Machine Learning practices and infrastructures

Tools shape ethical blind spots: The paper finds that interactive computing platforms like Jupyter Notebook and Google Colab are making certain critical stages of the ML lifecycle "invisible" to practitioners—stages that AI ethics research has shown are particularly important for preventing discrimination and inequity in deployed systems.

Infrastructure influences practice: ML practitioners' interactions with their tools don't just support their work; they actively shape how ML systems get built. The paper demonstrates an "infrastructural relationship" where the platforms practitioners use co-evolve with their practices, meaning tool design choices have downstream consequences for responsible AI development.

Learning happens through tools: The study of Stack Exchange questions reveals that interactive computing platforms serve multiple purposes beyond coding—they're central to how ML practitioners learn, coordinate with teams, and solve problems. This means these tools have outsized influence on what practices become normalized in the field.

Overlooked source of bias: While much AI ethics research focuses on data, algorithms, and organizational policies, this paper highlights that the tools practitioners use daily are an often-overlooked site where problematic practices can become embedded and invisible, potentially exacerbating existing inequities.