Can AI Help Humeans? The Laws of Nature Debate in Light of Automated Scientific Discovery

Robert Michels (LanCog, Centre of Philosophy, University of Lisbon)

13 October 2023, 16:00 (Lisbon Time – WET)

Faculdade de Letras de Lisboa

Sala Mattos Romão (Departamento de Filosofia)

Abstract: According to the standard Humean theory of the laws of nature, Lewis’s Best System Analysis, laws of nature have this status at least partly as the result of an optimal trade-off between scientific values such as simplicity and descriptive strength. This idea has recently come under pressure, since — as authors like Roberts and Woodward have pointed out — there might, pace what Humeans like to suggest, be no such trade-off in the way laws of nature are identified in the natural sciences. Recent developments in the field of automated scientific discovery, in particular regarding symbolic regression, promise to provide Humeans with an answer to this challenge and, as we will argue, might even allow them to in turn put pressure on rival theories of the laws of nature: Symbolic regression gives us a method for (re-)discovering laws which closely matches the Humean picture of what makes a law of nature a law of nature and in particular crucially involves a trade-off between simplicity and descriptive strength. In this paper, we discuss whether Humeans can indeed rely on symbolic regression to bolster their theory of laws of nature. (This is joint work with Niels Linnemann [University of Geneva].)