By Tara Hoke, M.ASCE
While generative artificial intelligence offers great potential to enhance the practice of engineering, professionals must always balance this promise against the risk of overreliance. Drawn from a recent court case, this column presents an example of the ethical and professional perils of placing too much trust in AI-generated output.
Situation
Litigation ensues between a general contractor and its electrical subcontractor over delays in the construction of a midscale hotel in central Texas. The contractor claims that the subcontractor’s failure to provide sufficient labor and meet project deadlines was responsible for the project delay, causing the contractor to incur hundreds of thousands of dollars in additional labor and equipment costs, contractual penalties, and other damages. The subcontractor assigns fault to the general contractor, asserting that the contractor had provided it with deficient design documents and then failed to approve an equitable increase in its fees.
The general contractor retains an engineer to perform a forensic schedule analysis of the project, hoping to establish the precise impact of the subcontractor’s delay on the overall timeline. Unfortunately, though the contractor had initially used critical path method software to manage the project schedule, it had failed to make regular updates as the project lagged further behind — meaning that CPM scheduling data did not exist for the period of time most relevant to the litigation.
The expert witness attempts to remedy this shortcoming by using other contemporaneous project documentation. He prepares a 60-page report attributing responsibility for almost the entirety of the project’s delay to the electrical subcontractor. The engineer’s report includes copies of project records that he relied on in reaching his findings, including payment submissions, inspection reports, and look-ahead schedules.
The report claims that the engineer’s method of supplementing the CPM schedules is “endorsed by numerous independent industry guidelines,” and he supports this contention with references from recommended practice manuals, textbooks, and nearly a dozen peer-reviewed articles and manuscripts.
Upon review of the expert’s report, counsel for the electrical subcontractor notices some surprising factual discrepancies. In at least three places, the engineer’s report points to language in attached project records that purportedly describe work as “pending” or “on hold,” but the records themselves contain no such notations.
Further review uncovers additional inaccuracies in the report. The report states that Section 4.7 of a technical guideline document supports the engineer’s approach to supplementing the missing CPM records, but the actual guideline does not have a Section 4.7. The report quotes a passage from a construction management textbook, but the excerpt contains language that does not appear in the cited work. Most strikingly, the report concludes with a list of references the author claims to have used as the basis for his methodology; however, two of those references are unrelated to construction schedule analysis, and two others — both purportedly ASCE journal papers — do not exist either in the cited journal volume, nor anywhere else in the ASCE publication library.
At deposition, the general contractor’s expert witness initially claims that he personally read and verified the citations used in his work. When pressed to explain the inaccuracies, however, the engineer states that he maintains a personal database of source materials, and he attributes the issue to copy-and-paste errors made in preparing his report. He admits to using a generative AI tool for writing assistance, but he flatly denies that either the report’s substantive content or its reference list was composed by this tool.
Notwithstanding this denial, the subcontractor’s attorney files a motion with the trial court to exclude the expert witness’s report and testimony. He claims the discrepancies in the engineer’s report “bear all the hallmarks of AI hallucinations,” and he asserts that the report’s use of fictional references and quotes was an attempt to mislead the court into believing that the engineer’s method was reliable, and that his opinions were supported by project records.
The opposing counsel argues that the report’s contents reflect extensive work and research by a qualified and competent engineering professional, and that the discrepancies were “incidental errors” that were unrelated to the substance of the report and did not affect the merit of its underlying analysis.
Ultimately, the court sides with the subcontractor’s attorney. Noting that the engineer had cited, claimed to rely on, and falsely attested to verifying the nonexistent references in his work, the court felt the engineer had “shattered his credibility” with the court and fallen “far short of the intellectual rigor expected of an expert witness.”
The expert witness and his report are excluded, and the court expresses willingness to consider an award of attorney’s fees and costs to the subcontractor for time spent disputing the expert report.
Question
If this case were considered by the Committee on Professional Conduct, how would the committee view the engineer’s actions under the ASCE Code of Ethics?
Discussion
Section 1h of the ASCE Code of Ethics instructs engineers to “consider the capabilities, limitations, and implications of current and emerging technologies when part of their work.” This ethical mandate reflects a constant tension in engineering practice: Engineers who avoid new technologies risk becoming obsolete, but those who adopt new technologies without fully understanding them may make costly, or even deadly, mistakes.
With regard to generative AI, such tools offer significant time and cost savings to engineering professionals, with their ability to swiftly produce design options and renderings, create checklists and workflows, or draft reports and other documentation. But because generative AI is not fundamentally a search engine, but rather a technology that creates content informed by a dataset, AI tools can unfortunately sometimes produce creative fictions — visuals that disregard physical limitations or realities, calculations that fail to recognize bad or outdated data, and answers that promote plausible falsehoods over literal truths. Mindful of the dictates of ASCE’s code, an engineer seeking to take advantage of AI’s efficiencies must also take care to remediate its risks, by ensuring that the tool’s sources are reliable and verifying its output with a careful personal review.
In the present case, if the CPC believed the flawed report was the result of AI hallucinations, the committee would likely feel the engineer had not met his ethical obligations to “consider the capabilities, limitations, and implications” of this technology. It would also likely be troubled by the engineer’s lack of candor in claiming that he had read all the cited references and that they supported his methodology; as such, the CPC might find that this falsehood violated the engineer’s ethical obligation under Section 1c to “express professional opinions truthfully and only when founded on adequate knowledge and honest conviction.”
Moreover, whether the falsehoods in the expert witness report were introduced by AI or through some other method, the CPC would likely believe that the engineer’s failure to verify the report’s contents caused significant harm to his client’s interests. As such, the committee might also characterize his conduct as a violation of Section 4a’s mandate to act “with integrity and professionalism” on behalf of clients and of 4g’s directive to “approve, sign, or seal only work products” that the engineer had properly prepared or reviewed.
New tools and future developments in AI hold tremendous promise to improve the accuracy of AI-generated content and its utility for the engineering profession, but engineers must always remember that new technologies are no substitute for engineering expertise, nor is any tool an excuse for engineers to cut corners on a professional or ethical obligation.
As stated in ASCE policy statement 573: “AI remains a tool that has great potential to enhance civil engineering processes. However, AI cannot be held accountable, nor can it replace the training, experience, and judgment of a professional engineer ... .”
Tara Hoke, M.ASCE, is ASCE’s general counsel and a contributing editor to Civil Engineering.
This article first appeared in the November/December 2026 issue of Civil Engineering as “Trust but Verify.”
