OVERVIEW: This conference session will provide my professional peers with an inside look at the process of designing an AI tool for use in the archives. While archival institutions are often seen as traditional, our team is actively exploring how to leverage cutting-edge AI technology to streamline the declassification review of historical records. I aim to explore both the potential benefits of AI in this context and the challenges and realities of implementing such tools effectively.
The World Bank’s Access to Information (ATI) Policy serves as the legal framework for reviewing documents to determine whether they should be restricted or released based on specific exceptions. While it might seem ideal to simply program an AI tool to automatically apply the ATI Policy, the reality is far more nuanced and complex.
Our team’s experience with reviewing records digitally began during the peak of the COVID-19 pandemic. Rather than assessing physical records, we digitized them into PDFs and annotated these files to highlight exceptions and flag pages for restriction. This process quickly revealed the need for a consistent naming convention to track document versions and streamline collaboration among team members. Once we returned to the office, we resumed reviewing the original physical documents, sending only the approved records for digitization. While the experience sparked ideas for alternative methods of reviewing restricted records, our immediate focus remained on addressing the backlog of public requests. This highlights the challenges of introducing new workflows in an organization while managing urgent deadlines for fulfilling public information requests. Last fall, our Information and Technology Solutions Vice Presidency launched an “AI Shark Tank” competition, inviting teams to pitch ideas for tackling real-world challenges with AI. My colleagues and I from the Archive’s Access to Information team seized the opportunity and participated with our proposal titled AI-Powered Access: Revolutionizing ATI Declassification for Transparency. During our pitch, we displayed bulky folders of onion-skin paper—an important visual for the competition judges unfamiliar with archival records. More than any polished PowerPoint slides, this hands-on demonstration of physical documents helped judges understand the painstaking, manual process we rely on for line-by-line reviews. We also emphasized to our judges of experienced ITS leaders the feasibility and scalability of deploying an innovative AI tool capable of handling the complex and sensitive content of our textual collection, adapting to new types of content, identifying patterns in previous declassification decisions, and ensuring consistency in restricted materials.
Winning the AI Shark Tank competition was just the beginning—the real challenge lay in transforming our vision into a practical, effective tool. The competition was followed by a 24-hour hackathon, which proved invaluable in reimagining workflows and setting the stage for development.
The hackathon, combined with ongoing meetings and operations, allowed us to build momentum and further advance our vision. The connections we forged with software engineers, developers, and system architects during this event have been crucial to advancing our AI project. One of the lessons I hope to share at the conference is the importance of nurturing partnerships and securing leadership buy-in for long-term initiatives. Complex projects like this require sustained effort, time, and often a temporary shift in focus.
Patience, open-mindedness, and collaboration are essential when embarking on complex initiatives. Working with the engineering team to translate archival practices into technical requirements has been a truly enriching experience. Initially, I believed that simply articulating our vision would be sufficient to bring it to life. However, this ongoing partnership has shown me the value of breaking down business requirements into incremental milestones.
Rather than a single product, our AI solution is evolving as a suite of interconnected applications designed to analyze records, support decision-making, and improve through multiple iterations. This iterative approach has been essential to ensuring the tool meets both immediate needs and long-term goals.
This session will provide participants with a better understanding of the following key points:
- AI Integration in Archival Workflows: Insights on incrementally redesigning archival processes to streamline declassification, analyze records, and improve decision-making in archival settings.
- Collaboration and Leadership: Practical tips on managing competing priorities, working with IT teams, securing leadership buy-in, and fostering cross-disciplinary collaboration for long-term success.
- Iterative AI Development: Lessons on building scalable AI tools through incremental milestones and continuous improvement.
- Real-World Application: A case study on how to transform an AI idea into a practical tool, from concept through collaboration to implementation.
Participants will leave with actionable strategies for applying AI in archival work and navigating the complexities of such initiatives
GARA CERTIFICATE CORE COMPETENCY AREA: "Record Considerations for Emerging Technologies"
TARGET AUDIENCE: Federal, Tribal, State, Local, Public Institutions of Higher Learning
FOCUS AREAS: Archives, Records Management, Technology/Tools
PRESENTER(S): Salma Berrada El Azizi, IT Analyst / Archivist, The World Bank