Reframing AI: Three studies of AI in the laboratory life of biotechnology

This dissertation challenges three common beliefs about artificial intelligence. First, it disputes the view of AI as autonomously learning from data, instead emphasizing the human effort and creativity involved in making AI systems work. Second, it critiques the digital-centric view of AI by drawing attention to the material culture in which AI is embedded—highlighting the role of physical objects, bodily practices, and infrastructure. Third, it questions the assumption that AI is primarily aimed at automating or augmenting knowledge work, showing instead that AI can reconfigure even menial roles in ways that make them central to organizational knowledge production. Chapter 2 investigates how data scientists participating in a hackathon coped with limitations in the available data while developing machine learning models to predict complex natural phenomena. The study identifies four core practices—transforming data, sourcing data, redefining phenomena, and proxy making. These practices reveal the iterative nature of data work and highlight how developers not only work with data but also reshape how the world is conceptualized in the process. The chapter refutes the belief that AI learns autonomously from data by demonstrating the interpretive and creative labor involved in model development. Chapter 3 is an ethnographic study of an AI development project focused on automating the phenotyping of cucumber traits for plant breeding. Initially, developers attempted to automate the knowledge of plant breeders. However, as they encountered the breeders' material culture—such as handling actual cucumbers and participating in on-site discussions—they shifted their approach. They moved from a modeling paradigm to one more aligned with ethnographic knowledge production. This transition illustrates how AI development is shaped by the embodied, situated knowledge of domain experts and challenges the assumption that AI is purely a digital phenomenon. Chapter 4 examines how AI reshaped the role of seed sorters in a seed processing facility. Traditionally seen as low-skilled laborers, seed sorters became key contributors to organizational knowledge once AI systems for seed evaluation were introduced. Through detailed ethnographic observations, the chapter shows how changes in the "apparatus of measurement"—including imaging devices and AI models—enabled seed sorters to gain new insights and participate in collaborative research. This chapter challenges the assumption that AI merely automates or augments expert work by demonstrating its potential to transform so-called menial roles into knowledge-producing ones. To reframe these assumptions, I draw on the theoretical lenses of material culture and performativity. Rather than seeing AI as an autonomous, digital tool for automating expertise, I propose understanding AI as a sociomaterial system shaped by the practices, artifacts, and bodies involved in its development and use. This perspective foregrounds human learning, the role of physical and sensory practices, and the co-constitution of technologies and organizational roles. It enables us to see AI not as a fixed product, but as an evolving apparatus that reshapes knowledge, labor, and meaning in context-specific ways.