Who Should Teach Artificial Intelligence About Culture?
Artificial intelligence can identify an object, translate a label, reconstruct a damaged site or help a researcher search thousands of records. But knowing what an image contains is not the same as understanding what it means.
That distinction is placing museums, libraries, archives and originating communities at the center of an important policy debate: if cultural collections help train or inform artificial-intelligence systems, who determines how those collections are described, licensed and interpreted?
The first global UNESCO–ICOM survey of AI use in museums, presented on September 17, 2026, gathered responses from more than 400 museums in 90 countries. Fifty-seven percent of respondents reported already using AI, yet 55 percent had no internal policy, strategy or guidelines governing its use. Only 17 percent provided AI training to staff.
The imbalance is consequential. Museums are experimenting with AI for administration, communications, research, documentation, exhibition development and visitor engagement. Yet many institutions are adopting these systems before deciding what responsible use requires.
Accuracy is one concern. A model may connect an artwork to an incorrect artist, period or community. It may repeat outdated terminology found in older catalogues or reproduce gaps created by centuries of unequal collecting. If women, minority communities or non-Western traditions were poorly documented in the historical record, an AI system trained on that record may present those absences as objective reality.
Provenance presents a deeper problem. A digitized object is not merely an image. It has a maker, ownership history, material composition, original purpose and relationship to a particular place or community. Removing that information may make the file easier to process, but it makes the culture less intelligible.
Copyright and licensing remain unsettled as well. In Europe, providers of general-purpose AI models must maintain policies intended to comply with copyright law and recognize qualifying reservations of rights connected with text-and-data mining. Separate EU transparency rules, applicable since August 2, 2026, address the marking and disclosure of artificially generated or manipulated material.
These rules provide a legal framework, but law alone cannot supply cultural judgment. An institution may possess the copyright to an image without holding the moral authority to separate it from the community or ceremony that gives it meaning. Indigenous knowledge, sacred material and living traditions require consultation that cannot be reduced to a licensing checkbox.
The European Heritage Hub is consequently arguing that heritage organizations must be treated as “trusted agents of cultural knowledge,” not simply as suppliers of training data. Its 2026 campaign calls for cultural institutions to participate in designing Europe’s public-interest AI ecosystem and informing the European Commission’s forthcoming strategy for cultural and creative sectors.
A responsible museum policy should therefore answer several basic questions before a collection is used with AI: Who created the underlying material? Who owns the rights? Is community consent required? Will provenance remain attached? Can the institution audit the output? Must synthetic reconstructions be labeled? What happens when the system produces an error?
This debate is not an argument against technology. AI can make handwritten archives searchable, increase accessibility, assist translation and help identify relationships across collections that no individual researcher could examine at comparable scale. The danger lies in treating efficiency as a substitute for knowledge.
Cultural institutions should not simply ask what AI can do with their collections. They should decide what conditions AI must respect before it is allowed to use them.
Sources:
International Council of Museums, “UNESCO and ICOM Global Survey Finds Museums Embracing AI, but Governance and Capacity Lag Behind,” September 17, 2026. This provides the survey population and the figures concerning AI adoption, institutional policies and staff training.
UNESCO, “Reimagining Museums in the Age of AI,” October 5, 2026, addressing authenticity, provenance, intellectual property, cultural representation and human oversight.
European Heritage Hub, “How Cultural Heritage Can Make Public AI a Reality,” October 3, 2026.
European Commission, explanation of the AI Act’s copyright obligations and text-and-data-mining opt-outs for general-purpose AI providers.
European Commission, “Code of Practice on Transparency of AI-Generated Content,” updated October 19, 2026.
Comments