The Global Data Supply-Chain for Artificial Intelligence and Its Paradoxes

Data are a crucial aspect of deep learning models and the groundwork for contemporary artificial intelligence. While it is common to point out biases within datasets, far less attention has been given to the global relations that make data and the limits of representing complex scenarios associated with conflict. This paper makes three interventions: (1) it describes how the idea of a global supply chain applies not only to the semiconductors needed for large-scale AI workloads but also to data; (2) it points out that a global data supply chain cannot be analogized to natural resources because data are representations; and (3) it considers the limitations of large-language models to address complexity and uncertainty through a ChatGPT analysis of the current conflicts in the Sahel. Without deeper consideration of the limits of data collection and uncertainties that will persist in conflict, military AI will not function as expected. Worldwide, we need deeper debate about the purpose of military AI, as the technology alone cannot be an end unto itself.