Alexander Martin Mussgnug

This paper considers cases where AI is used to automate tasks perceived to be “mere routine” and inconvenient to researchers. We term such uses “Convenience AI” and characterise the set of assumptions involved in making claims of convenience in relation to the role of AI in scientific discovery. While Convenience AI can save resources and give rise to novel forms of inquiry, our analysis underscores how its uncritical adoption bears risks for knowledge production in science, also because it often betrays the very notion of convenience that it is meant to foster. Critically, we show how the consistent association of Convenience AI with the goals of productivity, efficiency, and ease can itself contribute to the epistemic challenges posed by AI.

The rapid proliferation of artificial intelligence (AI) and machine learning (ML) in science raises profound epistemological and ethical questions. Over the past decade, philosophical scholarship has engaged extensively with the challenges surrounding AI applications. Research is, however, often somewhat divided. On the one hand, work in the philosophy of science centres largely around the epistemic dimensions of AI applications. For instance, philosophers of science have explored at length how the opacity of deep learning models impacts scientific discovery and debated the potential of fully automated knowledge generation through AI. On the other hand, moral considerations surrounding the development and use of AI are most comprehensively discussed in the field of AI ethics. Within it, scholars have investigated issues such as the fairness and ecological sustainability of AI applications, or the degree of moral responsibility attributable to AI developers. Epistemic and ethical concerns surrounding AI in science, however, are often inextricably linked. This is particularly apparent in the social sciences where AI applications are intended to foster better understanding of complex value-laden concepts and are employed to aid public policy, resource allocation, and social interventions. One such field is development economics, where ML models are used to predict poverty statistics. While these applications promise to revolutionize data generation, offer unprecedented opportunities to analyse complex social phenomena, and unlock novel possibilities for downstream international development aid and policymaking, they also present new challenges. What epistemic and moral standards should we hold machine learning poverty predictions to, in light of their high-stakes application in international development? How does the use of AI affect development economists’ understanding of poverty as a value-laden concept? How does the increasing ease and speed of data generation and analysis shape epistemic authority and moral deliberation in the study of poverty? What ethical norms should we rely on when novel technological tools, used by scientists to estimate poverty, inform policies that impact vulnerable communities? In three introductory chapters and four independent research papers, this dissertation explores these questions at the intersection of the philosophy of science and AI ethics. The first two papers provide detailed analyses of the concrete practices of machine learning poverty predictions. The first paper “The predictive reframing of machine learning applications” problematizes the common framing of machine learning applications in development economics as data prediction rather than measurement problems. The second paper “AI Operationalism” outlines a strategic but problematic role for researchers’ tendency to conflate concept and operationalization in machine learning applications in science, such as equating the concept of poverty with a particular poverty metric. The latter two papers engage with machine learning poverty prediction and medical AI in order to explore broader themes surrounding the ethics and epistemology of AI in science. My coauthored paper “Convenience AI” argues that the consistent association of certain AI applications with the goals of productivity, efficiency, and ease can lower critical scrutiny of scientific research processes and shift focus away from appreciating their broader epistemic and social implications. The last paper “Technology as uncharted territory” highlights how a tendency to view AI as a distinct and new moral realm can come at the expense of scientists, developers, policymakers, and ethicists engaging with established norms and virtues that were gradually cultivated to promote successful and ethical practice within concrete social contexts. Together these four papers aim to contribute insights that can aid both the responsible development and deployment of machine learning poverty predictions and the ethically and epistemically beneficial integration of AI in science.

Recent research illustrates how AI can be developed and deployed in a manner detached from the concrete social context of application. By abstracting from the contexts of AI application, practitioners also disengage from the distinct normative structures that govern them. As a result, AI applications can disregard existing norms, best practices, and regulations with often dire ethical and social consequences. I argue that efforts to promote responsible and ethical AI can inadvertently contribute to and seemingly legitimize this disregard for established contextual norms. Echoing a persistent undercurrent in technology ethics of understanding emerging technologies as uncharted moral territory, certain approaches to AI ethics can promote a notion of AI as a novel and distinct realm for ethical deliberation, norm setting, and virtue cultivation. This narrative of AI as new ethical ground, however, can come at the expense of practitioners, policymakers, and ethicists engaging with already established norms and virtues that were gradually cultivated to promote successful and responsible practice within concrete social contexts. In response, I question the current narrow prioritization in AI ethics of moral innovation over moral preservation. Engaging also with emerging foundation models and AI agents, I advocate for a moderately conservative approach to the ethics of AI that prioritizes the responsible and considered integration of AI within established social contexts and their respective normative structures.

Supervised machine learning has found its way into ever more areas of scientific inquiry, where the outcomes of supervised machine learning applications are almost universally classified as predictions. I argue that what researchers often present as a mere terminological particularity of the field involves the consequential transformation of tasks as diverse as classification, measurement, or image segmentation into prediction problems. Focusing on the case of machine-learning enabled poverty prediction, I explore how reframing a measurement problem as a prediction task alters the primary epistemic aim of the application. Instead of measuring a property, machine learning developers conceive of their models as predicting a given measurement of this property. I argue that this _predictive reframing_ common to supervised machine learning applications is epistemically and ethically problematic, as it allows developers to externalize concerns critical to the epistemic validity and ethical implications of their model’s inferences. I further hold that the predictive reframing is not a necessary feature of supervised machine learning by offering an alternative conception of machine learning models as measurement models. An interpretation of supervised machine learning applications to measurement tasks as _automatically-calibrated model-based measurements_ internalizes questions of construct validity and ethical desirability critical to the measurement problem these applications are intended to and presented as solving. Thereby, this paper introduces an initial framework for exploring technical, historical, and philosophical research at the intersection of measurement and machine learning.