<p>Current debates on artificial intelligence often treat predictive performance, fluency, and task success as sufficient indicators of intelligence. We argue that this misses something more basic. What is increasingly governed as intelligence is often a historically specific abstraction: a form of predictive capacity detached from sustained exposure to correction, resistance, and real-world effects. Yet the problem is not only conceptual inflation. It is also historical. Contemporary AI gains traction in institutions that increasingly lack the time, labour, and organizational patience required to form judgment in the stronger sense. Under such conditions, predictive adequacy becomes attractive not because it is equivalent to intelligence, but because it can substitute for slower practices of interpretation, verification, and deliberation that are no longer well sustained. We use answerability to name a thicker relation of exposure: to others, to contestation, to institutional limits, and to effects that cannot simply be offloaded once a judgment has been made. Our concern, then, is less with whether contemporary AI truly counts as intelligent than with why institutions increasingly treat predictive systems as if they do, and what this reclassification authorizes. We also suggest that the eclipse of older cybernetic vocabularies is revealing here: if cybernetics named feedback, control, and adaptive adjustment under constraint, “artificial intelligence” offered a more elevated and institutionally useful public ontology. AI, in this sense, does not simply imitate intelligence. It participates in a reorganization in which the time needed to form judgment is increasingly treated as an unaffordable interruption, while prediction can circulate with the prestige of intelligence and the authority of judgment.</p>