The use of artificial intelligence (AI) has been catapulted into the mainstream, and its application in the building services industry is on the rise.
There is a vision of the future where AI is seamlessly integrated into every aspect of a building’s function, and consultants are already exploring how it can be used to enhance the design, management and operation of building systems.
However, the rise of AI brings its own set of challenges and complexities, particularly around data security, system integration, regulatory compliance, and reliability and accuracy. In response, regulatory frameworks are being developed to ensure integration is safe and ethical.
While some countries, such as the US and UK, are taking a sector-specific approach to regulation, the EU introduced the Artificial Intelligence Act in June 2023, which addresses critical issues such as data protection, data usage and transparency.
Artificial intelligence glossary
Artificial intelligence: the simulation of human intelligence in machines that are programmed to think, learn, and make decisions
Machine learning (ML): a subset of AI that involves algorithms learning from, and making predictions or decisions based on data
Neural networks: a series of algorithms that mimic the operations of a human brain to recognise patterns and solve problems
Deep learning: a subset of ML involving neural networks with many layers, enabling the analysis of complex patterns in data
Generative design: an iterative design process that uses algorithms to generate a wide range of design solutions based on set parameters.
Cementing AI into building services
AI is already applied in many areas relating to building services such as: energy management and optimisation; predictive maintenance; building automation; occupant comfort; design and planning; security and surveillance; and emergency response and safety.
Arup global automation leader, director and Fellow Michael Beaven, says Arup has been using AI and machine learning (ML) for many years, including developing a tool telling engineers how significant cracks are in tunnel walls.
Beaven says the use of AI has now become endemic, because the interface with the technology is now easier for everyone to understand. ‘Previously,
the barrier to entry was high, but now people can use ChatGPT to write code,’ he says. Arup is leveraging ML to expedite problem solving, says Beaven. In a major commercial project in London, he says replacing traditional energy analysis models with regression-based AI slashed computation time. ‘What would have taken 18 months on a laptop, or four weeks in the cloud, takes just seven seconds,’ he says.
Foster + Partners began exploring ML in 2018 to understand how the integration of neural networks can help predictive models generate solutions easily. Martha Tsigkari, senior partner, and head of applied research and development, says it was able to predict deformation of passive materials under thermal conditions.
By analysing laminate layering, Foster + Partners reverse-engineered a process to train models to predict deformation patterns. This generative design approach, using distributed computing to run thousands of calculations, allows scalable applications such as designing a façade that deforms to create shading when its heated.
May Winfield, global director of commercial, legal and digital risk at Buro Happold, says AI allows the analysis of huge amounts of data generated by smart buildings. ‘We can feed data into an AI model and ask it questions. Some of the options could be ridiculous, but it will spark new ideas.’
A key benefit of AI, she says, is automating boring and repetitive tasks, allowing designers to add value elsewhere: ‘It allows our engineers to do the amazing creative work they do best.’
Several companies have launched their own in-house large language models (AI-based programs such as ChatGPT). Tsigkari says such a program called ‘Ask Foster and Partners’, allows engineers to access the company’s large archive using simple text questions.
‘One of the challenges of AI is that lots of good, organised data is required to train a system to predict things well,’ says Gavin Bonner, head of data and digital at Cundall. ‘AI is more specific than automation. It needs a data source that it reads and learns from, and then applies new context or new content based on that material,’ he explains.
While AI provides many great opportunities, there is the potential for unknown risk
Bonner says Cundall has built a cloud Lakehouse environment, which combines the benefits of large repositories of raw data with organised sets of structured data. The aim is to set the company up for effective ML across several disciplines.
While the computational power of AI promises to revolutionise the way buildings are designed, it comes at an environmental cost. The power needed to sustain AI is vast and growing; the International Energy Agency has estimated that electricity consumption associated with data centres, AI and cryptocurrency will grow from 2% of global energy use in 2022 to 4% by 2026.
Winfield says data centres need to become more efficient to handle the surge in demand that AI brings.
Cost is another barrier to adoption, says Bonner, pointing out that the deployment of AI systems requires substantial expenditure in data collection, storage, and analysis.
Winfield is concerned by the trajectory of AI if it is left unchecked. ‘With anything new and shiny, people tend to run at it head-first. But while AI provides many opportunities, there is the potential for unknown risk,’ she says.
Winfield believes large language models such as ChatGPT have scraped huge amounts of data from the internet, so there is the potential inadvertently to plagiarise designs or ideas. ‘There is a huge issue around copyright that companies must navigate,’ she says.
Confidentiality is another issue. Many building services firms work on unique, confidential projects. Feeding sensitive data into public AI systems could lead to breaches of confidentiality, with proprietary designs or models inadvertently exposed to competitors.
There are also safety elements to consider, says Bonner. ‘If you’re using an AI system to optimise an MEP design that’s related to safety, fire, structural design, it opens you up for a lot of scrutiny and you must be very transparent in the way that you are developing AI applications,’ he says.
‘If you ask the AI chatbot a question about your building, are you going to trust the answer? What if something goes wrong and there’s a massive leak in the building; whose fault is that? It’s an issue people are currently wrestling with.’
Beaven stresses that engineers are responsible for the AI output. ‘No matter where it comes from, it must always be checked,‘ he says.
Bonner says the EU AI act will help minimise risks. ‘There’s a lot you must comply with to ensure systems meet the requirements of the law,’ he says.
The Act requires that AI systems deemed high risk, such as critical infrastructure, meet multiple requirements and undergo a conformity assessment. Bonner says Cundall is ensuring AI processes comply with the act, which could be adopted elsewhere.
The rise of AI necessitates a new skill set, which firms must address. However, Bonner says the accessibility of AI through ‘low-code and no-code platforms’ like ChatGPT will enable a broader range of professionals to use AI.
The future
The industry experts don’t expect AI to replace building services engineers; the human element remains indispensable. Winfield says that AI’s strength lies in its ability to repeat patterns, but it cannot understand concepts or capture the nuances of human creativity.
‘Our AI strategy board analyses where the technology can benefit our business. At the moment ,we’re looking at quick wins that can allow our engineers to do what they do best – which is the thinking and creating amazing work,’ she says.
Tsigkari says AI tools should become facilitators rather than replacements of creative processes. ‘As with all disruptive technologies, AI can help us rethink everything that we have been taking for granted and effectively innovate in a new way.’
Beaven says engineers have to learn to drive the machine. ‘We need to be clear on what measures we give AI and who says it’s of real benefit for people. We need to interpret that ourselves. It can’t be the machine,’ he says.