Gunkel paper reads LLMs as Derrida's différance engines
TL;DR
- David J. Gunkel of Northern Illinois University argues in AI & SOCIETY that GPT-type transformer LLMs 'actualize' Jacques Derrida's concept of différance.
- The essay reframes LLM output as statistical differences computed across massive textual corpora, enacting Derrida's 'spacing, temporalization, and trace.'
- Gunkel walks the consequences through logocentrism, authorship, the metaphysics of presence, and the 'death of the author' in AI-generated text.
A philosophy journal has handed transformer language models a new label: différance engines. In AI & SOCIETY, Northern Illinois University's David J. Gunkel argues that GPT-type large language models "actualize Jacques Derrida's concept of différance." Différance, in the poststructuralist account, treats meaning as a system of differences and deferrals rather than stable reference.
The claim is mechanical, not metaphorical. Transformer architectures, Gunkel writes, "generate meaningful content by calculating statistical differences across massive textual corpora," foregrounding the moves of "spacing, temporalization, and trace" that Derrida theorized. From there the essay walks the philosophical fallout: challenges to logocentrism, authorship, and the metaphysics of presence, alongside the "death of the author" in AI-generated text.
The paper runs pages 1991 to 2001 in volume 41, issue 3. The two-way read is the hook. LLMs get explained through poststructuralism, and poststructuralism, Gunkel argues, is "clarified and rendered accessible through the technical operations of contemporary AI." A pair of researchers in our directory had circulated the piece before it reached the print issue.
Originally reported by link.springer.com
Read the original article →Original headline: The différance engine: large language models and poststructuralism - AI & SOCIETY