Bridges Across Time: Heritage, Infrastructure, and the Challenge of Building Better AI
I attended a fascinating lunchtime seminar this week on the potentials for archaeology and AI, which brought together perspectives from across different departments at Durham University. One of the talks was by Prof Jelena Ninic, a civil engineer whose work focuses on structural assessment and maintenance of infrastructure, particularly ageing transport networks such as railways and bridges. Her presentation discussed how heritage is not something separate from modern engineering but fundamentally embedded within it. Much of the infrastructure we rely on every day is in fact, historic. Across the UK and elsewhere, roads, bridges, railways, tunnels, often have origins in the nineteenth century, if not earlier. These are complex, evolving systems that have been repaired, adapted, and extended over decades or centuries. As such, they present a series of challenges that are as much archaeological as they are engineering in nature.
A key challenge is maintaining historic infrastructure. The UK already spends around £20 billion annually on road and rail infrastructure yet faces maintenance backlogs of £17 billion for roads and nearly £6 billion for bridges alone. Many of our assets are now reaching a critical stage of deterioration, having been built in the late nineteenth and early twentieth centuries and subjected to levels of use and environmental stress that far exceed their original design parameters. The consequences include for example, weight-restricted bridges, disrupted supply chains, longer journey times, and growing pressure on already stretched maintenance budgets. This is where the issue of preventative maintenance, which Jelena emphasised, becomes so important. Evidence suggests that planned, preventive maintenance can reduce whole-life costs significantly, compared to reactive approaches, while delivering wider benefits in reliability, safety, and sustainability. And yet, for many infrastructure systems, there remains a persistent tendency towards reactive intervention, driven in part by uncertainty and incomplete knowledge about the condition and history of assets.
Part of Jelena’s work is developing ‘digital twins’ of historic rail tunnels, to monitor their condition and support maintenance, and make this more cost and time efficient. There are two points to consider. Firstly, the importance of an archaeological understanding of historic structures, and secondly, archaeological expertise inputting into AI models that engineers are developing. Historic structures are much more complicated than modern tunnels, the latter typically being more uniform in shape and materials, whilst the former are more complex, layered, irregular, and present a much greater modelling challenge. Archaeology provides a crucial understanding of how historic structures were built, modified, and have deteriorated over time, not just how they look today. Archaeology can identify hidden construction phases, characterise historic materials and their behaviour, and reconstruct long-term processes of repair and decay. This enables digital twins to move beyond simple geometric models to become historically informed, materially accurate, and better able to predict performance and risk.
Another point that emerged was that this kind of complexity is precisely what makes heritage data so valuable in the context of artificial intelligence. Heritage datasets are, almost by definition, large, messy, heterogeneous, and complex. They are rarely standardised, often incomplete, exactly the kind of datasets that pose a real challenge for AI, and that makes them incredibly powerful training environments. Training AI models on difficult, real-world datasets, rather than clean, idealised ones, can lead to more robust and transferable systems. Heritage data forces models to deal with ambiguity, uncertainty, and variability. It requires algorithms to reconcile conflicting signals, handle missing information, and recognise patterns across widely different scales and formats. AI models trained to handle heritage datasets are therefore better equipped to support infrastructure monitoring, defect detection, and predictive maintenance.
A crucial point that underpins all of this is the continued importance of expert knowledge in the loop. Training AI on such data without domain expertise risks producing models that are technically sophisticated but fundamentally disconnected from material reality. Archaeologists, heritage scientists, and engineers working together, provide the essential ground truth, identifying uncertainty, contextualising data, and validating model outputs against real-world evidence.
This makes it clear that heritage science (and programmes like RICHeS, which I am convinced is going to have a significant legacy in terms of data as much as research) are central to the Industrial Strategy, providing the complex data, critical expertise, and interdisciplinary frameworks needed to drive innovation, reduce costs, and support the long-term sustainability of the UK’s infrastructure systems. Infrastructure and heritage are more often than not, one and the same.
Comments
Post a Comment