AI will only be ubiquitous in infrastructure when ‘explainability’ problem is solved

Building regulations all around the world demand that those designing and constructing a buildings and infrastructure have to take responsibility for safety and integrity, as well as other factors such as its accessibility to people with disabilities, energy efficiency and the carbon footprint of the resources used.

If AI is used at any stage of design or construction then the company using that technology has to take responsibility for the output, which means they have to understand how it has been arrived at. If they are using other people’s AI technology products and don’t understand, or can’t access information on, how it works, then they are putting their trust in a system that is just a ‘black box’ to them.

That means they either take responsibility for software and processes they don’t understand and so absorb huge amounts of risk on behalf of their AI supplier, or they don’t use it at all.

This issue isn’t restricted to infrastructure – financial services companies are grappling with the same issue, and with the suggestion that responsibility for explainability in the UK at least will fall to named senior managers under financial regulations there.

Explainability will be the focus of regulation– the UK’s six principles for safe use of AI includes a requirement that its use be transparent and explainable, for example. It will no doubt be a solvable problem, but it hasn’t been solved yet and is holding infrastructure firms back from what would otherwise be a fuller use of AI.

So what uses are actually being made of AI right now in the infrastructure and real estate sector? There aren’t as many as our current position in the AI hype cycle might suggest, but it is being used.

One thing it is very useful for is embedding building standards into project documentation. Building standards are a crucial way to ensure that structures are safe and compliant, but they are incredibly complex and subject to frequent change.

Managing this on paper is laborious and tricky and mistakes are likely – a single standard can run to several hundreds of pages and cross refer to other standards and multiple guidance documents which can also be complex. As an example, the UK Building Regulations relate to some 800 different standards. These standards relate to designers, contractors and individual trades contractors who will each be responsible for safety and compliance So, systems currently in use which embed these in project documentation; understand to some extent the content of them, and cross reference between one standard and another to flag issues or risks makes projects safer, more compliant and more efficient.

Project contracts are also dense, long and complicated, and AI is being used right now by designers and contractors to analyse contracts and highlight where a company faces risks. These could be commercial, in that the company is exposed to outsized penalties or delays; they could be technical, in that a project proposes an action which breaches building standards, or they could be legal, in that a company is exposed to more than its fair share of legal liability.

AI is also being used in the assessment of these risks. Predictive analysis uses huge datasets describing lots of past projects and identifies patterns to find out the areas where certain types of projects commonly go wrong. This can give companies early warning of delays, problems or vulnerabilities that they can use to avoid problems. This also applies to delay analysis and using large project datasets to understand causes of delays.

Here we need some clarity – we don’t consider Building Information Modelling (BIM) to be AI. This is a well-established and well-used technology that predates generative AI. But BIM systems can interact with AI in useful ways.

Video cameras and the software and hardware needed to run them are cheap, so the recording of 360 degree views of construction sites is feasible. What AI can do is use image analysis and track site progress, then feed that data into a BIM model or digital twin of the building. The BIM model or digital twin becomes effectively self-updating via image capture, which saves time and money and ensures that the BIM model or digital twin becomes an accurate ‘as built’ model.. This makes asset management and maintenance quicker, cheaper and more comprehensive.

There is no doubt that AI has the potential to have wide ranging effects in the infrastructure and real estate sector, where companies with AI and those without operate in different parts of the market. It is clear that the currently siloed and more traditional parts of the sector will find it more challenging to implement AI solutions. It may even drive consolidation in the sector.

But its use is so far a long way from being ubiquitous.

Take part in the 2025 global ‘Artificial Intelligence – Impact on the Built Environment’ survey, run by international law firm Pinsent Masons with Bentley Systems, Mott MacDonald, and Turner & Townsend. It focuses on the impact of AI on the built environment and, in addition to AI use cases, examines issues of AI governance and risks.