Transduction

In the AI Story

Transfer assumes pre-existing terms. Information moves from point A to point B; both points exist before the transfer begins and persist unchanged after it completes. Transduction does not presuppose its terms. In transduction, the process constitutes its terms as it propagates. When structure propagates outward from a seed crystal, each newly formed crystalline layer serves as both the result of the previous step and the seed for the next. The crystal is not a pre-existing form imposed on passive matter. It emerges through the transductive process itself.

Biological development is transductive: each stage of embryonic differentiation creates the conditions for the next stage, and the plan of the organism is not contained in any single cell but in the propagating process itself. Psychological development is transductive: each new cognitive structure creates the conditions for new experiences, which create the conditions for new cognitive structures. Social evolution is transductive: each new institution creates new forms of interaction, which create new needs, which create new institutions.

The coupling of human and machine intelligence, Simondon's framework reveals, is transductive. When a human engages in sustained conversation with a large language model, the human's initial prompt is not a fixed message but a metastable field — an incomplete structure laden with implicit intentions, unstated contexts, half-formed associations. The machine's response does not merely answer the prompt. It transduces it — converts it into a new form that reveals implications and connections latent in the prompt but invisible to the human. This response becomes a new metastable field for the human, generating a new prompt that could not have existed without the machine's response. Each exchange creates the conditions for the next.

The distinction between transfer and transduction has direct implications for how we understand AI training data, AI outputs, and AI copying. The common claim that AI plagiarizes human creators makes sense within the transfer model — the human creates, the machine copies, the information moves from creator to imitator. But within the transductive model, the claim is incoherent. What the machine does with its training data is not copying — it is the propagation of structural patterns across domains, producing new configurations that were not present in any single source. The output may be derivative, low-quality, or ethically problematic, but it is not a copy. It is a transductive product.

Origin

The term transduction originates in signal processing, where it names the conversion of a signal from one form to another: microphones transduce sound waves into electrical signals; loudspeakers transduce electrical signals back into sound waves. The medium changes but the information — the pattern, the structure, the relational organization carrying meaning — is preserved across the transformation.

Simondon generalized this technical concept far beyond its engineering origin. In his framework, transduction names any process of structured propagation that constitutes its domains as it advances. The generalization is developed throughout the principal thesis on individuation but most explicitly in its introduction, where he identifies transduction as the fundamental operation underlying all individuation.

Key Ideas

Transfer and transduction differ fundamentally. Transfer moves fixed information between pre-existing terms. Transduction creates its terms through the process of propagation itself.

Novelty originates in transductive processes. Genuine novelty does not originate in individual entities but in the processes that cross boundaries between them and constitute new domains.

Human-AI collaboration is transductive. The meaning produced in deep engagement with AI is not transferred between fixed minds — it is a structuring activity that transforms both participants as it propagates.

Transduction transforms both terms. The human who emerges from genuine transductive engagement is not the human who entered it. Neither is the system.

The transfer model generates false problems. Debates about AI ownership, authorship, and copying often rest on a transfer-model assumption that the underlying process violates.

Debates & Critiques

Critics have asked whether transduction, as Simondon deploys it, is a rigorous concept or a metaphorical extension that loses precision in its generalization. The technical meaning in signal processing is clearly defined; the philosophical meaning is deliberately capacious. Defenders argue that the capaciousness is not a weakness but a feature — transduction names a structural feature of many different processes, and its applicability across domains reflects genuine structural homology rather than loose analogy.