National Archives adopts Google Gemini AI for employees

National Archives embraces AI for efficiency and accessibility: The U.S. National Archives and Records Administration (NARA) is actively exploring and implementing artificial intelligence technologies to enhance its operations and public services.

AI initiatives and employee concerns: NARA has introduced several AI-related projects, sparking both excitement and apprehension among its staff.

• In June, NARA presented “AI-mazing Tech-venture,” showcasing Google’s Gemini AI as a productivity tool for employees.
• The agency plans to launch “Archie AI,” a public-facing chatbot, in December.
• Some employees have expressed concerns about AI’s role in archiving, citing potential risks to historical accuracy and institutional trust.
• NARA previously banned the use of ChatGPT due to data security concerns but now favors Google Gemini and Microsoft Copilot for their “more controlled environment.”

Demonstration and employee feedback: The presentation included a live demonstration of Google’s Vertex AI, trained on National Archives data.

• The AI was prompted to act as an “expert archivist” and answer questions about the JFK assassination.
• Employees raised issues about the AI’s self-designation as an expert and the potential for misinformation.
• Questions were posed about data privacy, copyright infringement, and the environmental impact of AI technologies.

Archie AI and public disclosure: NARA is developing Archie AI as a tool to improve public access to its vast collection of records.

• The chatbot will come with a disclosure stating that AI-generated summaries may not reflect NARA’s opinions and are not guaranteed to be accurate.
• NARA aims to connect users with records more seamlessly, regardless of their purpose (e.g., veterans, family historians, researchers).

AI applications in archiving: The agency is exploring various ways to leverage AI in its archival processes.

• AI is being used to transcribe Revolutionary War pension files, with a reported 90% accuracy rate.
• NARA plans to share AI-generated transcripts in its official catalog by the end of the year.
• The agency is developing a prototype AI research assistant powered by Google Vertex.

Broader context and implications: NARA’s AI adoption aligns with broader government initiatives but raises important questions about the role of technology in preserving and presenting historical records.

• The Biden administration has directed federal agencies to study AI and create usage policies.
• NARA’s AI initiatives could significantly speed up the digitization and accessibility of historical documents.
• The use of AI in archiving raises complex issues regarding accuracy, trust, and the potential transformation of how we interact with historical records.

Critical considerations: As NARA moves forward with its AI plans, several key questions and challenges remain.

• How will the agency balance the efficiency gains of AI with the need for human expertise and oversight in archival work?
• What safeguards will be put in place to prevent the spread of misinformation or misinterpretation of historical documents?
• How will NARA address the ethical implications of using AI to interpret and present sensitive historical information?