FIVE THINGS TO THINK ABOUT WHEN NAVIGATING THE USE OF AI IN YOUR MUSEUM:
1
The Power of the Real / Authenticity
The first time I held a Bronze-Age axe head in my hands, I was awestruck. This was something another human being had crafted and used, over 5000 years ago. It was a tangible touchpoint to the past. A former colleague called this the“ power of the real” – the magical ability of museums to put visitors in the proximity of real objects and products of human creativity and resilience. Museums provide experiences that most people can never have at home or online. That is a superpower that shouldn’ t be squandered. While generative AI can quickly produce images, videos and XR content that enhance exhibits and experiences, surveys show that visitors generally prefer engaging with museum staff and authentic objects as part of an overall meaningful and memorable social experience( Machado Corral et al., 2021). Keeping the“ real” front and center should be a priority. This applies particularly to social media. Think carefully before using AI images and video to attract followers— the very presence of a synthetic image( without even considering the off-putting“ uncanny valley” aspect of much of the current material) can dilute a museum’ s authentic voice, undermining audience confidence and loyalty.
2
The Things AI Does Well
Much of the concern about AI relates to the possibility of jobs being eliminated and staff replaced by very smart, yet not infallible, machines. An optimistic view would be that AI tools may be able to take over or improve some of the more repetitive and onerous tasks, freeing museum staff to spend more time engaging visitors and planning impactful experiences. There is no doubt that AI excels at many things. The integration of AI in museum DAMS( Digital Asset Management Systems) and CMS( Collections Management Systems) combined with the use of vector databases means faster and more powerful access to data. For example, a visitor could search for a“ painting with ancient boats” on an interface, drag a slider to determine how far afield the query should go, and the system might bring up related results like canoes, rafts, maritime tools or clothing— even if none of the assets have exact metadata word matches. AI systems like this are numerical and therefore language agnostic, so a visitor could enter a query in French and find Arabic-tagged artifacts. And by using an application like Smartify or Museumfy, that visitor can draw on the same information combined with image recognition to identify artifacts and receive relevant information on their phone.
By harvesting data from visitors, AI also promises personalized tours catering to specific ages and interests, and interactive exhibits that adapt to levels of ability. Museum visits can become more accessible through location-aware applications for visitors with hearing or visual impairments, or those on the spectrum. Data-driven insights into visitor behaviour can improve the overall user experience from ticketing to wayfinding. There is no doubt that AI excels at processing and using all that data, but that brings with it another issue …
3
What about Privacy and Ethics?
Large language models( LLMs) would not exist without the huge data sets needed to train them. And they need to continue consuming data to improve and become more“ intelligent”. ChatGPT, Grok, Claude— they all harvest your inputs and queries to build on their learning. It’ s crucial to think carefully and commit to due diligence when embracing AI in museums. When visitors enter information like their ages, interests, country of origin and so on, where does that data go? Who profits from it? Is the data anonymized? Is your organization seeking informed and clear consent from visitors, and providing access to the data privacy policy? In addition, AI is an inherently backwards-looking technology. It“ relies on historical data, reproducing old patterns instead of fostering progress”( Ziedler, 2024) and as a result is prone to bias. In his book“ The Alignment Problem”, author Brian Christian describes Google’ s early experiments with unsupervised machine learning, where it fed its model enormous datasets of text gleaned from newspapers and the internet. The goal was for the system to recognize patterns on its own, to develop a type of“ neural network” of language understanding. But it wasn’ t long before problems of bias arose. Asked to name the female equivalent of a male doctor, the machine answered“ nurse”. A shopkeeper in female form was a“ housewife”.
40 MATAHEF ISSUE 01