The US State Department will use an AI chatbot called StateChat to help select members of its Foreign Service Selection Boards.

Developed with Palantir and Microsoft, StateChat will generate candidate lists based on skills and grades, while final evaluations remain human-led.

The process must still follow the 1980 Foreign Service Act, but it’s unclear how the AI will ensure diversity and representation.

StateChat has been used internally since 2024, but this marks its first disclosed role in human resources.

The Foreign Service Association has asked for clarification, and neither Palantir nor Microsoft has commented.

🔗 Source: Reuters

🧠 Food for thought

1️⃣ Federal agencies are rapidly expanding AI use in human resources functions

The State Department’s use of AI for Selection Board member assignment reflects a government-wide trend toward automating administrative personnel functions.

A 2019 study identified 130,000 federal jobs across 80 occupations likely to be impacted by AI adoption, with administrative tasks being prime candidates for automation 1.

This aligns with broader government initiatives, as federal agencies requested $1.9 billion for AI research and development for fiscal year 2024 2.

The State Department’s approach mirrors similar implementations at other agencies. For instance, the Department of Homeland Security launched an AI chatbot for its 19,000 employees 3, while the GSA recently developed an internal AI tool to assist with employee tasks 4.

These developments represent a significant shift in how government manages its workforce, with AI increasingly handling functions previously requiring manual review of qualifications and credentials.

2️⃣ AI in personnel selection raises complex bias and representation concerns

The State Department’s emphasis on “unbiased selection” through AI presents a challenging technical and policy problem, particularly given the Foreign Service Act’s explicit requirement for diverse representation.

Research shows AI systems can perpetuate existing biases in hiring contexts. For example, the Word2vec model demonstrated gender bias reflecting stereotypes in its training data 5, highlighting the potential for automated systems to maintain rather than eliminate historical patterns.