Library Innovation Lab @ Harvard Law School · Public Data Project

AI Benchmarking Exercise

Federal datasets can be altered, delayed, defunded, or quietly taken offline. Assessing their level of risk is complicated work that requires both knowledge of the dataset as well as political, technological, and policy perspectives. This exercise—which can be done alone or in a group—measures how humans, humans assisted by AI, and AI agents alone perform as they each assess a group of datasets using a common rubric.