The gateway to European audiovisual heritage

Search in:

Building and keeping trust: Generative AI and archival materials

“How do we make sense of the responsible use of Generative AI when engaging with audiovisual materials?” This question was at the centre of the EUscreen webinar Generative AI: Confusion of facts or understanding of content?. On Wednesday 21 May, three esteemed experts in the field of Generative AI in archives and newsrooms provided critical reflections and practical guidelines on how to use AI within our sector. 

Watch the recording

Fortify the truth

The webinar was kicked off by Jacobo Castellanos, Coordinator of the Technology, Threats and Opportunities team at WITNESS. Jacobo opened by showing the evolution of synthetic media’s impact on human rights over the past eight years, referencing WITNESS’s early initiative, Prepare, Don’t Panic, which aimed to proactively address deepfakes. Through global consultations with various stakeholders – like journalists and activists – WITNESS has since shifted focus towards action, emphasising the need to “fortify the truth” by adding indicators of authenticity to content from vulnerable communities. 

They are not alone in this: through partnerships with organisations like C2PA (The Coalition for Content Provenance and Authenticity), standards are being developed for certifying the source and history of media content via cryptographic metadata, allowing for provenance without compromising privacy. Jacobo stressed the ethical dilemmas posed by provenance technologies, such as potential surveillance, and underlined the need for epistemic justice for content lacking technical verification.

History distorted

Being the managing partner at Topfoto –  an archive with millions of negatives, prints, and illustrations dating from 1900 onwards – Flora Smith discussed how photographic archives have a leading role in educating the public about how synthetic images can distort history. Around 34 million synthetic images are generated every single day, which are increasingly assumed to be real. Flora highlighted a public education campaign called Reality Checked and a documentary, The Moment of Truth, to raise awareness about AI’s impact on visual history. 

Flora discussed industry-wide efforts to redefine “authenticity” as “trustworthiness” and shared Topfoto’s proactive steps: ensuring no synthetic images enter their archive and openly communicating their ethics and provenance policy to clients. Flora described plans for blockchain-based archive anchoring and praised initiatives like Starling Lab and OpenOrigins that support provenance diversity outside of major tech monopolies. 

AI for journalism: A double-edged sword

Xavier Eutrope offered an institutional perspective from INA, France’s national audiovisual archive, focusing on AI’s systemic impact on journalism. Xavier described INA’s role in mapping AI’s journalistic applications, from content collection and verification to distribution, and emphasised that AI is not inherently good or bad but a pharmakon – a word used by the ancient Greek to describe something that is both remedy and poison. 

Drawing on recent events, Xavier illustrated how AI tools could either compromise anonymity or protect it, depending on use and context. INA’s AI tool evaluation criteria was introduced – such as use value, technological maturity, and legal risk – and an ethical framework to guide responsible implementation. Xavier concluded that journalists face unique challenges: AI threatens jobs, credibility, and independence, and must therefore be wielded with a clear-eyed understanding of its double-edged nature.

AI generated content to complement authentic archival material

After hearing from our three experts and their view on responsible use of Generative AI, we moved towards a panel discussion and Q&A with the audience, moderated by Johan Oomen, Head of Research & Heritage at the Netherlands Institute of Sound and Vision and chair of EUscreen. 

The Q&A centered on nuanced and practical concerns from participants working with cultural heritage. Xenophon, a computer scientist working on digital cultural preservation, asked about good practices on how to use metadata and digital assets for educational reconstructions of historical sites, while maintaining historical accuracy and trust. In other words: if we use AI to generate historical decors to complement real authentic heritage materials, how can we make sure the generated content comes across as accurate and plausible yet is not mistaken as to be real? 

The experts emphasised that sometimes, real footage is simply not available while still needed for educational purposes. Nowadays, synthetic content can be so realistic that it is often falsely perceived as real. Therefore, it is important to label generated content correctly and signal to your audience that it is generated and set the context under which it was created. 

In addition, the experts agreed that historical accuracy has to be established authoritatively. Certain institutions already have this credibility because of the research that they’ve done, because of their methodologies and because of their scientific approach. If you want to establish historical accuracy in synthetic content, you leverage that credibility from institutions that have this authoritative position.

Practical takeaways

Participants left with a clear message: 

  • Trust is essential: When you’re using AI in your archival practice, building and keeping trust is key. This means being open about how images and information are created or altered.
  • Show where things come from: Use tools like metadata and blockchain to clearly show the history and origin of digital content. This helps people understand what is real and what isn’t.
  • Small organisations can lead: You don’t need to be a big tech company to make a difference. Small archives like Topfoto are showing how to take ethical and practical steps to protect truth in the digital age.
  • Think before using AI: Always ask why you’re using AI and whether it helps or hurts. Make sure it supports your values and doesn’t create new problems, like privacy risks.
  • Work together: Journalists, archivists, technologists, and educators need to collaborate. By sharing tools, ideas, and standards, we can make sure AI is used responsibly and usefully.

Other useful links

Title
Date
Provider
Description