Vadim Keyser

In biomedical measurement, biomarkers are used to achieve reliable prediction of, and useful causal information about patient outcomes while minimizing complexity of measurement, resources, and invasiveness. A biomarker is an assayable metric that discloses the status of a biological process of interest, be it normative, pathophysiological, or in response to intervention. The greatest utility from biomarkers comes from their ability to help clinicians (and researchers) make and evaluate clinical decisions. In this paper we discuss a specific methodological use of clinical biomarkers in pharmacological measurement: Some biomarkers, called ‘surrogate markers’, are used to substitute for a clinically meaningful endpoint corresponding to events and their penultimate risk factors. We confront the reliability of clinical biomarkers that are used to gather information about clinically meaningful endpoints. Our aim is to present a systematic methodology for assessing the reliability of multiple surrogate markers (and biomarkers in general). To do this we draw upon the robustness analysis literature in the philosophy of science and the empirical use of clinical biomarkers. After introducing robustness analysis we present two problems with biomarkers in relation to reliability. Next, we propose an intervention-based robustness methodology for organizing the reliability of biomarkers in general. We propose three relevant conditions for a robust methodology for biomarkers: (R1) Intervention-based demonstration of partial independence of modes: In biomarkers partial independence can be demonstrated through exogenous interventions that modify a process some number of “steps” removed from each of the markers. (R2) Comparison of diverging and converging results across biomarkers: By systematically comparing partially-independent biomarkers we can track under what conditions markers fail to converge in results, and under which conditions they successfully converge. (R3) Information within the context of theory: Through a systematic cross-comparison of the markers we can make causal conclusions as well as eliminate competing theories. We apply our robust methodology to currently developing Alzheimer’s research to show its usefulness for making causal conclusions.

In experimental settings, scientists often “make” new things, in which case the aim is to intervene in order to produce experimental objects and processes—characterized as ‘effects’. In this discussion, I illuminate an important performative function in measurement and experimentation in general: intervention-based experimental production (IEP). I argue that even though the goal of IEP is the production of new effects, it can be informative for causal details in scientific representations. Specifically, IEP can be informative about causal relations in: regularities under study; ‘intervention systems’, which are measurement/experimental systems; and new technological systems.

In this paper, prediction markets that encourage traders to bet on matters of life and death are used to explore the varieties and dynamics of moral repugnance. We define moral repugnance as morally charged feelings of revulsion that correspond (correctly, incorrectly, and indeterminately) to moral reasons and contexts. Rich variations of moral repugnance and their dynamic qualities are presented by investigating the contextual frames in which they arise. These contextual frames constitute interacting conditions composed of information about states of affairs, moral reasons, and feelings of revulsion. Through careful study of two medical prediction markets that encourage betting on death, we can observe the interaction between these causal conditions to see how the varieties of moral repugnance emerge. We also present three interesting results that arise from analyzing the dynamics of moral repugnance in response to prediction markets. First, a prediction market can elicit several conflicting types of moral repugnance at the same time. Second, moral indeterminacy can arise in two different ways when judging prediction markets. Finally, some prediction markets can generate a moral endogeneity problem, a disruptive feedback loop between a given prediction market and the morally relevant outcome it predicts.

This discussion centers on two desiderata: the role of measurement in information-gathering and physical interaction in scientific practice. By taking inspiration from van Fraassen’s view, we present a methodological account of perspectival measurement that addresses empirical practice where there is complex intervention, disagreeing results, and limited theory. The specific aim of our account is to provide a methodological prescription for developing measurement processes in the context of limited theory. The account should be useful to philosophers of science, who are interested in the intersection between representation and intervention; scientists, who are interested in methodological suggestions for theory-development and reliability; and interdisciplinary researchers, who are interested in the intersections between the pandemic, built environments, and social processes. We apply the process-view of measurement to COVID-19, specifically, measuring replication in the SARS-CoV-2 virus. The aim is to show that our account tracks key elements—manipulation/intervention, independence, invariance, and theory-development—by organizing unfolding measurement processes. Additionally, we use our account to make prescriptive suggestions for measurement practice in the COVID-19 context by discussing the need to broaden measurement perspective on interaction, manipulation, and production.

