Large language models (LLMs) are increasingly described as mirrors, colleagues, and rivals in academic research. This paper takes these three roles as phenomenological entry points rather than fixed categories. As mirrors, LLMs reflect sedimented linguistic, cultural, and epistemic patterns embedded in their training data, often revealing biases and structures that remain implicit in human discourse. As colleagues, they appear to participate in processes of writing, reasoning, and interpretation, giving rise to new forms of collaboration that challenge traditional notions of authorship and expertise. As rivals, they destabilize established epistemic norms, raising concerns about evaluation, originality, and the reliability of knowledge production. Drawing on Edmund Husserl, Maurice Merleau-Ponty, and Don Ihde, the paper argues that these roles are not independent, but emerge from a deeper reconfiguration of epistemic agency. LLMs do not simply reflect, assist, or compete with human cognition; they mediate and redistribute it.
In this respect, they can be understood as a novel form of technical memory, in the sense developed by Bernard Stiegler, where processes of retention and anticipation are increasingly externalized and operationalized. By generating plausible continuations of text, LLMs function as anticipatory systems that pre-structure the field of possible thought. Thinking thus becomes a distributed process shaped by suggestion, iteration, and selection, rather than the expression of an isolated subject. This transformation resonates with contemporary concerns about cognitive deskilling, as discussed by Shannon Vallor, insofar as epistemic labor is progressively delegated to computational systems.If LLMs act simultaneously as mirrors, colleagues, and rivals, how should responsibility, authorship, and expertise be redefined? The paper proposes to understand them as constitutive elements of a hybrid epistemic ecology, in which human and artificial contributions are inseparably intertwined. By articulating the tensions between these three roles, the paper offers a phenomenological account of LLMs as mediators of thought, capable of reshaping both the practice and the norms of research in the digital humanities.
Speaker: Riccardo Valenti