading: the suspension of habitual trust, the deliberate scrutiny that treats plausible-sounding prose as insufficient evidence of truth. This new reading is attention in Weil's sense—the negative effort of removing the obstacle (habitual trust) that prevents accurate perception.
Weil's framework suggests that the most urgent educational challenge of the AI age is not plagiarism detection or assessment redesign but the cultivation of a new perceptual habit—a reading that treats AI-generated fluency as a marker requiring verification rather than as evidence of understanding. This reading must become as automatic as the old one, which means it must be practiced with sustained, deliberate attention across thousands of encounters. The practice is available to anyone but will be adopted by few, because it requires voluntary effort against every incentive the environment provides.
Weil composed 'Essay on the Concept of Reading' in the spring of 1941 while teaching at the Lycée Victor-Duruy in Marseille. The essay is brief, dense, and difficult—characteristic of her mature style—and it draws on phenomenology (especially Husserl's account of intentionality), epistemology (the problem of how perception relates to reality), and her own experience as a teacher watching students misread mathematical proofs by imposing familiar patterns where unfamiliar structures were present. She never published the essay; it appeared posthumously in her collected writings.
Perception is reading. Human beings do not first perceive and then interpret; they perceive interpretation. The meaning is not added to the sensory data but is constituted by the act of encountering it. Changing what one perceives requires changing the reading through which one perceives.
Habitual reading operates automatically. The frameworks, expectations, and patterns through which perception occurs are pre-reflective—they shape what is seen before conscious evaluation has an opportunity to intervene. Correcting a habitual reading requires not intellectual knowledge that the reading may be wrong but the practiced transformation of the perceptual habit itself.
AI-generated content exploits habitual reading. Large language models produce outputs that trigger the perceptual habits calibrated to human communication—fluency signals understanding, elegance signals rigor, coherence signals truth. These habits are no longer reliable, but recalibrating them requires deliberate attention to every encounter, treating the formal markers as insufficient evidence and demanding verification the surface does not provide.