Museums and AI: Balancing Innovation and Integrity

On 7 October 2024, the European Heritage Hub gathered 400 attendees both in person and online for the second Hub Forum in Bucharest. At the invitation of Europeana, Paul Keller delivered a short pre-recorded video provocation on Museums & AI: balancing Innovation and integrity. Read the text of Paul’s speech below or view the recording of the presentation here:

The emergence of the current generation of AI systems has important implications for the role of cultural heritage institutions. While other actors view new technologies through the lens of innovation alone, cultural heritage institutions must balance the assessment of risks and opportunities based on concerns about the integrity and authenticity of their collections.

Some promising applications of AI technology in our field include helping with issues such as multilingualism, improving the availability of metadata or understanding the connections between works, collections, and context.

However, all of them function by making things up based on processes that none of us can examine or understand in detail. As a result, allowing AI to interact with the collections of cultural heritage institutions poses a great risk to the integrity of these collections, which have been curated and preserved as authentic sources of knowledge.

It is important to build firewalls around existing collections to ensure that they are not contaminated by AI outputs before it is too late. In the long run, every machine translation, every bit of machine-generated metadata, and every upscaled image added to a collection will undermine its value as an authoritative source of human knowledge.

The nature of collections as authoritative sources of human knowledge and culture presents us with a great opportunity. Today, there is an almost insatiable demand for data, including the data contained in cultural heritage collections. This data is considered especially valuable because it comes from well-curated collections of authentic human knowledge and culture.

Museums should rethink their relationship with those who want to use the data embedded in their collections by asking:

• How can we use this demand for data to strengthen our institutions?
• How can institutions strengthen their core mission of preserving collections and making them available for human inquiry?
• How can we best serve the demand coming from machines to empower people and society at large?

The data in heritage collections is valuable to a wide range of users, from researchers to open-source AI developers or the world’s largest corporations. The latter have the means to help digitize more of our collections while gaining access to the data they contain.

This combination of demand for data and willingness to expend resources gives cultural heritage institutions agency to decide who gets access to the data in their collections and under what conditions.

As large AI companies approach cultural heritage institutions and other sources, it is important to keep the following considerations in mind:

• First, we must ensure that the digitization of collections results in public access to the data for everyone on the same terms. No more digitization projects that result in exclusive access that further entrenches the position of already dominant players.
• Second, the current situation provides an opportunity to digitize collections that are still in copyright but no longer in commercial circulation. Here cultural heritage institutions will need to work with rights holders to provide access to as complete collections as possible on terms that benefit both rights holders and cultural heritage institutions.
• Third, when providing access, we must work within the boundaries of copyright law. Cultural heritage institutions should not build additional walls around works that are in the public domain. They belong to everyone, and restricting access to them will only hurt those trying to catch up with the big AI companies.
• Fourth, it follows that charging for access should be limited to works that are still in copyright. Here we need to find ways to share the revenue in a way that fairly rewards rights holders while also helping to preserve our collections.

To seize this opportunity, we need to think differently about scale. Faced with a small number of giant companies, we need to scale up, bundle collections across borders, partner with publishers and other rights holders, and together make them offers they cannot refuse.

As demand for data outstrips supply, we should be able to set the rules of access. We should strive to ensure that the hunger of the machines is harnessed to grow the digital commons.

We should do this with the goal of making more of the knowledge and culture in our collections available to all of humanity. Both directly, through our institutions and platforms, where integrity and authenticity should continue to be our primary concern. But also indirectly, through the AI systems that others will build on top of this data and the innovative use cases that will emerge.