VIOLENT HERMENEUTICS: AI AND THE WEAK TECHNICAL EVENT

Abstract

This paper investigates how large language models (LLMs), particularly those based on transformer architectures, inadvertently restage key insights from Heidegger’s and Lacan’s theories of language while simultaneously revealing their limitations. By foregrounding the associative and metaphorical dynamics at the heart of both LLMs and continental accounts of language, the essay situates machine learning within a broader hermeneutic framework. It discusses how vectorial associationism parallels Heidegger’s alsrelation and how attention mechanisms resonate with Lacanian metaphor, enabling local reconfigurations of meaning. Yet, unlike human language, the model’s ‘worlding’ remains a contingent performance under inscription, without an inner horizon of disclosure. What emerges is not Badiou’s Event but a weak technical event: a temporary redrawing of adjacency and accessibility within latent space. This fragility of articulation underscores the infrastructural role of metaphor, poetics, and style in shaping machine outputs, making literary practices newly central to the technological condition. In turn, I argue that human–AI interaction is best understood not through the lens of ideology or subjectivity, but as a form
of literacy or hermeneutics where inscription reorganizes the calculable field.