Program for language translation capabilities

Section 1(n) of the State Department Basic Authorities Act of 1956 (22 U.S.C. 2651a(n)) is amended— (1) by redesignating paragraph (5) as paragraph (6); and (2) by inserting after paragraph (4) the following: (5) Program for language translation capabilities (A) Establishment The Chief Artificial Intelligence Officer shall establish a program for the utilization of multi-modal generative artificial intelligence language translation capabilities within the Department. (B) Matters to be included The program required by subparagraph (A) shall— (i) include an automated human-in-the-loop review and verification process option and a machine-only process option, allowing users to choose which process to utilize in adherence with Department policies; (ii) include real-time training and fine-tuning of translation models for use within different geographic regions and Department mission areas; (iii) be available on unclassified and classified information technology networks; (iv) be capable of generating original content in non-English languages; and (v) be available at all United States and overseas missions of the Department. (C) Department policies The Chief Artificial Intelligence Officer shall ensure the deployment and use of artificial intelligence for language translation capabilities as part of this program adhere to Department policies, including the conditions where it is appropriate to use machine-only processes or automated human-in-the-loop review and verification processes. (D) Definitions In this paragraph— (i) the term automated human-in-the-loop review and verification process has the meaning of an automated process within an artificial intelligence language translation system that requires human linguists to review and verify translations performed by an artificial intelligence model for accuracy prior to returning translated materials to a user; (ii) the term “machine-only process” has the meaning of an artificial intelligence language translation capability that delivers a translation to a user without review by a human linguist; and (iii) the term “multi-modal” has the meaning of multiple modes or formats of content to be translated, including audio, text, video, and text contained within images..