How to Be Smart About Using AI in Construction
Caution signals from construction attorneys
Among the hazards to using AI without verifying its accuracy is that a general contractor or owner picks the wrong winner in a bid competition.
Photo: BestForBest/Getty Images
Peyton Aldrich
To get up-to-date on risks and possible downsides of using artificial intelligence in construction, ENR turned to two experenced construction lawyers who are are familiar with issues related to AI: Christopher Horton, a partner at Smith Currie Oles LLP and Peyton Aldrich, an associate at the firm. They spoke with correspondent Elaine Silver about possible inaccuracies, how to sharpen queries and what they call a 'defensible trail.'
What is the most important thing construction companies need to know about using AI?
Horton: Treat AI like a scalpel and use it strategically for specific inefficiencies, not as a catch-all solution. Take the time to conduct due diligence before adopting it to understand the tool’s security, capabilities, and limitations.
Aldrich: Don’t forget that AI is a tool, not a replacement for human oversight. Without proper checks, you risk relying on inaccurate outputs that could lead to costly mistakes.
There are plenty of strengths and advantages, right?
Horton: AI shines in streamlining project planning, such as in crafting bid package narratives or reviewing proposals to spot improvements. Our clients are also using it for notetaking in meetings, although security must be ensured. It’s important to vet tools for security and implement them with clear guidelines to maximize efficiency.
Can you further explain that?
Aldrich: Companies should train staff to use AI critically and set policies to avoid ad hoc use. Define tasks to automate, evaluate tools carefully and provide training on inputting data and spotting errors. Here’s a personal example: I use the Westlaw AI feature a lot. So, when I'm starting a research project, if it's not an issue that I have a lot of knowledge on what I will do, I'll actually take the prompts that Westlaw gives. That shows us, hey, this is how our AI works. This is how to ask the best kind of question, and I will copy and paste that into ChatGPT, and tsay, this is the issue that I'm trying to look up.
I use one AI to help generate a question to get the most accurate response from another.
What are some of the biggest pitfalls in AI use?
Horton: Understanding its security features and whether what you're using is closed loop or open loop. It’s very important not to put any information into AI that would include any confidential proprietary attorney, client or work product or privileged information.
Christopher Horton
Aldrich: AI also can generate biased or inaccurate outputs, misinterpret contract clauses and cause "hallucinations"—where AI produces false information such as made-up cases or regulations. That could lead to picking the wrong bid winner due to flawed data, creating new liabilities. It’s a real risk if you don’t verify the results carefully.
Why is the construction industry particularly vulnerable?
Looking for quick answers on construction and engineering topics?
Try Ask ENR, our new smart AI search tool.
Ask ENR →
Aldrich: Construction involves complex contracts and high-stakes decisions, so AI errors, such as misinterpreting a contract provision, can lead to missed deadlines or financial losses. Construction is vulnerable because it deals with lengthy, custom contracts and data-heavy processes such as risk assessments and subcontractor evaluations. AI’s tendency to overlook nuanced contract terms or produce biased outputs from incomplete data can lead to serious consequences, including compliance issues or unfair decisions.
Horton: The industry’s fast-paced nature tempts firms to rely on AI without verification, especially for bids or schedules. But unverified outputs can violate contractual obligations or expose proprietary data, particularly if AI tools aren’t secure.
So to train AI most effectively...?
Horton: Training involves learning the tool’s limits and security features. It’s not just about using AI but knowing when to trust it and when to verify, which comes from structured guidance and policy-driven implementation.
What is a defensible trail that you mentioned to me?
Aldrich: A defensible trail documents how AI was used, including inputs, outputs and human oversight. This record strengthens your position in litigation or regulatory reviews, proving that consistent and proper procedures were followed.
That's important because ...?
Horton: Without a defensible trail, you can’t justify AI-driven decisions in court or arbitration, especially if privileged information is compromised. Opaque AI algorithms also make it hard to defend decisions in disputes that could lead to increased legal and financial exposure.
What AI policies do you have in your law firm?
Horton: Our firm has a strict AI policy I developed as AI committee chair. We only use approved tools such as Westlaw, ChatGPT, and Microsoft Copilot, ensuring they’re secure and closed loop to protect confidential data. No proprietary or privileged information goes into unverified tools and we’re also testing generative AI like Thomson Reuters’ CoCounsel for future use.
What is the one thing every construction executive needs to do to use AI programs well?
Horton: Verify, verify, verify.
