From Reflection to Repair: A Scoping Review of Dataset Documentation Tools
Analysis of 59 publications reveals why documentation tools fail to gain adoption in AI development.
Deep Dive
Researchers from MIT's Robotics and AI Institute, ESPOL, and IBM published "From Reflection to Repair," a systematic review of 59 dataset documentation publications. Their mixed-methods analysis identified four persistent patterns hindering adoption: unclear value propositions, decontextualized designs, unaddressed labor demands, and deferred integration. The study proposes shifting Responsible AI tool design toward institutional solutions and outlines actions for the HCI community to enable sustainable documentation practices.
Why It Matters
Poor dataset documentation leads to biased AI models; fixing these tools is critical for responsible AI development.