The Relevance of Carnap’s Inductive Logic for Supervised Classification

Abstract

Recent advances in the field of machine learning have cast new light on the long-standing debate within the philosophy of science between falsificationists and inductivists. As a result of many failed attempts at justifying inductive inference, falsificationism has tended to be preferred over inductivism. However, Gillies (1996, 2003) claims that the success of certain machine learning procedures necessarily leads to the conclusion that inductivism has shown to be successful . In this thesis I will challenge this claim by examining how criticism voiced against the inductive logic of one of the central figures of inductivism, Rudolf Carnap, might still be relevant for modern day machine learning procedures. Since a general claim about all machine learning procedures would be prone to counterexamples this thesis will be limited to only linear models for supervised classification. I will consider how three well-known critiques that strongly undermined Carnap’s inductive methods might still apply to such classification models.