Research Interests: Performance validation/error assessment practices for automated and semi-automated inference systems where ‘gold standard’ truth data is difficult or impossible to obtain, the ways in which these practices have (and have not) evolved over time, how they privilege certain metrics over others (e.g. quantitative over qualitative, representative cases over edge cases), and how they reflect the valuation of different types of labor (e.g. coding vs. manual review) within the development community. Read more about Sandip Bhatt