Nevertheless, the operations linking documents, researchers and computational models can alter the circumstances in which problems are formulated. This is the more exact site of computational reciprocity. Its two sides are not equivalent: on one side, people establish questions, choose criteria, encounter places and accept responsibility; on the other, trained systems retrieve, rank, classify, transform and generate representations according to their technical architectures and deployment conditions. These operations may be mutually consequential without being biologically symbiotic. Manovich’s account of generative AI as a medium gives this relation formal specificity. Probabilistic variation enlarges the field of available comparisons while making exact control difficult; extensive cultural patterning makes certain associations readily available while potentially favouring aesthetic or conceptual convention. An unexpected comparison is therefore not evidence of insight until its source, scope and exclusions are examined. This difficulty also changes the question of artistic agency. Bomba and De Angeli use situated studio practice to contest a model of authorship in which all causality originates within a solitary individual; Rozental and de Rooij likewise show why the opposition between passive tool and independent collaborator neglects the prolonged adjustments made in practice. Neither argument requires treating software as conscious. Their most transferable insight concerns the distributed circumstances of production, which should not be confused with distributed ethical responsibility.