For those who are not familiar with it, Europeana is a European initiative working to make digital cultural heritage more accessible, usable and open to everyone.
Through Europeana, millions of artworks, images, books, documents, recordings and digitised objects from museums, libraries, archives and cultural institutions across Europe can be discovered and reused.
But Europeana is not just an online platform. It is also a community of professionals, institutions, aggregators and networks that have been working together for years on the digital transformation of cultural heritage.
Today, this community is also at the heart of the common European data space for cultural heritage: an infrastructure designed not simply to collect data, but to make it understandable, trustworthy and useful.
It is from this community that the paper The case for Public AI: making it happen with cultural heritage has emerged.
The paper brings together the work of the Alignment Assembly on Culture for AI, a collective discussion that has involved more than 400 cultural heritage professionals since May 2025.
It does not claim to have all the answers. That would be impossible, especially at a time when the questions seem to change almost as quickly as the technologies themselves.
But it does try to build a shared position. And, above all, it makes one thing very clear: the cultural heritage sector cannot simply stand by while artificial intelligence changes the way we access knowledge.
Artificial intelligence is not new to museums, archives and libraries. For many years, it has been used to recognise images and handwritten texts, transcribe recordings, enrich metadata, classify collections and make them easier to explore.
What has changed is the scale.
With the rise of large language models and generative AI, these systems no longer simply help us search for information. Increasingly, they are becoming the place where we search for it. They select, summarise and reorganise what we know.
In other words, they are becoming one of the main interfaces through which we encounter culture, history and the world.
And this is not just a technical matter.
It means deciding which content becomes visible and which remains at the margins. Which languages are represented. Which sources are considered trustworthy. Which version of a story is presented to us.
Ultimately, it means influencing how meaning is produced and how power is distributed.
Today, much of this power is concentrated in the hands of a small number of large private companies. These companies use enormous quantities of online content, including material digitised and made available by cultural institutions through public investment.
But we often do not know exactly how this data is being used, under what conditions, or how communities and the institutions that have preserved and shared it will benefit.
Sharing is not the problem. Cultural heritage is preserved and made accessible precisely so that it can circulate, be studied, interpreted and create new value.
The problem begins when this relationship becomes purely extractive.
Institutions bear the costs of digitisation, preservation, cataloguing and digital infrastructure. Their data is then collected at scale to build commercial products, often without attribution or transparency, and without part of the value created returning to the public ecosystem from which that data came.
The paper proposes addressing this imbalance through a vision of Public AI.
“Public” does not necessarily mean developed and controlled entirely by the state. It means an ecosystem in which technologies have public attributes, serve public functions and are subject to democratic forms of control.
An AI that is accessible, open and interoperable, but also transparent and accountable. One that responds to the needs of society, not only to the logic of the market.
This is an important distinction.
Public AI is not the same as “sovereign AI”, which tends to place national or European competitiveness at the centre. Nor is it enough for a model simply to be open source.
The question concerns the whole system: data, computing power, models, applications and the rules through which they are governed.
This is where cultural heritage can play a much more important role than we might imagine.
The paper identifies four concrete areas of action.
The first is data quality. Museums, archives and libraries hold collections that are curated, documented and often multilingual. Their contribution is not limited to providing images or texts on which a model can be trained.
They can provide context, provenance, connections and information that help us understand where content comes from and how it should be interpreted.
It might sound like a detail, but it is not. Data without context can easily become incorrect information. Or a story only half told.
The second area concerns the conditions under which data can be accessed and reused. Cultural institutions can act as responsible data stewards, helping to establish fairer relationships with those developing AI systems.
Reciprocity can take many forms: attributing sources, being transparent about how content is used, contributing to infrastructure costs, or sharing tools, expertise and computing capacity.
Being open should not mean giving up every form of control. And protecting something does not necessarily mean closing it off.
The third area is the development of smaller, specialised systems.
We do not always need enormous, general-purpose models that consume vast amounts of energy and resources. In many cases, models developed for a specific purpose and built using trusted data may be more useful, easier to understand and more environmentally sustainable.
This would also allow cultural institutions to be more than sources of training data. They could take part in designing and governing the technologies they will eventually be expected to use.
The fourth area is perhaps the one most closely connected to the public mission of culture: helping people understand artificial intelligence.
AI literacy is not only about technical skills. It means understanding how these systems work, recognising their limitations and asking questions about the data they use, the biases they may reproduce and the responsibilities involved.
Cultural institutions have a long history of making complex subjects accessible. They can do this through exhibitions, workshops, educational programmes, public events and new forms of participation.
They can become places where artificial intelligence is not simply celebrated or feared, but discussed.
The paper also highlights another fundamental issue: diversity.
If AI systems are trained mainly on English-language content, they will inevitably present an incomplete view of the world. Europe’s cultural collections, with their linguistic, geographical and cultural richness, can help reduce this imbalance.
Of course, cultural data is not neutral or perfect either.
Collections reflect decisions, absences and power relations built over time. Some communities have been described by others. Some voices have been excluded. Many descriptions still carry the language and prejudices of the past.
But cultural institutions, working together with the communities represented in their collections, have the knowledge and experience needed to recognise these issues and address them openly.
This is why the paper is not a romantic defence of cultural institutions. It is an invitation for them to take on a new responsibility.
The common European data space for cultural heritage can become a trust infrastructure, bringing together institutions, communities, researchers and technology developers.
It can help create well-documented and legally reusable datasets, define access conditions, support training and build more balanced relationships with Europe’s AI infrastructure.
But to do this, the cultural heritage sector must speak with a clearer voice. And a more political one.
In the end, this is not about choosing between innovation and regulation, or between excitement and fear.
It is about deciding what kind of innovation we want to support, who it should be accountable to, and what kind of society we want to build through it.
Culture cannot shape the future of artificial intelligence alone. But it can help make that future more pluralistic, more transparent and closer to people.
This paper is a starting point. And perhaps that is its greatest value: reminding us that the future of AI has not yet been written, and that cultural institutions should not simply adapt to technologies built by others.
They can help change their direction.