132. The Algorithmic Translation of Expertise: Machine Learning and Expert Practices
Wanheng Hu, Cornell University
Posted: February 28, 2022 Accepted Languages: English/Inglés/Inglês
This panel explores an enduring theme in Science and Technology Studies in light of an emerging sociotechnical trend: the cultivation of machine learning algorithms to cope with expert tasks. In recent years, machine learning and artificial intelligence (AI) have been introduced to intervene in a growing range of routine yet complicated tasks that typically require significant professional knowledge, skills, and experience, such as medical diagnosis, drug development, securities trading, and criminal justice. Unlike the traditional symbolic approach to AI as exemplified by expert systems, today’s machine learning techniques in AI rely on large quantities of training data often annotated by humans, from which the algorithms will “learn” hidden patterns on their own and can then generate predictions over new data.
While studies of expertise have long been central to the STS scholarship, the use of machine learning to conduct expert practices raises complicated, novel issues around knowledge, expertise, and credibility, among others. The development of machine learning systems involves a process of translating human expertise into a new algorithmic form, and this panel seeks to bring into conversation scholars interested in the “algorithmic translation of expertise.” The panel welcomes contributions exploring such questions as: What are the assumptions about human expertise that underlie machine learning? How are domain experts mobilized for developing task-specific machine learning algorithms? How does the credibility of these algorithms get negotiated, and, in turn, how do they affect the credibility of human experts? Finally, how will the rise of machine learning reshape expert practices as well as our understanding of expertise?
