Skip to main content

AI is a very broad field, however, and has much wider possibilities for practical applications than the generation toys that are currently drawing the public’s attention and causing disruptions that are only going to be the tip of the iceberg. For the past decade, AI has increasingly been sitting as a quiet layer underneath systems we already take for granted in our daily lives, such as voice-recognition software and translation tools, never mind what’s powering certain organisations to streamline all sorts of internal operations. It’s just invisible to us because in those cases it’s often functioning correctly, subtly enhancing or augmenting an experience without our knowledge.

This is also true of museums and visitor centres around the world, which have been experimenting with systems and technologies, and in some places already permanently implementing them to bring cataloguing, curation, and the visitor experience into the 21st century.

Consequently, this is an overview article to spark thoughts about what is being tested and implemented right now, as well as what that might mean for the future. It isn’t going to examine any specific technologies or systems in detail to evaluate their pros and cons but rather look at how AI could help and where in an organisation it could be integrated to be most beneficial. We will, however, mention a few ethical considerations at the end that anyone contemplating AI technologies should bear in mind before they make any decisions.

Enhancing The Visitor Experience

Institutions are having to evolve due to greater expectations from potential visitors and more friction that holds them back from visiting a place. For many people, the Internet now gives them access to “everything” – often for free – so an institution has to work harder to convince certain demographics that there is still great value in going in person to a museum or visitors’ centre and physically experiencing the collections.

Institutions can use AI to enhance the visitor experience by offering options that help to encourage interactivity and immediate information seeking, without the expense of having to hire more staff to talk to and guide the visitors directly. This can take the form of multilingual translations; chat functions in an app or on a museum device that can provide quick answers to simple questions; and voice-activated technologies that would allow a visitor to ask a question verbally and receive a spoken answer, which has the added benefit of offering more accessibility to people with visual impairments.

More intricate systems allow for AI-powered exhibition objects, such as recreations of historical figures that visitors can interact with, which will naturally lead to each person having a slightly different experience. AI systems can also enable personalised experiences that guide visitors to exhibitions or objects on display that would most appeal to them, which helps with visitor retention and repeat visits. One method for doing that is to have a system scan visitors’ social media accounts (with permission), such as Instagram, to analyse their posts and determine their aesthetic preferences (a caveat being that you might direct people into aesthetic bubbles and guide them away from objects that will challenge them, which is often also an important part of a museum experience).

The possibilities for new kinds of visitor experiences are endless. Different museums are trying different approaches and we’re right at the beginning of many new methods that will emerge to draw, educate, and entertain visitors.

Making The Most Of The Collections

Institutions are sitting with vast collections of artefacts and artworks that have never been researched or put on display because they haven’t been properly examined and catalogued. Instead, they sit in storage, untouched, often for decades. Once trained, AI systems can sort and classify these objects a lot faster than a human can, plus they are able to perceive details that humans might miss, which means better categorisation for future research or exhibition purposes (think of botanical samples or fossil pieces, for example).

AI can also help with digitisation efforts, especially for written materials where optical character recognition won’t work because it’s not in a font that’s easy for the system to recognise but is rather handwriting or logo-syllabic script. Instead of just scanning a page as an image file and saving it in the institution’s archives, never to be seen again, a trained AI system can interpret the characters on the page and transcode them into a text format that can be searched or fed into another system to be translated into another language or reproduced as audio for accessibility purposes.

Restoration projects can also benefit from AI and there have already been some tests on a handful of damaged or incomplete artworks, as well as experiments with training AI models to get them to figure out a missing piece of art correctly. In future, these systems will get better at working out what should be there or what was there. Similar systems will also be used in the examination of incomplete fossils to get a better idea of what is missing.

Finally, AI systems are able to see obscure patterns and make unusual visual connections so all this data can be analysed by a system to find new ways to interpret and display the pieces in the collection, as the system will make connections between pieces and group objects in a way that wouldn’t necessarily occur to a human as we’re predisposed to focussing on a time period, style, or geographic location. This can lead to novel exhibitions that will draw intrigued visitors because the approach to the curation is so different.

Streamlining And Optimising Operations

Data gathered about visitors to an institution can be analysed in many different ways to increase operational efficiency. This includes monitoring visiting patterns and tracking attendance to plan future events and exhibitions based on what might draw a crowd in terms of content or even time of year, as well as analysing ticket sales to see how many pre-sold tickets go unused to allow the institution to release more tickets for sale earlier. Physical traffic patterns around an exhibition space can also be analysed to see if, and how, the movement of people results in flow blocks in high-traffic areas or if there are low-traffic areas that have dead spots that therefore should be given some attention and reworked to increase visitor numbers in those spaces.

For institutions that have gift shops, purchasing data can be analysed to streamline operations, save money, and increase sales. What is more popular? What doesn’t sell? Are certain kinds of purchases seasonal? Can the selection of items for sale be reduced so there are fewer different items, which all come with logistical costs (sourcing, ordering, shipping, displaying), but there is more stock of items that do sell well so profits will increase?

Ethical Considerations

It is one thing to build a customised system that has been trained on your institution’s data, although it’s still very important to consider biases in the data that might then be introduced into the system and to adjust for that. The controversies currently swirling around publicly available systems largely centre on the data sets that these tools have been trained on, although the issue is also valid for private systems.

In the case of text-generating tools, it is frequently information that has been scraped from the Internet, often with no oversight or curation. This means that the training data will contain biases that will be replicated in its output. At its worst, this has led to racial and gender biases. The data will also have an end date based on when it was scraped and if it isn’t updated regularly it will lack new information that might inform scientific or historical accuracy. The information that is scraped is likely also information that wasn’t made available for AI training with permission from its owners: just because something is publicly visible doesn’t mean you can just take it and use it for something else.

Similarly, most text-to-image generators have been trained on vast repositories of art on the Internet. In particular, these include websites that allow artists all over the world to build and showcase a portfolio but there have also been private companies that develop art software – and which operate cloud systems where artists can save their work in progress and library items privately – that have accessed the art to train a system without the artists’ knowledge or consent.

The same is true of systems that are now able to generate audio, such as sound effects or music. The news hasn’t focussed on them as much as other types of generating systems but they also exist and may have been trained on data sets that weren’t ethically compiled.

If you opt for using an external piece of software and/or external data set it is imperative that the data was all sourced ethically from people who gave permission for their work to be included in the system for AI processing and training. The law hasn’t caught up with this yet but eventually, there will be frameworks in place. Until that happens it’s important to be educated on the issues and to make morally sound decisions for yourself and your institution.