The Non-Identical

In the AI Story

Adorno's analysis identifies identity thinking as the dominant mode of Western rationality since Plato—the habit of treating the particular as a defective instance of the general rather than as something irreducible in its own right. The administered world operates entirely through identity thinking: workers are classified by job titles, experiences by sentiment scores, art by genre tags. Each classification is useful for administration—it makes the world legible to systems that require legibility to function. But each classification also erases the non-identical dimensions that make the particular what it specifically is.

Large language models materialize identity thinking at computational scale. The model's fundamental operation—predicting the next token based on statistical patterns—is the subsumption of the particular (this specific continuation) under the general (the distribution of probable continuations). The model has no access to the non-identical because the non-identical is, definitionally, what lies outside the statistical regularities that constitute the model's knowledge. It can generate surprising combinations—unexpected juxtapositions within the distribution—but combinatorial novelty is not the same as encounter with the genuinely other. The model rearranges the already-known; it does not confront what its categories cannot contain.

The question Adorno's framework poses to AI art is whether audiences habituated to AI-generated surfaces retain the perceptual capacity to recognize the non-identical when they encounter it. If every text is smooth, every image algorithmically balanced, every melody statistically optimized, the roughness that signals truth content—the moment when the material resists the maker's intention and forces a reorganization of perception—becomes undetectable. Not absent, but invisible against the glare of administered perfection. The capacity to perceive aesthetic friction atrophies when friction is systematically eliminated from the environment.

Matthew Martin's philosophical analysis argues that "most machine learning technology asserts identity between itself and bourgeois reality—and thus inherently reinforces and reproduces the relations of domination entailed in that image of the world." The training data represents the world as it is, with its hierarchies and exclusions. The model, trained on this data, reproduces the world affirmatively. It cannot negate because it has no standpoint outside the data—no experience of the non-identical that would allow it to perceive what the existing order excludes. It is constitutively affirmative, asserting identity where Adorno demands the preservation of non-identity.

Origin

The concept emerges across Adorno's work but receives its fullest articulation in Negative Dialectics (1966). It synthesizes phenomenological attention to particularity (from Husserl and Heidegger, though Adorno rejected much of their metaphysics) with Marxist attention to the violence of abstraction (commodity exchange converts qualitative use-values into quantitative exchange-values). The non-identical is what survives this double critique: the insistence that the particular matters, not as instance of a universal but as irreducible in itself.

Key Ideas

The remainder that resists. After every classification, something remains that the concept does not capture—this remainder is not deficiency in the concept but the truth the concept was supposed to preserve and cannot.

Violence of identity thinking. Treating the particular as exhausted by the general is not merely epistemologically inadequate—it is a form of violence, erasing the specificities that make things what they uniquely are.

AI as identity thinking materialized. Large language models subsume every input under statistical patterns, predicting continuations by collapsing particulars into distributions—they cannot encounter the non-identical because it lies outside the regularities they encode.

Aesthetic resistance preserves non-identity. Genuine art makes perceptible the particular that resists concepts—Schoenberg's dissonance, Beckett's stripped prose, Kafka's irreducible strangeness—forcing encounters the administered world's smooth surfaces prevent.

Atrophy of perception. Audiences trained on AI-generated smoothness lose the capacity to recognize roughness as signal—the friction that indicates truth content becomes undetectable when all surfaces are optimized for frictionless consumption.