When AI Meets the Archive: How IWM Is Reimagining Access to Sound and Image Collections

What happens when you have more than 20 million documents, 11 million photographs, and countless hours of recorded voices—but only limited ways for the public to find meaning in them?

For Imperial War Museums (IWM), that challenge isn’t theoretical. “At this scale, interpretation and accessibility is a massive obstacle and a huge opportunity,” explained Emily Dodd, Head of Collections Management at IWM. The museum, a family of five sites across the UK, has always experimented with new technology to connect audiences with personal stories of conflict. Now, its teams are testing what artificial intelligence can bring to that mission.

Making Voices Heard

The first project focuses on IWM’s sound archive: 20,000 hours of oral history interviews stretching from the First World War to the present day. “This collection really helps you to connect with the past, to listen to the voices of the people themselves who lived through war and conflict,” said Anthony Richards, Head of Documents and Sound.

But there was a bottleneck: transcripts. Richards and his team once calculated it would take 22 years to produce them manually. “Of course, that’s completely unrealistic,” he admitted. AI offered another path.

Edward Kay, Head of Platforms and Development, explained how the team used stored interview notes to guide the AI, dramatically reducing errors: “Just from spot-check analysis, we were getting something like 99% accuracy for the content.” Within weeks, the archive had transcripts, translations into major European languages, and even automatically generated interview summaries and timelines.

The real magic, however, lies in the new ways audiences will be able to interact with the material. “You can now ask a question in natural language— and the system will surface relevant clips, complete with citations,” said Richards. One result? A veteran’s admittance of fear during their flying training during the First World War. What once felt like a buried treasure hunt is now becoming searchable, discoverable, and deeply human.

Teaching Images to Speak

IWM’s second AI experiment takes on photographs—millions of them, many with scant metadata. “This project involves using AI to help catalogue these photographs,” explained Nathan Doherty, Curator (Photographs) Cold War and Late Twentieth Century Conflict. The system generates titles and descriptions, which are then validated by volunteers through a dedicated portal.

For Callum Brogan, Digital Producer (AI and Crowdsourcing), the human layer is essential. “Our approach remains human-focused whilst also taking advantage of the capabilities offered by a tool like AI. Remote volunteers will gain new digital skills, while IWM benefits from fresh insights that enhance the quality and diversity of our data.”

The model isn’t static, either. Kay described how the feedback loop works: “We’ll be able to compare the output of the AI with the output of the volunteers and then create a virtuous cycle—feeding that difference back into the model to improve future results.”

Balancing Innovation and Integrity

For all the excitement, the team is quick to note the risks. “We can’t put the integrity and authenticity of this material at risk or enable bias and hallucination,” Dodd stressed. The museum has taken care to use only content where rights are cleared, keep data secure, and separate AI-generated information from staff-authored records in its CMS.

Challenges remain, especially when scaling up. Richards wants to expand AI-powered search across thousands of interviews. Doherty worries about subtle errors in photo descriptions slipping past reviewers. And Dodd underscored a central tension: “If we lose that trust in collections, we don’t really have anything else. That is our purpose.”

People, Not Tech, at the Centre

For all the talk of algorithms and pipelines, the team returns to a consistent theme: AI is only as powerful as the people around it. “This has all been led by people, not tech,” said Dodd. “We brought in colleagues from across the museum, as well as tech partners, to debate every step. That gave us a whole new way of working.”

Richards sees a poetic irony: “We’re using artificial intelligence to make a very human collection a lot more accessible. There’s something quite nice about that.”

What the IWM team has built is not simply a suite of AI experiments—it’s a blueprint for how museums might responsibly harness technology, while keeping trust, authenticity, and human connection at the forefront.

–

Callum Brogan, Emily Dodd, Nathan Doherty, Anthony Richards and Edward Kay fromImperial War Museums spoke at the Digital Collections Summit 2025.