Are Predictive Models Epistemically Superior to Causal Ones?

Abstract

This thesis investigates the complex and often contentious question of whether predictive models are epistemically superior to causal models. It challenges the simplistic dichotomy that often frames this debate, arguing that the epistemic superiority of a model is not an intrinsic property but is contingent upon the specific epistemic goals and context of inquiry. Drawing on a multi-disciplinary framework that integrates philosophy of science, epistemology, and computer science, this thesis proposes a novel, multi-dimensional framework for evaluating the epistemic standing of predictive and causal models. This framework is grounded in a nuanced understanding of epistemic virtues, such as accuracy, robustness, and explanatory power, and is contextualized within Judea Pearl’s Causal Hierarchy. We argue that a richer understanding of the epistemic landscape reveals a more symbiotic relationship between prediction and causation than is commonly acknowledged. The thesis develops a novel perspective, the “Epistemic Parity and Complementation” (EPC) principle, which posits that while predictive and causal models may have different epistemic strengths, they are often epistemically complementary and can be leveraged in a synergistic manner to achieve a more comprehensive understanding of complex phenomena. Through a combination of theoretical analysis, mathematical modeling, and empirical case studies, this thesis aims to provide a more sophisticated and holistic understanding of the relative epistemic merits of predictive and causal models, ultimately arguing against a universal claim of epistemic superiority for either approach.