Ludovica Schaerf

Can, and should we, put a computational/mathematical concept on trial? What if its impact on culture and the environment is increasingly felt worldwide, even if it’s a depersonalised, distributed, abstract, and strange object? Does it bear responsibility? With this performance, we subject latent space(s) to a juridical process. Four researchers, each working on distinct dimensions of latent spaces, namely technical, cultural, philosophical, material, and aesthetic, while attending a round table, are thrown into a scholarly trial where prosecution, defence, and witnesses stage an intellectual confrontation. This trial — where the accused is absent, yet everywhere in its immaterial pervasiveness — makes explicit the performativity and theatricality that are dominant yet understated in the academic and scholarly worlds.

Representation, in its most rudimentary sense of ‘standing for’ something else, operates as a fundamental yet profoundly complex mediation within contemporary AI models. These systems do not simply process information but re-elaborate it through a sequence of transformations, producing what we call deep representations: encodings that are simultaneously layered, interconnected, and polysemic. Drawing on Stable Diffusion as our case study, this work traces the representational journey from world to model, examining how each layer of mediation reshapes what ‘standing for’ means. We begin with the mediations that precede the model: data as capta rather than given, digital image as cultural object and computational substrate, dataset as epistemic border, and pre-processing as structural constraint. We then turn to transformations within the model itself: the training, loss functions, and material economies that shape learned representations, differing fundamentally from standardized compression. We conclude by questioning whether the inherited vocabulary of representation remains adequate to describe what these systems actually do.

Metaphors surrounding artificial intelligence increasingly inform contemporary discussions on the topic, shaping both technical conceptualizations and cultural narratives. This paper explores literary metaphors by Jorge Luis Borges and Italo Calvino, arguing that their narratives offer critical conceptual frameworks for understanding deep learning’s representational structures. Drawing on five emblematic metaphors from Borges and Calvino—the Aleph (dimensional compression), the Library of Babel (combinatorial archive), the Chessboard (rule-based combination), the Atlas (emergent potential), and the Garden of Forking Paths (branching potentiality), this paper suggests that these literary metaphors reflect key phenomena within latent spaces and attention modules, such as the reduction of knowledge in one abstract point, dimensionality, and the combinatorial or potential logic. By aligning literary analysis with computational architectures, this paper highlights the significance of literary imagination in articulating deep learning’s capabilities and limitations, challenging simplified or anthropomorphic interpretations.