Artificial Intelligence or AI is constantly in the news, promising to transform the future (for good or ill). But can advanced computing techniques also help us with the past? And, specifically, can they help museums to open up their historic collections?
Most museums, including ours, have vast collections and are only able to show a very small percentage in their public galleries at any one time. We’ve been fortunate to be able to create the Hawking Building, at the Science and Innovation Park in Wiltshire, a vast repository for more than 320,000 objects from our collection, which is open for public tours, school and research visits.
The next stage for us is to help people find what we hold in our wondrous collection. It is most timely then that the opening of the Hawking Building coincides with the end of our major three-year research project called Congruence Engine. The project has been looking into how collections of all kinds across the country can be linked together via their digital records. This pioneering project was one of five ‘Discovery Projects’ funded by the Arts and Humanities Research Council under its ‘Towards a National Collection’ programme to research how this ambition could be fulfilled.
Our project took industrial history and heritage as its subject, particularly looking into the histories of the textile, energy and communications sectors. This short film is designed to convey the spirit of our project’s investigation:
In many ways, heritage organisations are not well prepared for the age of big data. Curators most often work with very brief digital data to describe the objects in their care and on display. The many objects that have never been displayed often languish with very basic records, making them difficult to find via normal digital searches.
Curators get to know their objects better when they research them for display, but the databases that store information about individual objects have not, so far, been designed to store the connecting interpretations that are created for exhibitions. And nor do the databases as a rule include the interpretive text of publications, including the Science Museum’s once famous published collections catalogues, which are mainly just on paper.
The same rule applies to wider scholarship relating to our objects and their histories. All of these kinds of narrative, technical and contextual information has been kept in different siloes. This is only now coming to seem strange, and it is our world of big data that makes it seem strange.
In the Congruence Engine project, we set about applying a whole range of computational techniques to seek ways to link collections, including the very new generative AIs, including ChatGPT, that only became available a year into the project. Here are some of the things we made during the research:
- We created a prototype app that you can show an old photograph that includes and object – such as a phone – and it will produce images and details of related objects in the Science Museum Group and National Museums Scotland collection.
- We mined the old catalogue shown above for the names of objects, put them into an orderly list, and posted it online (on Wikidata) so that anyone with similar collections of looms or sewing machines will be able to find what each other have.
- We made a chatbot that can channel the author of a 100-year old wool manufacturing textbook to answer any question you might have about the related machines we have in our collections.
- We found a way to link the brief online descriptions of our objects to relevant articles and books. In future it will be possible to ‘trip over’ mention of an object as you read the history or, in the opposite direction, click from the object to the literature that brings it to life.
- We created complex digital ‘family trees’ that map people’s working lives to the machines they may have used and where, the things they made and the people they worked with.
- We showed that supervised generative AI is pretty good at producing catalogue entries for archives, though very bad at writing history!
- We used digital maps to show who bought what machines, and then looked alongside at other factors such as where the people who operated machines came from.
You can now read all about the project in our final published report, which can be found here. And, if you’re interested in the broader picture, please take a look at my plenary talk from the final conference here.
Machine learning, it turns out, can be pressed into service to make collections more accessible. Probably that means that digital methods will become increasingly important for museums and curators. We know that many ethical and environmental concerns arise from the new technologies, as anyone who follows the news will know. But at least with Congruence Engine and its sister projects we are well informed to take the next steps.
Congruence Engine was funded by the Arts and Humanities Research Council.