The launch of Chat-GPT in November 2022 prompted a Cambrian explosion of discussion about how Artificial Intelligence would transform how we work, communicate and interact with the world around us. But the adoption of AI, as it turns out, has been far more lethargic to date than these claims suggested. 

In museums in particular, interest in AI has been high but practical applications appear limited. In a survey by UK Heritage Pulse on use of AI in the heritage sector, only 24% of respondents had used AI as part of their role, with most utilising it as a tool for speeding up internal processes around idea generation and future planning, for data analysis or to help with marketing efforts. Most of these applications relate to the inner workings of the museums, not yet impacting the visitor experience or analysis of collections. 

But there are also opportunities to utilise AI in a more visitor-facing manner and to expand the accessibility and interpretation of collections, with some larger museums starting to trial new approaches tentatively, which may point to an increasingly interactive future. 

Research examining the opportunities for AI in museum has identified a number of key areas where AI could be transformative. These include: visitor engagement, collection analysis, and audience analytics.

Audience analytics is perhaps the most straightforward application, but poses interesting ethical questions. Processes such as sentiment analysis and audience prediction can aid institutions and designers in understanding the needs of audiences better and designing spaces for that fit the required capacity. 

At The National Gallery in London, AI is being utilised to predict demand for temporary exhibitions to ensure that an appropriate amount of space is allocated to each project. They can also forecast attendance 12 to 18 months in advance of opening and, after 3 weeks of an exhibition opening, AI can predict final ticket sales.

Sentiment analysis can capture the depth of feeling towards an exhibition, providing valuable feedback. Combining data from Trip Advisor and Google reviews with social media engagement can offer insight into what aspects of an exhibition were most engaging and be utilised to gauge where museums can improve their offering.

Data can also be collected to provide information to visitors about how busy a space is or even to analyse which objects in a collection attract the most interest, creating heat maps which illustrate patterns in visitor behaviour. However, whilst museums might have access to data from wi-fi connections or similar sources which allow for this granularity of detail, they are still reluctant to utilise it due to lack of explicit permission from visitors. Museums must always balance the risks against the rewards and ensure that their actions are transparent to avoid any diminution of trust on the part of visitors.

Visitors often start their journey on a museum website, offering multiple opportunities for AI to offer additional support where museum staff may be unable to assist. Visitors could have support planning their visit, asking questions of an AI-powered chatbot which can make tailored suggestions and provide specific information.

There are even suggestions that AI could transform interpretation, creating digital solutions which can be tailored to the individual needs, language and age of each visitor. Graphic displays, lighting and content can be altered automatically according to audience needs, making exhibitions more accessible and engaging. 

Large language models are a type of AI algorithm which can generate language based on a data set it’s been given. These models could be trained on data specific to a prominent historical figure or text, allowing visitors to ask it questions and receive tailored answers. This approach has been used by the Shoah Foundation to allow students and researchers to ask questions to holocaust survivors and receive answers specific to their testimonies. 

When it comes to analysing large collections, AI presents numerous opportunities, and numerous challenges. For museums with large collections, thousands of objects can be hidden away from visitors in the archives for years on end. Here AI can help these collections be more accessible by visualising objects in 3D or making objects searchable on museum databases. 

At the Metropolitan Museum of Art (MoMA), researchers have found that visitors to their website struggle to search through their online collections due to a lack of tags. Tags are part of the metadata that comes with an object, they are the key words that describe the style or characteristics of a work of art, allowing the object to appear when a search mentions those terms. Tagging images is often dull and laborious, but at MoMA they are trialling using AI to automatically tag images, helping make their online database easier to use. 

There are limits to the success of this project as tagging is often subjective, meaning that an AI algorithm may not assess a painting in the same way as a person. But there are opportunities to develop this work, even opening opportunities for curators to look at paintings differently by sorting artworks by shape, colour or line, creating previously hidden points of comparison.

Despite these exciting possibilities, there are understandable hesitations around the use of AI in the cultural sector. There are multiple reasons for this, including a lack of specific expertise, the high cost of adoption, ethical concerns about the use of data and anxieties around the sustainability of training AI models. While AI offers a great deal of opportunity, it also introduces an element of risk into museum processes, requiring careful consideration by individual museums about how they should engage. 

Understanding the regulations that govern AI usage, such as GDPR regulations for the use of data, and also understanding the level of AI literacy within each organisation creates a solid base from which museum scan start to explore the potential of AI in their organisation.