SESSION

Automating Government Transparency in Declassification and FOIA Tasks

OVERVIEW

EXPERIENCEIn Person
TYPEBreakout
TRACKData Science and Machine Learning
INDUSTRYPublic Sector
TECHNOLOGIESAI/Machine Learning, ETL, Governance
SKILL LEVELIntermediate
DURATION40 min

In the conduct of normal government business, communications may be classified to protect national interests for a period of 25 years in most cases. When that period elapses, official records are declassified unless a review indicates they should be exempt. Unclassified documents are also subject to public disclosure through Freedom of Information Act requests. The existing manual review process is quickly becoming untenable. To get ahead of this challenge and ensure timely document release, we trained an open source supervised classification model to recognize human declassification decisions from 2020-2021. We tested our model on cables that underwent a declassification review in 2022. Our model performed to an accuracy of over 98%, resulting in over 63% of the manual workload preserved. Successful integration of the technique, the platform, and the business process keeping humans in the loop demonstrated increased performance in 2023 while improving the consistency of the reviews.

SESSION SPEAKERS

Samuel Stehle

/Data Scientist
US Department of State