Intersecting perspectives: artificial intelligence in museums

Artificial intelligence (AI) is a trending topic across all professions and sectors—and museums are no exception.  

Some cultural institutions are eager to incorporate this technology into their practices and tools, not only to enrich the visitor experience but also to support in-house professionals.  However, the implementation of AI in museums raises major operational and legal challenges. 

In this new Experts’ Corner, Charlotte Clergeau, Digital Projects Officer, and Léa Karouby, Junior Legal Officer at France Muséums, present a practical case  of how AI has been applied in the recent exhibition jointly organised with Louvre Abu Dhabi, From Kalila wa Dimna to la Fontaine, Travelling Through Fables, presented at Louvre Abu Dhabi from 26 March to 21 July 2024.  

Can you tell us about the solutions that were implemented in the exhibition?  

CHARLOTTE: Covering an area of 600 square metres, the exhibition explored the rich heritage of fables, highlighting their educational role and social relevance. Within this framework, several interpretive tools were developed, one of which incorporated generative AI. 

An interactive screen invited visitors to generate their own narratives by selecting a writing style, characters and a moral, producing a customised fable generated by AI.  

In the framework of this exhibition and in museums in general, what types of AI might we be dealing with?  

CHARLOTTE: the most popular type in museums is Generative AI or GenAI. This technology makes it possible to generate various types of content: images, texts, videos or music. GenAi is embedded in various formats, such as a chatbot for example, also known as conversational AI. 

Several technologies can also be merged for a more complete experience. For example, by combining image recognition with a conversational chatbot so that the system can recognise an image and generate related content as part of a question-and-answer session with the visitor. 

So, there are different kinds of AI, and the way they are used varies according to the area in which they are applied. 

How and why do cultural institutions use this technology?    

CHARLOTTE: There are three main ways in which AI is used in museums today: 

• For mediation and visitor assistance: helping visitors understand exhibits or the topic of an exhibition during the visitor journey and/or enriching the experience with interactive and personalised approaches 

• For creative purposes: artists can use the technology to create artworks  

• For curatorial purposes: the technology can be used to help us to manage collections and to facilitate the exploration of digitalised catalogues. 

AI can also be used to analyse visitor data, for example. 

What were the challenges that arose regarding the exhibition?  

CHARLOTTE: on the production side, when we launch the production of an interpretive tool incorporating AI, for example the one used for this exhibition, (Chat GPT-4 developed by Open AI), operational challenges quickly arise. The main challenge lies in the current difficulty of accessing AI training data. 

For example, for a multilingual tool, if AI lacks sufficient data in a particular language, this can affect the quality of the output, which will fall short of responding to the needs and expectations of the audience. 

Data-related issues also arose where the literary styles of the fables were concerned. For our tool, we asked AI to write fables in the style of Jean de La Fontaine or Ibn al-Muqaffa.

This required an imitation of the literary style in terms of its form, its vocabulary, its characters and its distinctive landscapes. Because we don’t have access to the training data, we have to formulate very precise AI prompts, even though they might seem extremely basic.  

It’s also crucial to be able to control the texts that are generated, especially as regards the local context, to avoid undesirable content, ill-adapted turns of phrase and odd translations. There are two effective solutions: using a “post-prompt” or “negative prompt” to eliminate predefined sensitive subjects; and designing the user interface with limited selection criteria to prevent people from using the tool inappropriately.  

LÉA: Legal issues have also been raised, and we must bear them in mind for future uses of AI in museum projects.  

The role of the legal department is to ensure that visitor aids and interpretive tools are compliant with current regulations.  

For the exhibition on From Kalila wa Dimna to la Fontaine, Travelling Through Fables, we were eager to take these regulations into account.  

The main challenges are: 

During project development

• TRANSPARENCY and COPYRIGHT: Ensuring that the training data is legally compliant and does not infringe copyright. Checking that authors and rights-holders have not opposed the use of their work by AI (opt-out clause, article L.122-5-3 of French copyright law) and that the material generated does not constitute counterfeit. 

For this exhibition, the economic rights issue didn’t arise because La Fontaine’s works are in the public domain. However, the moral right continues to apply after the artist’s death and can be claimed by copyright holders. Moreover, an ethical challenge could have also been posed around the question of “bringing back to life” deceased artists.  

Also, when using platforms such as OpenAI for a mediation project, protected data may be used unknowingly, which means that risks must be assessed on a case-by-case basis. 

Once the project is up and running: 

• PROTECTION OF AI OUTPUT: For this exhibition, the prompt was very restrictive which meant that human intervention was insufficient to merit any claim that the output constituted an original “work”. 

For the exhibition, our main operational and legal challenges were related to the generation of text, the central element of the device. However, it is important to remember that these challenges also apply to image generation, which could be considered for an exhibition. For the generation of images by AI, it is essential to comply with the rules above-mentioned and to check the general terms and conditions of sale (GTCs) of the platforms to ensure that the images are free of rights. 

CHARLOTTE: It’s also essential to inform the user when content is AI-generated. At France Muséums, we are eager to remain as transparent as possible and to make our visitors aware of how the technology is being used.  

With this in mind, we decided to call the tool an “AI Fables Generator”. We also included an exhaustive description of the tool including a word of caution warning users of possible inconsistencies arising from the use of AI. Instead of maintaining an illusion where this technology is concerned, we clearly state that it remains fallible however innovative it may be. 

LÉA: This is the direction taken by EU regulations on AI, and future international regulations will undoubtedly do the same, which confirms that we are adopting the right approach where AI-based interpretive tools are concerned. 

How does the use of AI call into question our current practices in the museum sector?

LÉA: The development of AI in museums has allowed us to re-examine the application of copyright. As a legal counsel, it’s always beneficial to work on these new types of mediation to assess the risks involved and raise new questions on current and future regulations. We have also organised an informal discussion on this subject with the legal directors of our partner museums. 

Charlotte: I’m convinced that AI is a promising technology for museum mediation, as evidenced in the exhibition From Kalila wa Dimna to La Fontaine, Travelling Through Fables. Although increasingly popular and widespread, AI is still under development. New modes of use appear all the time, which means that its limitations and potential for museum professionals remain to be defined.  

At France Muséums, incorporating AI into our interpretive tools led us to re-examine our mediation goals and helped us to ensure that the way it was used in the exhibition was appropriate, namely by drawing on the concepts on display to offer visitors an enjoyable hands-on experience exploring the ideas and aesthetics of fables.  

When used as an interpretive tool or as part of a visitor journey, AI must serve the teams who constantly strive to improve the visitor experience. Its innovative potential must encourage us to be even more exacting and ambitious where mediation and accessibility to the public are concerned. 

Our multi-disciplinary team is available to project leaders for legal and operational management support and is able to provide training on the use of these new technologies. Our goal is to rethink our practices and to create coherent, well-designed experiential tools for museum visitors. 

Discover our areas of expertise