When John Andres, director of technology for ANDRES Construction Services, began to experiment with AI services like ChatGPT, he was skeptical of the program’s accuracy and trustworthiness.
“One of the first things I did was throw our standard operating procedures (and) company handbook (into the program),” Andres said. The company has a no tobacco policy listed in the handbook, so Andres tested the AI by asking if employees could smoke at their desks.
AI’s answer? Yes, employees can smoke at their desks but should remember to “be polite about it.”
“It didn’t have the context that tobacco is smoking,” Andres explained, “(which eroded) trust right away.”
Andres, a user of AI-powered document intelligence and compliance platform Document Crunch, shared this story during a breakout session sponsored by the platform at BuiltWorlds’ recent AI/ML Conference. The sold-out event marked a sharp increase in attendance from the 2024 AI/ML Conference, highlighting a growing interest in AI across the industry, as over 130 industry leaders and innovators gathered to dig into the complexities and opportunities of AI and machine learning in construction.
Speakers ranged the gamut from developers to end users, while presentations and panels covered a similarly wide array of topics. However, despite the myriad perspectives shared during the event, one point of agreement made clear throughout the conference was that trust in AI is a concern among contractors.
Giving AI Construction Context
Construction is an industry rife with context. According to Document Crunch CTO Trent Miskelly, AI providers in the industry, Document Crunch included, will need to train their solutions to know that they are responding to users within the construction industry.
“We know that (construction) very specifically relies on a language of the industry and we need to make sure that our solutions are catering to that,” Miskelly said.
Andres pointed out that the word “submittal” is not always recognized by word processors as a “real” word, making it difficult for AI solutions not specifically trained for the construction industry to understand. Additionally, many commonly used acronyms, such as “LD,” short for liquidated damages, hold very specific meanings for industry professionals. If the AI feature is pulling from non-construction specific data or isn’t trained or properly prompted to apply a construction perspective to questions being asked, the answers given can get muddled.
ChatGPT, for example, pulls from information publicly available on the internet. When we asked what LD stands for, ChatGPT answered that the acronym could mean learning disability, long distance, logical design, lethal dose, load or low definition. However, those in construction will tell you that, in the industry, LD stands for liquidated damages.
Kelsey Gauger, vice president of operational excellence for Suffolk Technologies and a 2025 AI/ML Conference speaker, echoed the importance of trust in AI during another session that highlighted a partnership between Suffolk and Trunk Tools. Following a successful pilot program, Suffolk recently announced it would be deploying the AI-powered document and data management platform across all projects.
Prioritizing Early Accuracy
When deciding which AI solution to pilot, the “number one thing (Suffolk) looked at was accuracy,” Gauger said, emphasizing the importance of getting things right from the get-go.
“When we go out to the jobsites, you really get one shot,” Gauger added. “If you roll out something and the first impression is not good, the answers aren’t right, (then the workers are) not going to come back to you.”
However, in order to develop AI tools that are better-suited for construction, continued usage and good data are key to training AI solutions. Miskelly encouraged conference attendees to use AI as a brainstorming tool. While AI will not always be right at this stage, he believes that construction professionals are “discerning” and will catch the errors.
“The more that you’re doing that type of work inside of these solutions, the more those solutions will learn from the way that you operate,” Miskelly said, “and it will become part of the DNA and fabric of your business.”
Finding a Way to Trust AI
Gathering data from across the industry for AI to reference and train on is tricky when construction companies are worried about proprietary information being shared with outside players. Even collaborators on a single project can be reluctant to share data believing the information gives them an edge, admitted Rajitha Chaparala, vice president of data and AI at Procore, who spoke on a panel focused on data and AI implementation. AI solutions require company-specific data to operate effectively, but AI companies must include safeguards to build trust with users and clients who are concerned with risk.
When asked about situations where competing companies both use Document Crunch, Miskelly was quick to reassure conference attendees that the solution does not share customer data between clients. “It’s not like we’re connecting accounts,” Andres added. “It’s still very much this is our side, this is our tool.”
If the recent AI/ML Conference is any indication, there’s a growing interest in exploring AI’s potential to augment and enhance productivity and efficiency across construction. Meanwhile, questions linger about how effective these solutions will be and how to incorporate them in a way that mitigates risk. “We have a lot of hope and we have very little trust,” Miskelly remarked. “I think this is the biggest problem we need to be solving in the near-term so that we can get to a point (where) people can rely on these solutions.”
Discussion
Be the first to leave a comment.
You must be a member of the BuiltWorlds community to join the discussion.