The opportunities and challenges of AI in construction

While AI offers huge potential for error reduction and productivity improvement in construction, it should be used in concert with clear policies and strategies, human oversight and assurance procedures to reduce the risk of AI-introduced error, John Priestland told GIRI’s September members’ meeting.

GIRI’s technology report on AI and error reduction in construction was published in July, and it revealed that while the industry’s use of AI is relatively limited currently, often as a result of concerns about the reliability of AI outputs, it is expected to have a transformative impact on the sector within the next five years. 

“The glide path of AI and its capabilities is almost vertical,” said John Priestland, chair of GIRI’s Technology Working Group and the author of the report. “The models are going off the scale with these technologies and techniques, multi-context protocols and AI agents, and it is hard to imagine where AI will be a year from now. So it is not wholly hyperbole for the former chairman of Google to call AI the most important development in the last 500 to 1,000 years. What’s important is the ethical and appropriate adoption of AI.”

In construction, said John, AI has huge potential, and the report identifies seven key areas where AI could make a difference – if it can be assured that outputs are correct and compliant with standards. These include generative AI for design and documentation; verification tools; predictive AI for real-time risk management; AI agents to streamline back-office workflows; AI agents to enable parametric 5D BIM; to enable a systems engineering approach; and integration with robotics for physical delivery.

AI for error reduction

To analyse the potential of AI for error reduction, John explains: “We went back to the iconic 2016 GIRI research report and the 17 identified root causes of error. Thanks to a mapping exercise by an HS2 team, which took those 17 root causes and grouped them into nine categories, we were able to consider the potential for AI to make a difference for each of these nine categories of error.”

The nine categories are: 

1. poor planning and design; 
2. lack of skills and training; 
3. poor communication; 
4. materials issues; 
5. environmental or site conditions; 
6. inadequate project management; 
7. quality control and compliance; 
8. technological or equipment failure; and 
9. human error.

“To date, none of these categories has seen very much progress on AI, but the four that are the most interesting are poor planning and design, inadequate project management, quality control and compliance, and human error. So the logic is that if emphasis is put on those four types of error and we encourage progress in these areas, we will end up with greater impact.”

Within these categories of greatest potential, the report identifies five key ways in which AI could reduce error: through insight and prediction; automation and assurance; monitoring and verification; communication and collaboration; and decision-making and control. “However, we will only be comfortable using AI in construction if we can be confident that it won’t make mistakes.”

AI errors, or ‘hallucinations’, happen frequently, including citing legal cases or scientific references that don’t exist. “This is a big problem, and it is reasonable for people in construction to be wary of using AI to generate, for example, environmental compliance reports, in case AI gets something wrong. This is why GIRI has focused not only on how AI can reduce errors in construction but also on how we can work to help reduce errors in AI to make it suitable for use within high-hazard industries such as construction.”

Reducing AI errors

The first step, he explained, is to carry out a risk assessment before using AI. “So, for example, you wouldn’t use AI in safety critical processes, but you could introduce it for lower-hazard processes initially and make sure that assurance is built into your quality plan or risk assessment.”

The second step is utilising chain-of-thought reasoning. This is where an AI system documents the steps taken to produce the requested output, which can be reviewed to identify any potential issues. “Chain-of-thought reasoning opens up the AI black box.”

The third step is ensuring that there is always a human in the loop. “This means that your workflows should always have a human sitting at the top of the process who can identify errors so that the AI assists rather than replaces human judgement.”

The fourth step is to use assurative AI, which employs rules-based checks or knowledge graphs to verify AI outputs against standards. “In this way, we can use AI to not only check itself, but also to check other outputs we may have produced. This opens up the potential for generative AI use in a safer way in construction.”

AI applications for organisations

John closed his presentation by highlighting the ways in which different types of organisations should be focusing their use of AI. 

“For contractors and others working in construction, focus your AI investment on high-impact error sources. Carry out risk assessments before using AI tools, embed assurance into your workflows, and establish internal policies and safeguards to prevent unauthorised and inappropriate uses of AI.”

Software providers are encouraged to develop AI systems that see error reduction as an important use case and develop verification tools, assure outputs from generative AI, and integrate AI into lessons learned and quality management systems. 

Finally, said John, GIRI will promote AI for error reduction across the sector. “We are going to update last year’s best-practice casebook to include AI case studies, and we will be an advocate for safer and assured uses of AI in construction.”

Because the AI revolution is coming. “AI will not solve construction problems alone but used wisely it could become a powerful force for good and reduce error and improve outcomes across the industry. If you’re not using it now, you will be in five years.”