Jundi Cui

Large language models are commonly judged by whether they possess minds or merely reproduce statistical regularities. This opposition leaves a prior question: can intelligence operate in a technical system that is not a subject? This article develops the concept of objectified intelligence to answer it. Discursive intelligence is an organized capacity to respond to relations of evidence, relevance, and inference whose significance is established in public practices. Training can help form a technical organization capable of sustaining the relevant selective dependence. Such an organization realizes the capacity when changes to evidential relations redirect its response while variations that preserve those relations leave the response correspondingly stable. When products of the relevant practice have helped form that organization, the capacity has acquired technical-object form. Existing evidence supports several dimensions of this attribution within bounded domains and makes the full criterion empirically testable. Objectification differs from subjectivation. A capacity is objectified when relations formed in practice are reorganized in a technical bearer that can carry them into new exercises. It takes subjective form when its exercises enter the normative history of a continuing position, so that earlier commitments constrain later activity and authorized revision changes what the position may maintain. An LLM that sustains the relevant dependence can help form discursive content without thereby becoming a subject of judgment. Under proprietary service provision, this separation takes a political-economic form. Platforms control a reusable general capacity, whereas downstream institutions particularize its exercises and answer for the judgments they authorize. The resulting control–accountability asymmetry is one capitalist organization of objectified intelligence, not an inherent consequence of objectification.