Mel Andrews

This paper contends with the notion that the methods of machine learning (ML) are unique among the tools of science in enabling a form of theory-free inductive inference. I challenge these assertions of epistemic distinctness, attributing the prevalence of these views to an untenable conception of scientific objectivity: what I term a theory-free ideal, in homage to its normative counterpart. ML, as a formal method of induction, must rely on conceptual or theoretical resources to get inference off the ground. By means of two case studies, I argue that this theory-free ideal has a deleterious effect on the epistemic standing of ML-involving science.

Machine learning (ML) refers to a class of computer-facilitated methods of statistical modelling. ML modelling techniques are now being widely adopted across the sciences. A number of outspoken representatives from the general public, computer science, various scientific fields, and philosophy of science alike seem to share in the belief that ML will radically disrupt scientific practice or the variety of epistemic outputs science is capable of producing. Such a belief is held, at least in part, because its adherents take ML to exist on novel epistemic footing relative to classical mathematical or statistical modelling approaches utilised in science. Namely, they take modelling with ML to be a “theory-free” enterprise, in the sense of not resting essentially on input from human conceptual grasp on the target phenomenon and domain expertise. I take this view to arise from the further, and more deeply entrenched belief that data is worldly and objective; i.e., data is viewed as recapitulating or representing with perfect fidelity the structure or properties of the systems in nature it is sampled from. Yet most contemporary philosophers of science take on board, in one version or other, the thesis of theory-ladenness or theory-mediation of observation or measurement. From this, it follows that most philosophers of science (and, I will venture, many scientists) hold irreconcilable views on the nature of data, an internal tension which threatens the integrity of their appraisals of ML in science. Taking the thesis of theory-ladenness on board---and its implications for the nature of data seriously---it follows that there is no reason to believe that ML differs fundamentally in its epistemic footing from established mathematical modelling approaches in the sciences. I show that usage and interpretation can differ from “standard practice” in scientific projects which wield the tools of ML without accepting a difference in the epistemic or representational status of such tools.

Much has been written about the free energy principle (FEP), and much misunderstood. The principle has traditionally been put forth as a theory of brain function or biological self-organisation. Critiques of the framework have focused on its lack of empirical support and a failure to generate concrete, falsifiable predictions. I take both positive and negative evaluations of the FEP thus far to have been largely in error, and appeal to a robust literature on scientific modelling to rectify the situation. A prominent account of scientific modelling distinguishes between model structure and model construal. I propose that the FEP be reserved to designate a model structure, to which philosophers and scientists add various construals, leading to a plethora of models based on the formal structure of the FEP. An entailment of this position is that demands placed on the FEP that it be falsifiable or that it conform to some degree of biological realism rest on a category error. To this end, I deliver first an account of the phenomenon of model transfer and the breakdown between model structure and model construal. In the second section, I offer an overview of the formal elements of the framework, tracing their history of model transfer and illustrating how the formalism comes apart from any interpretation thereof. Next, I evaluate existing comprehensive critical assessments of the FEP, and hypothesise as to potential sources of existing confusions in the literature. In the final section, I distinguish between what I hold to be the FEP—taken to be a modelling language or modelling framework—and what I term “FEP models.”