Oscar Westerblad

We claim that scientists working with deep learning (DL) models exhibit a form of pragmatic understanding that is not reducible to or dependent on explanation. This pragmatic understanding comprises a set of learned methodological principles that underlie DL model design-choices and secure their reliability. We illustrate this action-oriented pragmatic understanding with a case study of AlphaFold2, highlighting the interplay between background knowledge of a problem and methodological choices involving techniques for constraining how a model learns from data. Building successful models requires pragmatic understanding to apply modelling strategies that encourage the model to learn data patterns that will facilitate reliable generalisation.

Scientists strive to understand the world. Traditionally, philosophers of science have thought that this is a matter of constructing explanations, based on theories and laws, thereby gaining understanding of phenomena by explaining them. This thesis takes a radically different approach, instead relating the notion of understanding to the activities that scientists perform. Scientific understanding is not just a matter of representing or explaining the world, but a matter of practical and intelligent doing. Philosophers of science have continued to sell short the project of making sense of scientific understanding, limiting their views to the articulated and well-formed products of inquiry, like theory and explanation. In taking a pragmatist approach to the epistemology of understanding, I reorientate our theorising about understanding from products to processes, and from explanations to methods and concepts. I thereby provide an ecumenical account of scientific understanding that can accommodate the dynamics and heterogeneity of scientific practice. In Chapter 1, I argue against the dominant focus on explanatory understanding in philosophy of science, thereby making room for my own account of operational understanding. Chapter 2 provides an account of how concepts are essential constituents of understanding. Concepts frame entities so that scientists can think and talk about and interact with them intelligently. Chapter 3 gives an account of understanding-how, through my notion of operational understanding, which relates to the methods that structure and guide successful performances of activities. In Chapter 4, I take up the task of describing the psychology of understanding, the kind of state that understanding is, which I argue is accounted for in terms of skill. In Chapter 5, I develop an account of how cognitive devices — tools for thinking, like models and thought-experiments — facilitate and produce understanding.

Reconstruction is a central notion in Dewey’s account of inquiry and in his metaphilosophical commitments. In his work, Dewey made a call for reconstruction of philosophy, in the reconstruction of central notions of the discipline, like knowledge, logic, truth, the good, reason, and experience. Inquiry itself is reconstructive, according to Dewey, involving the transformation of an indeterminate situation into one which is determinate and understood. Dewey’s philosophical views should therefore be of interest to those taking part in the recent turn towards revisionary philosophical methodologies, like conceptual engineering, explication, and amelioration. In light of the recent developments in revisionary methodologies, I aim to explore Dewey’s conception of concepts in relation to the conceptual engineering literature, suggesting that it provides a useful conception of concepts for conceptual engineering, and to suggest a Deweyan model of philosophical inquiry where concepts are the objects of inquiry. In both cases, I hope to show a productive interaction between Dewey’s ideas and contemporary discussions of conceptual engineering.