Abstract

Our proposal brings together two approaches—design theory and design practice—to critically interrogate current modes of algorithmic prediction. We take the diagrammatic nature of machine learning as an entry/meeting point. This offers a design-driven understanding of computational prediction, and allows us to propose alternative strategies rooted in speculative methods, divinatory practices, and imaginative storytelling. Machine learning algorithms operate in multi-dimensional mathematical space; they create knowledge through operations, comparisons, and transformations of vectorised data. The shape of this space—and therefore the scope of predictions that can be made from it—is constrained by the training data and by the wide range of statistical operations and linear algebra that machine learning performs. In this current scenario, causality is superseded by a correlation-based type of rationality that predicts occurrences of phenomena as literal 'patterns' rather than searching for causes and allowing for contingency. This process has profound implications for what counts as knowledge as it forecloses the space of potential—what might happen or might not happen. By drawing on selected aspects of Deleuze's thought we discuss ways of 'diagramming' alternative narratives of these spaces of potential in order to reclaim algorithmic prediction as a productive mode of speculation; one that is able to predict radically new futures. The aim is to design a form of diagram-making that is liberating, enabling of the new and, crucially, able to actualize the very potential otherwise captured by contemporary apparatuses of algorithmic prediction. Invited talk. Presented at the Panel "Design Anthropology: Uniting experience and imagination in the midst of social and material transformation". Art, Materiality and Representation Conference. British Museum. With David Benqué.