From depot to database: how AI is breathing new life into old coins

From depot to database

How AI is breathing new life into old coins

M and KU Leuven are dusting off old coins from their collections for NumisMatrix: an AI project that aims to halve the time taken to register coins and medals. This paves the way for forgotten histories to be rediscovered: could it be the springboard for new research?

AI: the numismatist for repetitive tasks

The Flemish Government supports NumisMatrix as an Innovative Partner Project. ‘It could still go in any direction,’ emphasises Soetkin Vanhauwaert, Assistant Curator of the KU Leuven art collection. ‘As our understanding develops, we are devising a tailored methodology, specifically for numismatics,’ adds Benedicte Dierickx, Collection Manager at M. ‘For the registration process, each coin is photographed, measured and weighed by hand. Crucial metadata.’

Ludovik Segaert, digital applications specialist at M, says: ‘That data set is then neatly fed into a VLM, a Vision Language Model. This is an AI that can “see” and interpret images.’ The KU Leuven research group Augment is responsible for this AI aspect, testing and fine-tuning various models and prompts.

Soetkin Vanhauwaert, Assistent conservator van de kunstcollectie van KU Leuven, t Benedicte Dierickx, Collectiebeheerder bij M en Ludovik Segaert, medewerker digitale toepassingen bij M — Photo: Kimberley Dhollander for M Leuven

Is AI stealing your jobs?

A chuckle. ‘Only the work we’re happy to hand over. It actually leaves us more time for the real thinking,’ says Ludovik.

Benedicte: ‘AI excels at pattern recognition, especially of mass-produced objects, like coins. We hope to speed up basic identification and achieve 95 per cent accuracy, but we will always need experts to cover that 5 per cent gap.’

Ludovik adds: ‘AI is also a people-pleaser, so the right prompts are crucial.

“Hey AI, identify this coin, but if you don’t know, just say so.”’ Hallucinations and bias are always a risk, so human spot checks will remain vital for detecting plausible but entirely fabricated AI responses.

Met het quickPx toestel kunnen we het object ook wegen en meten, wat het proces versnelt — Photo: Kimberley Dhollander for M Leuven

People as critical quality control

Citizen science is a form of scientific research in which non-professional members of the public collaborate with professional researchers. It is a key component of NumisMatrix, as collection managers, students, experts and volunteers are all involved in identifying coins. DigitGLAM, the digital humanities research group, is responsible for the project’s communication and participatory aspects.

Together with a professor of archaeology and other experts, the citizen scientists are tackling what AI cannot identify: variants, archaeological finds and pieces damaged during the world wars. And KU Leuven’s collection also contains non-Western coins, with very different scripts and classification systems. It remains to be seen how NumisMatrix performs in this regard.

Quick, before the knowledge is lost

Coins are disappearing from circulation. And so is the knowledge. An ageing community of experts poses a silent threat to numismatics: the niche collectors and metal detectorists are still around, but young people are hardly finding their way into the field. ‘We’re worried that this expertise will simply disappear,’ Benedicte nods, ‘and AI can help us bring this knowledge together and make it digitally accessible.’

When asked where NumisMatrix might be in five years’ time, Soetkin’s answer sounds wonderfully ambitious:

‘The ObsIdentify for old coins! A widely accessible, non-commercial platform where anyone can identify their finds. In my dreams, that would be the ideal outcome for NumisMatrix.’