Navigating AI – Why Heritage Can’t Do This Alone

AI is set to change the world as we know it. It has already started to change the way we work, including in the heritage sector! Our Deputy Head of Heritage Information, Susan Hamilton , reflects on the impact of AI on our past.

Across Historic Environment Scotland and the heritage sector, we’re constantly negotiating the rapid changes that technology brings. When I started in the sector, we still dealt with huge volumes of paper files. In fact, one of my first roles was diligently processing paper records into our database.

Fast forward a few years and we’ve all seen, and perhaps been slightly scared by, the headlines about the computing algorithms we call Artificial Intelligence (AI) and how it is set to change the world.

This is already evident in changing the way we work and the technology’s impact on art, culture, and the creative industries. Anyone scrolling social media might have noticed a recent increase in AI images of varying quality, which have been generated for a wide range of purposes. A recent example, showing Edinburgh Castle, is available in this article on Stable Diffusion.

While Chat GPT, copyright infringement and “deepfakes” may be grabbing the headlines, there are many other techniques and applications of AI in circulation. It is transformative – but things can get overwhelming when keeping up to date with the rapid growth of this type of technology. ?

Benefits of AI within the Heritage Sector

So, what about Heritage and the work we all do in our part of the historic environment sector? In our team, we’re taking an approach that is best described as “curious, yet cautious.”

We’ve acknowledged that there’s little point in carrying on with existing practices and as if nothing is changing. That’s a surefire way to be left behind whilst carrying reputational risk, as the public and our partners will expect us to know how to handle AI.

Where it is appropriate, we’re using AI techniques to enhance understanding and make our processes more efficient. We’re using a technique called Natural Language Processing (NLP) to pull useful information out of unstructured text – text written in sentences and paragraphs just like this article. Colleagues in our Heritage teams have spent time over the last few years testing and using another technique, machine learning, to increase our knowledge of archaeological sites in Arran. In this example, AI was combined with work on the ground, to test survey results.

Introducing the Turing Way

We’re aware we can’t keep up with this rapidly growing area by ourselves, so we are learning from some of the best. We are adopting the recommendations of the Turing Way in our own data science work, so when we use techniques such as Data Mining and Natural Language Processing, our methods are recorded along with the results.

This is important so we, and others, can reproduce and retest our findings. A way to think of this is a bit like a recipe: we record what ingredients we used, in what combination and with what equipment, so someone else can come along and make the same dish again. Regulating our input and output data also makes our AI model more reliable if we can reproduce our initial results. Transparency and recording our work are vital to ensure public trust in the information that we hold and make available to the public.

The Turing Way was developed by the Alan Turing Institute, the UK’s national institute for data science and artificial intelligence. We’ve also participated in a Turing Institute project aimed at professionalising roles in data science; to help future generations navigate this area and to ensure the language we use to describe roles and tasks is consistent.

We’re keeping an eye on initiatives such as Designing Responsible Natural Language Processing at the 英国爱丁堡大学 and are partners in their Centre for Doctoral Training. Programmes like this aim to ensure the safe and responsible use of AI techniques. We hope to work with their PhD students on data projects, benefiting from their access to the latest techniques and their computing expertise while providing “real world” problems for the students to work on.

Building Human Knowledge to Improve AI Learning

Closer to home, we’re working with colleagues across our teams on a The Scottish Government initiative to develop our data maturity, which will identify and embed data skills and knowledge across the organisation. This hopefully means all of Historic Environment Scotland will be able to benefit from the expertise being developed across our broad range of teams and the wide variety of work which we do. This is already beginning to reap rewards! A recent information-sharing session resulted in a team learning about the existence of a tool that links data behind the scenes, preventing them from having to figure out a similar answer manually.

We’re also supporting team members in more formal learning on subjects including Data Science as part of a 英国斯特拉斯克莱德大学 Graduate Apprenticeship scheme. This mixes on-the-job experience with academic learning and will ensure we have the most up-to-date knowledge.?

However, the recent focus on AI and data science is making us realise that human expertise and knowledge must remain at the heart of our work. It is our deep knowledge of the information we hold, and of the way generations of research is represented in our database and archive, that will alert us when something doesn’t seem quite right. This is vital when anecdotal evidence is telling us that AI-generated reports and images may already be circulating within the sector. To provide a simple example, we know that we have a lot of work still to do in classifying the records in our database.

Currently, any data science, or artificial intelligence “pointed” at this dataset will not provide comprehensive information and may risk providing incomplete answers. Defining Period or Century indexing terms on database entries will enable a more thorough AI analysis of our records.

While we are still in the very early days of exploring this technology, we believe this pragmatic mix of data literacy, expert knowledge and cautious evidence-led use of appropriate AI tools and partnerships is the best way forward in this rapidly changing environment.