How GC giant EllisDon is grappling with rapidly advancing AI 

When ChatGPT, an artificial intelligence language model, was released in 2022 it sent shockwaves around the world and the construction industry. Everybody wanted to see what it could do for their business.

As an early adopter of AI, EllisDon’s team was eager to try it out but it was quickly banned. 

“It was something we shut down almost immediately,” said Patrick Lalonde, Senior Director, Digital Project Delivery, EllisDon. “Very quickly people were starting to upload contracts and sensitive documents into that open environment.”

Instead, they partnered with Microsoft Copilot to create a private domain with those same large language models built into it. Essentially, it’s a safe environment for people to use these large language models without compromising sensitive information. 

It’s just one example of how Canada’s construction sector is learning to integrate AI into their operations as the technology rapidly evolves. 

Beginning with BIM

For Lalonde, who’s been with EllisDon for the past 15 years, the transformation has been dramatic. When he joined the company, the virtual design and construction team was something new.

“We were just trying to figure out how to use BIM, what it was and how we could apply it to construction,” he said. 

It started with 3D models for designers but it soon became clear that these models were valuable to the construction team as well. They could be used for logistics, scheduling and more. As the approach snowballed, it bled into the broader construction technology ecosystem. It became a catalyst for much of what came next. 

Lalonde explained that everything relied on the model: laser scanning, reality capture, drone flying and more. 

“It all connected back into the model or using the model philosophies, like common data environments or central ways of working,” he said. 

The model and its affiliated data has become a product of its own, handed over to a project owner after the physical asset is completed. It’s a replica of the structure that can be used over its lifetime. 

AI is born

EllisDon was introduced to AI with the rest of the world and quickly had to figure out how to use it and what their rules were.

In addition to using large language models in a closed, secure environment, early uses included reality capture and visual intelligence. Sites already had cameras documented sites, but those were then plugged into computer vision which began to automatically recognize objects, which helped automate progress tracking. 

“Now that has expanded to getting our data infrastructure in place and more preparing for larger advancements in AI,” said Lalonde, who noted that this includes developing language models that can interact with EllisDon’s own internal data. 

This includes establishing an AI governance policy and implementing data classification to determine what info is sensitive, what’s private and what can be made public.

EllisDon is strict about what AI tools it lets into its ecosystem. They are piloted by a small team and validated before being introduced into the broader organization. 

“This doesn’t mean we aren’t receptive to AI tools, we are just very calculated about which ones we do bring in,” said Lalonde.

If a good tool doesn’t exist, EllisDon has systems in place to build it in-house. They have a partnership with Palantir which powers their internal AI capabilities.

“It’s kind of a one-stop-shop for a lot of stuff,” said Lalonde. “It gives us a lot of flexibility to build our own custom AI applications.” 

Cautious collaboration

The evolution of AI in construction can only go so far before it hits a data wall. Large language models have been able to feast on mountains of publicly available data, but data specific to the construction process is being cautiously guarded by builders who worry sharing it might give away their competitive advantage. 

“I don’t think everyone fully understands what competitive advantage this data or AI has yet, so I think there is a natural tendency to assume everything has a competitive advantage to keep cards close to the chest,” said Lalonde.  

He explained that good data on RFIs, schedules and the nuances of construction could help train AI to improve the entire industry, but a lot of it is required. Builders have been meeting to share their ideas on AI and figure out what data they’d be willing to share. 

“The safety world has kind of done this already, for the greater good of the industry, to make everyone safer, to share safety data across these companies,” said Lalonde. “That hasn’t happened yet for all the other data that AI might need to be trained on.”

Risking overreliance

Beyond the nuts and bolts of implementing AI, EllisDon believes one of the biggest risks is foundational: Overreliance on the technology could impact knowledge transfer and succession planning.

“Reliance on AI is something we need to be careful with,” said Lalonde. “If we are starting to rely on artificial intelligence to make all the decisions for us, without critical thinking or human input, we could find ourselves in serious trouble.” 

He explained that the company has long-term employees who have risen their way up the ranks after starting in the field. They have hands-on experience to pass on. But if younger workers simply solve problems using AI tools, they aren’t developing those skills. 

“If they have these AI tools in their hands and don’t have the opportunity to make those critical decisions and get that experience, then they won’t have it going forward,” he said. “We will lose all from the older generation.” 

AI should be used for boring, repetitive tasks that free up workers to focus on bigger things. Lalonde believes that despite the power of AI tools, EllisDon wants its employees to still maintain those critical thinking skills. 

“If everything is shopped out to AI, nobody is learning anything and you won’t have any critical thinkers left. That would be a very big problem for the industry.”