Writing Against the Machine: Computational Authorship and Historical Writing

Abstract

Historians generate knowledge through the labour of composition – through the friction between interpretation and evidence that makes claims open to scrutiny and challenge. This essay argues that when composition is bypassed, that structure disappears. Generative AI raises this issue in urgent fashion. Current large language models produce what the essay terms ‘stochastic history’: prose that replicates the surface forms of historical explanation without enacting the disciplinary reasoning behind them. Such prose flattens temporal complexity into chronological adjacency, inherits narrative patterns without deliberating over them and reproduces hegemonic framings without the mechanisms – archival friction, peer contestation and historiographical consciousness – through which the discipline revises itself. Recent studies measuring AI's applicability to historical work capture transmissive tasks while remaining blind to the interpretive core; approaches that identify textual markers of historical thinking detect symptoms that can be simulated, not the compositional process producing them. The consequences extend beyond the profession. When stochastic history circulates in classrooms, policy research and public media, non-specialists encounter pasts stripped of contingency and contestability – pasts that naturalize present arrangements rather than rendering them open to challenge. The defence of writing advanced here is methodological rather than nostalgic: It preserves the conditions under which historical claims can be scrutinized and revised.