Traditionally, in the literature on robustness analysis objects are classified as genuine phenomena (natural objects, events, and processes) or artifacts (results produced in error). But much of biological measurement requires the manipulation of local experimental conditions in order to produce new effects. These types of intervention-based regularities are neither natural objects nor artifacts; characterizing them as either fails adequately to address key ontological properties as well as their role in scientific practice. It is argued that a new classification, based on methodological considerations, can be useful in order to characterize experimental productions. Experimentally-useful context-sensitive objects are referred to as ‘artefacts’. To show how the new classification works and why it is instructive for scientific practice, two case studies are discussed. First, the puzzle of arsenic-consuming living organisms is analyzed, where under a set of specific experimental conditions a given organism was found to replace phosphorous with arsenic in its DNA. Second, ecological epigenetic measurement is discussed to show the complexity of variant effects in the context of lab, field, and computer measurements. It is argued that the line between artifacts and artefacts is fluid because theoretical, experimental, and practical considerations vary.

I show that in complex methodological contexts, representational and intervention-based roles require re-conceptualization. I analyze the relations between representation and intervention by focusing on the role of intervention in mediating representations. To do this, first I show how applied scientific practice challenges the simple distinction between representational and intervention-based roles of experiment/measurement. Then I discuss the complex interaction between representation and intervention applied to methodology in biomarker measurement.

Scientists often use multiple independent methods of identification to distinguish reliable results from those produced in error. This process is referred to as ‘robustness analysis’. I argue that even though robustness analysis is useful for differentiating natural phenomena from artifacts, it fails to differentiate experimentally produced effects from artifacts. I argue that to bypass this problem, we can re-frame the role of robustness analysis to focus on cross-comparison between methods of production. Focusing on the production relation provides information about how changes in conditions alter given effects, without first having to make a distinction between effect and artifact.

In experimental settings, scientists often “make” new things, in which case the aim is to intervene in order to produce experimental objects and processes—characterized as ‘effects’. In this discussion, I illuminate an important performative function in measurement and experimentation in general: intervention-based experimental production. I argue that even though the goal of IEP is the production of new effects, it can be informative for causal details in scientific representations. Specifically, IEP can be informative about causal relations in: regularities under study; ‘intervention systems’, which are measurement/experimental systems; and new technological systems.

In this discussion I show that philosophy of science concepts, especially where examples and thought experiments are limiting, can be enriched with artistic examples. I argue that artistic examples show abstract components and relations that can then be used to engage with philosophical concepts. First, I discuss a useful representational model for thinking about the process of science as analogous to the process of art. I set up philosophy of science as not only open, but also closely connected to art by using Giere’s and van Fraassen’s discussions of the connection between scientific and artistic perspectives. Second, I show how artistic examples can be engaging and informative for teaching philosophy of science concepts. I apply two artistic examples to the concept of quantum measurement ‘indeterminacy’: Jackson Pollock’s artistic process and Mark Dior’s fallen hemlock. Finally, I use anecdotal experience of creating art with a philosophy of science class in order to apply indeterminacy to social properties.

Recent work by Keyser in applied epistemology of experiment has focused on the iterative ‘production’ of knowledge: knowledge stabilizes within a given physical context and it is iteratively tested within that context to meet standards of reliability. This implies that in a given physical context (e.g., laboratory), the inferences, methods/techniques, and physical products form coherence relations with one another. We apply this epistemological stabilization account to the martial arts in order to argue that the context of stabilization dictates the training methods, techniques, and even the philosophical concepts of a given martial arts system. We describe key elements of stabilization pertaining to rules, training methodology, warrior philosophy, and spectator/media tropes. The account of stabilization requires a case study that shows how a dynamic system is stabilized. We describe the stabilization of modern mixed martial arts by tracking it on a time scale, but also in different contexts (e.g., Pancrase, Pride, UFC).

We explore multi-scale relations in artificial intelligence (AI) use in order to identify difficulties with coordinating relations between users, machine learning (ML) processes, and “sociobuilt contexts”—specifically in terms of their applications to medical technologies and decisions. We begin by analyzing a recent COVID-19 machine learning case study in order to present the difficulty of traversing the detailed causal topography of “sociobuilt contexts.” We propose that the adequate representation of the interactions between social and built processes that occur on many scales ought to drive interdisciplinary approaches for ML modification. Next, we describe ML algorithm development as a process that is partly dependent on methodological stabilization for reliability and partly on coordinating relations. In the coordination between user, ML process, and sociobuilt contexts, we propose that new methods can be explored that promote the inclusion of patients and communities for the purpose of cross-checking portions of the ML process. Finally, we suggest that the advantages of responsible innovation emerge through the iterative entanglement of ethical, methodological, and ontological considerations within the broader conceptual infrastructure of epistemic responsibility.