Making Archives AI Ready: A Brief Overview of ARA Guidelines

This article is part of our AI and Archives series - exploring the applications of AI in modern archiving.

Artificial Intelligence (AI) is becoming increasingly integrated into archival software, with archivists exploring how it can make archival work more efficient. However, the rapid development of AI has created uncertainty about how organisations should prepare their archives for its growing use, highlighting the need for clearer guidance on its responsible implementation. In response to these concerns, the Archives and Records Association (ARA) commissioned the FLAME (AI for Libraries, Archives and Museums) project, led by Professor Colavizza and Professor Jaillant.

Earlier this year, the ‘AI Preparedness Guidelines for Archivists’, co-authored by Colavizza and Jaillant, was published, providing practical advice on preparing archives for AI. This article outlines the key recommendations from the ARA guidelines.

 AI Applications in Archives

The guidelines distinguish between two types of AI used in archives. Task-specific AI, which performs defined functions such as locating names, dates or sensitive information, and generative AI, which can draft archival descriptions and summarise records. However, the authors argue that AI can only be useful when collections are made ‘AI ready’. They identify four key areas of preparation, including completeness; metadata and access; data types and formats; and application-specific metrics.

 Four Areas of Preparation for AI Use

 1.     Completeness

Completeness means documenting whether a collection is fully or partially digitised and explaining any gaps within the collection. This context helps the AI understand the scope and limitations of the collection, allowing the AI to give more accurate archival descriptions. The guidelines also stress the importance of documenting exclusions and potential biases, including under-represented communities or historical periods, so that AI-generated outputs do not present an incomplete collection as a comprehensive historical record.

 2.     Metadata and Access

Metadata and access focuses on maintaining consistent metadata practices. The guidelines emphasise the importance of consistent metadata, noting that even basic item-level descriptions, including dates, creators and titles, can improve AI-assisted retrieval and description. Information about provenance, relationships between records and access conditions is also essential, alongside collection-level descriptions and contextual information.

3.     Data Types and Formats

Data types and formats refer to preserving the provenance of digital records by maintaining their original file names, complex folder structures and file formats, while also creating a separate but linked derivative copy for AI use. However, the guidelines also recognise that different types of records, such as text documents, photographs and audiovisual material, may require different formats for AI to process them effectively. It is therefore recommend to standardise the derivative copy into AI-friendly formats, for example by converting text documents into UTF-8 plain text. This will allow AI tools to analyse the collection more efficiently, while also protecting the original records from accidental alterations.

 4.     Application-specific Metrics

Finally, application-specific metrics means establishing clear use cases for AI and measurable criteria to monitor its effectiveness. These criteria should reflect the AI’s intended objectives. For example, generative AI’s effectiveness at drafting descriptions can be measured by the staff time saved per record when compared to writing descriptions manually. Ultimately, these measures are important as they help archivists determine where AI is delivering practical benefits and where it shouldn’t be used.

Key Takeaways for Archivists

Overall, the guideline highlights that successful AI adoption depends on careful preparation, appropriate safeguards, and human oversight. This means that AI cannot operate effectively without the archivists' involvement, as its reliability is dependent on the quality of the preparation carried out before its deployment and the ongoing monitoring by archivists once it is in use.

 You can access the full ARA guidelines here: https://www.archives.org.uk/ai-preparedness-guidelines-for-archivists