The Time of the Latent: Evaluating Cultural AI Through Human–AI Creative Trajectories

Generative AI is increasingly evaluated as a cultural technology, yet most evaluation still treats the artifact as the primary unit of analysis. Building on computational hermeneutics, this paper argues that cultural evaluation should also examine the temporal path through which a human–AI artifact becomes meaningful: intention, prompting, branching, variation, rejection, selection, montage, and reception. I call this path a latent trajectory: not a direct measurement of internal model states, but an observable trace of how a creative process traverses, narrows, and interprets AI-generated possibilities. The paper introduces the Latent Trajectory Protocol, a qualitative framework for documenting situatedness, divergent exploration, convergent refinement, temporality of judgment, interpretive cooperation, curatorial closure, and semantic safety. A compact AI-cinema worked example shows how the protocol can transform a practice-based creative process into an evaluable trace. The central claim is that a positive cultural AI is not one that maximizes output, but one that preserves the human time of exploration, refusal, and responsible choice.