The difference a method makes: methods as epistemic objects in computational science

ABSTRACT

Computational science is intrinsically interdisciplinary; the methods of one scientist may be the objects of study for another. This essay is an attempt to develop an interdisciplinary framework that can analyse research into methods as a distinctive kind of epistemic orientation in science, drawing on two examples from fieldwork with a group of specialists in computer modelling. Where methods for simulation are objects of research in their own right, they are distinct in kind to the objects of simulation, and raise a different set of sociological and philosophical questions. Drawing on the historian Hans-Jorg Rheinberger’s theory of epistemic objects, I ask: what kind of epistemic object does a method make, and how is research organized around it? I argue that methods become objects of research as purposeful things, in terms of their enrolment in the intentional structure of the experimental system. And, as methods research tends to be interventionary, in the sense that its mode of study creates and modifies its objects, we therefore observe a practical recursion, a dynamic of scientific reinvention, a ‘tuning’ of experimental systems that sheds light on the form of these systems’ historicity, their differential self-reproduction.

KEYWORDS:

• Epistemic things
• science and technology studies
• methods
• scientific practice
• abstraction
• radical empiricism
• recursion
• computational science

Disclosure statement

No potential conflict of interest was reported by the author.

Notes on contributor

Matt Spencer is an Assistant Professor at the University of Warwick's Centre for Interdisciplinary Methodologies. He has a background in social anthropology and conducts research in science and technology studies.

Notes

1 The term ‘experimental system’ is taken from Rheinberger (1997b), on whose theoretical framework I rely in this paper. Methods are commonly taken as objects of sociologists’ or philosophers’ study: the power of Rheinberger’ theory lies in enabling the analysis of methods as scientists’ own objects of study.

2 Rheinberger uses both ‘epistemic thing’ and ‘epistemic object’ more or less interchangeably. To avoid multiplication of terms, I follow Knorr-Cetina and primarily use ‘object’ for the purposes of the present discussion (Knorr-Cetina 2001).

3 One of the fundamental issues in numerical methods is finding ways to solve problems that are defined in calculus (i.e. in terms of the continuum) in the discrete mathematics that describes the operation of digital computers.

4 One of the central observations of Latour and Woolgar’s classic study of endocrinology research practice was this relation between becoming a fact and shedding of modalities that refer it to its conditions of production (Latour and Woolgar 1986). ‘[T]here is an essential difficulty writing the history of a fact; it has, by definition, lost all historical reference.’ (1986, 108)

5 Although Moore’s Law continues to drive increasing density of transistors on a chip, the power of individual processors has stabilized, so Moore’s Law translates to increasingly parallel computing rather than faster individual processors (Sutter 2005).

6 ‘Specification’ in this context refers to the ‘high level’ design of a component, what it does.

7 When physical systems are modelled in simulation work, simulations are not simply created in their full complexity. They are, instead, built up iteratively, developing a set of verified, validated and benchmarked ‘precursor’ simulations of increasing complexity, that compare against other data models, encoding some of these into automated tests for new functionality, and developing visualisation and diagnostic techniques to provide means of working with the output data. These processes test the relationship between the simulation and the ‘target’ system it represents. There is therefore a threefold relation of representation in any theory-derived simulation, triangulating the physical theory with the target system and the computational system. I focus here on just one side of the triangle: that between physical theory and the computational system, because it is the way that this relationship is organized by the abstractions of the modelling framework, that was at stake in the Firedrake project. Other infrastructural projects could unpick the other sides. For example, the development of automated testing frameworks organize verification and validation of simulations, by automating judgement against comparator datasets.