AI in Museums: Why Human Expertise and Trust Still Matter
AI can expand what museums and creative teams can make. But transparency is only the beginning. Cultural organisations still need accountable people to determine what is accurate, appropriate and worth saying.
The museum AI debate has been framed too narrowly
Much of the discussion about generative AI presents two positions.
One portrays AI as an inevitable answer to limited budgets, production pressure and changing audience expectations. The other treats its use as fundamentally incompatible with authentic creative or cultural work.
Neither position is adequate for museums.
AI can increase creative and production capacity. It can help teams test ideas, prototype interactions, process material and investigate new approaches to accessibility, language or personalisation.
It can also produce plausible errors, obscure authorship, reproduce biases and create uncertainty about rights. Used without sufficient judgement, it can make interpretation feel generic precisely where visitors expect knowledge, specificity and institutional care.
The useful question is not simply:
Was AI used?
It is:
Who was responsible for the result?
Transparency matters—but it is not a quality standard
The distinction has become more urgent.
The European Commission’s Code of Practice supports implementation of Article 50 of the EU AI Act. Relevant transparency obligations concerning certain AI-generated and manipulated content became applicable on 2 August 2026. [European Commission guidance]
Clear marking can help audiences understand when they are encountering synthetic or manipulated material. It can also encourage organisations and suppliers to document their processes more carefully.
But disclosure cannot answer every important question.
A label cannot tell a visitor whether an interpretation is historically accurate. It cannot demonstrate that community knowledge has been respected, that a source was reliable or that an image is culturally appropriate. Nor can it explain whether a creative decision strengthened the experience or merely made production faster.
Transparency is a necessary layer of trust. It is not a substitute for quality.
Museums do more than provide information
A museum is not valuable because it can produce a large volume of fluent text.
Its value comes from selection, context, evidence and stewardship.
An object exists in relationships with provenance, people, memory, contested interpretation and material reality. A historical account may involve uncertainty that should be acknowledged rather than smoothed away. A reconstruction may be visually persuasive while relying on incomplete evidence.
These are not merely data-processing problems. They require judgement.
Generative systems are particularly good at producing apparently complete answers. Museum practice often requires the opposite discipline: recognising where evidence is partial, where terminology is disputed and where several perspectives must remain visible.
Human expertise is therefore not a decorative layer added after automated production. It establishes the boundaries within which a tool can be used responsibly.
Human creativity is more than making assets
Creativity is sometimes discussed as though it were simply the labour required to produce a picture, script, sound or animation.
In an interpretive experience, creativity starts earlier.
It includes deciding:
· What is the central idea?
· Whose perspective is present?
· What should a visitor notice?
· Which details carry emotional or historical meaning?
· When should the technology recede?
· What should remain unresolved?
· How does the experience relate to the physical place or collection?
A tool can assist with expressions of those decisions. It cannot assume responsibility for making them well.
Audiences do not experience a production workflow. They experience the cumulative effect of hundreds of choices: tone, pace, language, imagery, movement, sound and the relationship between the digital layer and the real object or location.
Efficiency may help a team make more. Creative intent determines whether more was worth making.
Start with institutional purpose
The Cleveland Museum of Art describes its approach as using AI to invite curiosity and create new pathways into its collection while preserving human elements at the centre of the museum experience. [Cleveland Museum of Art]
That is a productive place to begin because it does not treat AI as the objective.
The objective is curiosity, learning or access. AI is one possible means.
The same test can be applied to almost any proposed use:
· Does it help visitors engage more deeply with a collection?
· Does it improve access to content that was previously difficult to reach?
· Does it support another language or mode of engagement?
· Does it reveal meaningful relationships within an archive?
· Does it help staff investigate an idea before committing substantial resources?
· Is the result better for the visitor—or merely easier for the producer?
Technology earns its place in an experience through the answer.
Archives show why human and machine capabilities need each other
Digitised collections create an obvious opportunity. Museums can hold thousands or millions of records, images and fragments of knowledge that could never be presented in one gallery.
Computational methods can assist with classification, relationships and discovery. The Metropolitan Museum of Art has previously described combining machine suggestions with contributions from people to refine collection knowledge. [The Metropolitan Museum of Art]
The principle remains useful: scale and judgement solve different parts of the problem.
A system may identify possible connections across a large dataset. People assess whether those connections are meaningful, accurate and appropriately expressed. Curators, archivists and communities provide knowledge that cannot be recovered simply by generating more fluent prose.
Used well, technology can help institutions make more of their collections discoverable. Used carelessly, it can make uncertainty harder to see.
Accuracy must be designed into the process
Fact-checking should not be a final emergency step.
A responsible workflow needs to establish:
1. Which sources are authoritative.
2. Where uncertainty must remain explicit.
3. Who owns factual approval.
4. What rights apply to every asset.
5. How AI involvement is recorded.
6. What cultural or community review is needed.
7. How visitors can distinguish evidence from reconstruction.
8. Who can correct or withdraw material if a problem appears.
This becomes especially important when external creative and technology partners are involved. A museum should know which decisions remain with the institution, which sit with the supplier and how evidence and approvals pass between them.
AI does not remove the need for that process. It makes the process more important.
Toolkit callout!
Put these principles into practice
Download our toolkit: The Museum AI & Digital Interpretation Trust Checklist—15 practical questions for commissioning or approving an AI-assisted interpretation project.
Accessibility needs the same care
AI-assisted translation, transcription, description and personalisation could help more visitors access cultural content.
But an automatically produced translation can be grammatically fluent and still fail to capture local usage, historical terminology or cultural meaning. An image description may identify objects while missing the reason an image matters. Personalisation may increase convenience while narrowing an encounter in unintended ways.
The answer is not to dismiss these applications. It is to combine them with native-language review, accessibility expertise, representative testing and clear responsibility.
Access is not achieved because content exists in another format. It is achieved when that format works for the people expected to use it.
The Panivox position: AI-assisted, human-accountable
At Panivox, we believe AI can form part of a serious creative practice.
It can widen the field of ideas, make experimentation more accessible and support teams in producing richer combinations of story, image, sound and interaction. Those possibilities are genuinely exciting.
But museums should not be asked to choose between innovation and integrity.
The stronger model is AI-assisted, human-accountable:
· People establish the purpose.
· Experts ground the content.
· Creators shape the experience.
· Rights holders are respected.
· Institutions make the final judgement.
· Technology extends what the team can achieve.
· The organisation remains accountable to its audience.
This does not diminish technology. It places technology where it can be most useful.
Museums hold trust because they do more than communicate. They care for evidence, acknowledge complexity and connect people with material, memory and knowledge.
Any creative technology used in that setting should strengthen those responsibilities—not offer a shortcut around them.
Exploring how AI, interactive storytelling or digital interpretation could support a collection or visitor experience?
Contact Troy at Troy@panivox.com to discuss an approach that is creatively ambitious, evidence-led and human-accountable.
Comments