‘Drop your ML tools’: the role of domain experts’ material culture in ML development

Scholars studying machine learning (ML) development emphasize the importance of collaboration between domain experts and developers, claiming domain expertise is critical for aligning ML systems with real-world problems. Moreover, they argue this collaboration requires recognizing that domain expertise is not merely a cognitive skill to be elicited but practices deeply embedded in the context of experts’ work. Despite this realization, the role of the material culture of domain experts—the artifacts and tools central to their practices—in ML development remains underexplored. Drawing on a 25-month ethnographic study at BioTech, a global plant breeding company, we show how developers’ engagement with the material culture of domain experts changed their approach to developing ML. Our findings reveal that immersion with the material culture of domain experts first prompted developers to critically reflect on their own assumptions, and have subsequently informed their new methods, evaluation criteria, and modeling practices. This study contributes to the literature by explaining how the material culture matters for ML development.