Send message to Authors

Short abstract:

This paper explores how we can understand and engage critically with contemporary initiatives in the incorporation of Large Language Models (LLMs) into military warfighting assemblages.

Long abstract:

This paper asks how we might make sense of contemporary initiatives in the incorporation of Large Language Models (LLMs) into warfighting assemblages. In promotional scenarios, LLMs are envisioned to automate – and accelerate – the generation of so-called Courses of Action (COA) or plans for operational command. Empirically the paper is based on a reading of Palantir’s battlefield management system called Artificial Intelligence Platform for Defence (AIP), Scale’s “AI Digital Staff Officer” Donovan, and DARPA’s novel warfighting concept Mosaic Warfare. Through an analysis of US military doctrines, we situate these systems in a long-standing Western – and in particular American – military imaginary that has placed ‘the battle’ at the center of how war is understood and practiced. By showing how these imaginaries are inscribed in Palantir, Scale and DARPA’s visions of LLM-enabled warfighting, we trace how the automated generation of ‘objective’ knowledge about the enemy is fundamental to the martial dream of immersing analysts and operators in data worlds. Mapping the ‘end-to-end’ generation of COAs, from data production and analysis to decision-making, we argue, raises critical questions regarding the automation of military (operational) logics. Within these automated processes of trial and error is a highly distributed and messy agency, which paradoxically reproduces the logics of militarism at the same time that warfighting is made faster, deadlier, and less controllable.