Revolutionizing heritage management: the Emerging role of AI

Heritage management, traditionally reliant on human diligence, is being revolutionized by AI, with successful projects like the reconstruction of Rembrandt's "The Night Watch" and Beethoven's 10th Symphony. With AI's rapid market growth and its increasing use in heritage, the UK's National Lottery Heritage Fund commissioned Dr. Mathilde Pavis to write a report that explores AI's potential, limitations, and ethical implications in this sector.

Until now, the workload of managing our collective past has fallen to diligent teams of heritage staff and volunteers, manually transcribing old handwritten texts, surveying maps for historic sites and doing everything else that “heritage” entails. Now, however, the “Artificial Intelligence” (AI) craze may be opening new possibilities for what computers can do for us. Indeed, researchers have already utilised AI to succeed where humans have previously failed.

High profile examples of AI tools in heritage include the full reconstruction of Rembrant’s The Night Watch and the completion of Beethoven’s 10th Symphony (both done under expert guidance). In the Netherlands, over 6000 volunteers have helped to train an AI that can automatically find archaeological features in mapping data.

But keeping track of AI’s development is a challenge in itself: AI’s global market size has doubled in the last two years, a pace expected to continue up for years to come according to Statista.com. To keep track of how these powerful tools could shape the heritage sector, the UK’s National Lottery Heritage Fund commissioned a report from international ethics and data researcher Dr Mathilde Pavis. Here’s the breakdown:

How do AI systems work?
Whether you’re asking ChatGPT to create panels for your museum exhibition or trying to create metadata for your digital catalogue, the underlying AI all works the same way: the system is given a large dataset and then trained to see patterns and correlations. When it’s been given enough data and training, the AI will be able to look for those patterns itself – much faster than a human could. It’s worth noting that AI can refer to anything from the systems that drive autonomous vehicles to off-the-shelf generative AI chatbots such as ChatGPT.

For a basic example of how AI tools can help the heritage sector, Dr Pavis points to the Living With Machines project, who have applied AI to the tricky task of extracting geographical data from collections. Their technique – “smart annotation” – produced 25,000 georeferences in just 3 hours when applied to the British Library’s newspaper title catalogue, and they hope to apply the technique to other types of metadata in the collection too.

Can I trust AI?
Yes – and no. Whilst the possibilities may seem endless, it’s important to recognise that AI systems do make mistakes, particularly if they’re poorly trained or given small data sets. When researchers were using algorithms to find archaeological features in the Netherlands, they found that the system was initially unable to find the pre-existing charcoal kilns as there was no training data for them.

Secondly, bias is a very significant concern when working with AI systems. Discrimination has been a known issue with AI systems for many years now, and has been a tricky problem to get around. These biases are impossible to fully avoid, making it important for heritage institutions using AI to not reinforce harmful misinformation by accident.

It’s important to remember that modern AI tools are not all-knowing. They rely on large datasets, thorough training, and expert supervision – without all these, they’re just misinformation generators!

Is AI ethical?
The ethics of AI systems is still very much a debated topic, but especially important when thinking about usage in the heritage sector. The possibilities to save time and money by outsourcing to AI are highly tempting, especially when budgets are squeezed – but many AIs have been trained on datasets without explicit permission (ChatGPT is currently involved in multiple lawsuits regarding its training data). The European Union is introducing the “AI Act”, aimed at creating legally binding requirements for AI systems, and many countries around the globe are investigating new regulations to keep control over the industry.

For heritage institutions, provenance and traceability are vitally important. AI systems are often seen as “black boxes”, since their inner workings are either intentionally hidden or too complicated for non-experts to understand. When using AI tools, Dr Pavis recommends working with traceable data sets such as the institution’s own collection, and to avoid sensitive content such as ancestral remains, contributions from children or items of spiritual significance.

So what can I use AI for?
With the recent breakthroughs, the sky really may be the limit on what AI could be used for – here’s three general areas where AI could make a big difference for the sector:

  • Heritage and collections management. AI tools can assist staff and volunteers in heritage and collections management by making content, information, and collections easier to find – such as generating metadata, extracting information from images, and highlighting historic biases.
  • Visitor experience. Content and experiences that are driven by AI have great possibilities for the heritage sector. Automatic transcriptions of events and videos can be done quickly and efficiently with off-the-shelf AI tools, offering museums and heritage sites simple accessibility solutions. Some museums have even created their own chatbots, designed to help users with common questions.
  • General business operations and management. AI systems can help to handle operational and management tasks, such as generating summary reports, composing emails, or refining content for marketing material. For larger organisations, AI systems have the potential to forecast performance of exhibitions – a system that the National Gallery in London has already begun using.

In the evolving landscape of heritage and AI, balancing risks is crucial. Dr. Pavis offers practical solutions, emphasising the importance of rigorous testing against discrimination, inaccuracies, and transparency issues. Despite these pitfalls, the message is clear: AI systems are here to stay, and it’s up to us to use them responsibly.

This article was originally published in English. Texts in other languages are AI-translated. To change language: go to the main menu above.

Doneren