This feature is a part of “The Dotted Line” series, which takes an in-depth look at the complex legal landscape of the construction industry. To view the entire series, click here.
Two and a half years ago, Construction Dive asked ChatGPT, OpenAI’s large language model, to generate a construction contract on a design-bid-build job for a 600-unit, mixed-use project in San Jose, California. The document, attorneys said, could be legally enforceable — it addressed scope of work, payment terms, a termination clause and sections on indemnification, insurance and change orders.
But it also came with missing clauses and poor risk management. Those same attorneys cautioned that using artificial intelligence to solely generate a contract was akin to opening Pandora’s Box — the risk was simply not worth it.
Construction attorneys now say the tech is here to stay and that its capabilities have improved as it has worked its way into myriad aspects of daily life. But after two years of improvement, can it now write an ironclad construction contract? More to the point, should construction attorneys turn to it to do so?
“Oh my God, no. I would be terrified if I heard of anybody using it to generate a contract,” said Megan Shapiro, a construction attorney and partner at law firm Radoslovich Shapiro in Sacramento, California.
While AI does present uses for construction attorneys, experts cautioned that using it as a wholesale, one-stop-shop for contracts can only lead to trouble.
“It's not something where you press a couple buttons and you have a contract that's ready to go,” said Michael Vardaro, managing partner of New York City-based law firm Zetlin & De Chiara.
The hallucination issue
Shapiro didn’t mince words when it came to AI tools’ accuracy. She said it was “shocking” how wrong AI-generated content could be and how often that occurred.
Indeed, so-called hallucinations have been a massive problem for large language models. While that’s been the case ever since generative AI made its mainstream debut in 2022, a bug developers pledged would become less prevalent with use and the tech’s evolution, The New York Times reported in May that for newer AI systems, the hallucination rates reached as high as 79%, depending on the test and tool.
For attorneys, this presents a significant business and ethics problem as they look to balance speed with accuracy. Case in point: the advent of LLMs has given rise to “AI slop” in the legal system, or filings with bogus or misrepresented court cases, the Times reported on Nov. 7. Currently, a database that tracks AI hallucination cases in legal systems across the world has included, at the time of publication, about 600 cases — the majority of them in the U.S.
Due diligence
These issues present reputational and business risks, said Trent Cotney, a partner at law firm Adams & Reese and the head of its construction team. Should an attorney file a brief with AI-fabricated documents or claims, the presiding judge could consider sanctions.
“Not to mention the fact that there's a reputational hit, and your client obviously is not going to be happy with that either,” Cotney said.
Shapiro said that what keeps her up at night is the thought that her clients could be using AI on their own without discussing it with her. She fears that if someone at a construction firm doesn’t want to pay to call an attorney on that legal issue, then that person may attempt to figure the problem out on their own.
“I think the risk to them is getting substantially worse, because I think consumer confidence in the output is increasing, despite the fact that I think the quality of the output is decreasing,” Shapiro said.
Finding an edge
With diligence and a careful eye, however, attorneys said there are ways AI can help legal professionals with their daily tasks.
Vardaro, for example, said one of the benefits of AI was its ability to corral documents and seek out sections contained within them, a job that was left to associates years ago.
“Now, AI can kind of do that and give you kind of the hits in that project file within minutes, where that might have taken weeks, if not months, to do,” Vardaro said.
Experts also called out its ability to review contracts for specific provisions. Cotney, who refers to himself as an early adopter of the tech, will use it as a gut check, particularly when it comes to verifying references and citations.
“I do use AI for purposes of cross checking what I've done, identifying if there's something else to think of that I haven't thought of, or helping me fine tune certain things or compile certain things,” Cotney said.
Shapiro will sometimes use AI in her work as a research tool, but that’s where it stops for her. It will help her find and identify resources and paths to explore that she wouldn’t get in other ways.
“Thinking of it as a law clerk or a low level associate, a young attorney and reviewing it that way, I certainly think that there's value there,” Shapiro said.
At the same time, attorneys said AI tools need to be prompted by people who already know the law, in order to flag if something that it produces is suspect.
“It's not a completely useless tool,” Shapiro said. “It's just a tool that needs to be utilized by a user who has the knowledge and experience to leverage it in a safe, effective, efficient way.”
