Editors’ Summary: In this paper, River Rain and Houda Lamqaddam consider how recommender systems can be reframed as speculative tools for humanistic inquiry rather than commercial personalization. By demonstrating how a fine-tuned computer vision pipeline maps visual similarities across 2,341 animated films, they highlight how algorithmic proximity can trace artistic influence and macro-level aesthetic shifts. Specifically, they identify how the stochastic nature of machine learning can be embraced as an interpretive provocation rather than a benchmark error to be corrected. They propose a methodology, combining a VGG16 network with vector search, that attempts to solve the problem of exploring large-scale digital heritage collections without relying on rigid, predictive “ground truth.”
