Artificial Intelligence in the form of different machine learning models is applied to Big Data as a way to turn data into valuable knowledge. The rhetoric is that ensuing predictions work well—with a high degree of autonomy and automation. We argue that we need to analyze the process of applying machine learning in depth and highlight at what point human knowledge production takes place in seemingly autonomous work. This article reintroduces classification theory as an important framework for understanding such seemingly (...) invisible knowledge production in the machine learning development and design processes. We suggest a framework for studying such classification closely tied to different steps in the work process and exemplify the framework on two experiments with machine learning applied to Facebook data from one of our labs. By doing so we demonstrate ways in which classification and potential discrimination take place in even seemingly unsupervised and autonomous models. Moving away from concepts of non-supervision and autonomy enable us to understand the underlying classificatory dispositifs in the work process and that this form of analysis constitutes a first step towards governance of artificial intelligence. (shrink)

Data science is characterized by engaging heterogeneous data to tackle real world questions and problems. But data science has no data of its own and must seek it within real world domains. We call this search for data “prospecting” and argue that the dynamics of prospecting are pervasive in, even characteristic of, data science. Prospecting aims to render the data, knowledge, expertise, and practices of worldly domains available and tractable to data science method and epistemology. Prospecting precedes data synthesis, analysis, (...) or visualization, and is constituted by the upstream work of discovering disordered or inaccessible data resources, thereafter to be ordered and rendered available for computation. Through this work, data science positions itself in the middle of all things—capable of engaging this, that, or any domain—and thus prospecting is a key driver of data science’s ongoing formation as a universal science. (shrink)

The article focuses on stories and storytelling practices as explanatory resources in standardization processes. It draws upon an ethnographic study of the development of a technical standard for data sharing in an ecological research community, where participants struggle to articulate the difficulties encountered in implementing the standard. Building from C. Wright Mills’ classic distinction between private troubles and public issues, the authors follow the development of a story as it comes to assist in transforming individual troubles in standard implementation into (...) an institutional issue for the ecological scientific community. The authors present the “hands-on” social science collaboration in this study as an example of a mechanism for supporting institutionalization of issues. Finally, the authors argue that narratives can serve as effective organizing principles within institutional settings, thereby providing an approach to understand the practical, substantive difficulties that occur in work with data in the sciences. (shrink)

The multifaceted work of the late Susan Leigh Star is explored through a selection of her writings and essays by friends and colleagues. Susan Leigh Star (1954–2010) was one of the most influential science studies scholars of the last several decades. In her work, Star highlighted the messy practices of discovering science, asking hard questions about the marginalizing as well as the liberating powers of science and technology. In the landmark work Sorting Things Out, Star and Geoffrey Bowker revealed the (...) social and ethical histories that are deeply embedded in classification systems. Star's most celebrated concept was the notion of boundary objects: representational forms—things or theories—that can be shared between different communities, with each holding its own understanding of the representation. Unfortunately, Leigh was unable to complete a work on the poetics of infrastructure that further developed the full range of her work. This volume collects articles by Star that set out some of her thinking on boundary objects, marginality, and infrastructure, together with essays by friends and colleagues from a range of disciplines—from philosophy of science to organization science—that testify to the wide-ranging influence of Star's work. Contributors Ellen Balka, Eevi E. Beck, Dick Boland, Geoffrey C. Bowker, Janet Ceja Alcalá, Adele E. Clarke, Les Gasser, James R. Griesemer, Gail Hornstein, John Leslie King, Cheris Kramarae, Maria Puig de la Bellacasa, Karen Ruhleder, Kjeld Schmidt, Brian Cantwell Smith, Susan Leigh Star, Anselm L. Strauss, Jane Summerton, Stefan Timmermans, Helen Verran, Nina Wakeford, Jutta Weber. (shrink)

This is a co-authored dialogue which explores epistemological and metaphysical questions raised by a social constructivist approach to science.

The narrative of the digital phenotype as a transformative vector in healthcare is nearly identical to the concept of “data drivenness” in other fields such as law enforcement. We examine the role of a prescription drug monitoring program in California—a computerized law enforcement surveillance program enabled by a landmark Supreme Court case that upheld “broad police powers”—in the interprofessional conflict between physicians and law enforcement over the jurisdiction of drug use. We bring together interview passages, clinical artifacts, and academic and (...) gray literature to investigate the power relations between police, physicians, and patients to show that prescribing data appear to the physician as evidence of problematic patient behavior by the patients, and to law enforcement as evidence of physician misconduct. In turn, physicians have adopted a disciplinary approach to patients, using quasi-legalistic documents to litigate patient behavior. We conclude that police powers have been used to pave data infrastructure through a contested jurisdiction, and law enforcement have used that infrastructure to enroll physicians into the work of disciplining patients. (shrink)

As the ability to meaningfully process increasingly large quantities of data has improved, the need for systems to support the aggregation and subsequent use of disparate smaller datasets is correspondingly greater. The GBIF is just one such project among a larger group seeking to aggregate the smaller, focused, and disparate sources of information generated for the work of science. GBIF is simultaneously an effort to coordinate and aggregate digital species occurrence data and digitize natural history collections into a single global-scale (...) resource for biodiversity work. The basis for this paper is a critical study of the GBIF database and data portal as a socio-technical system. (shrink)

Residual categories are those which cannot be formally represented within a given classification system. We examine the forms that residuality takes within our information systems today, and explore some silences which form around those inhabiting particular residual categories. We argue that there is significant ethical and political work to be done in exploring residuality.