Zhengguan Chen
University of Bern
This paper explores a novel approach to the debate on kindhood and its ontological status by drawing on a philosophically less familiar concept vital in deep learning: latent space. In machine learning, latent space is defined as a representation of data in the compressed form where similar points cluster together in space. It is widely used to reveal similarities, predict future values, and, increasingly, generate new data. The paper argues that there is a striking parallel between the deep learning models’ capacity to discover patterns of data and human’s ability to recognise material objects: the way the points with data features cluster together reflects the way humans classify objects into kinds. Building on this parallel, it provides a top-down strategy for analysing kindhood and implies that kinds cannot be fully constructed and explained by their closely related properties and the partial essences of kinds should be taken to assume the existence of kinds in our categorial ontology.
Chair: Satbhav Voleti
Time: 03 September, 15:20-15:50
Location: HS E.002
SOPhiA 2026 