Artificial intelligence is everywhere.
Across industries, leaders are experimenting with various AI technologies to see where they fit. In recent years, generative AI specifically has created plenty of buzz.
The AI wave has not skipped over civil engineers, many of whom are exploring where AI can take the industry. But not all AI tools are created equal, and applying them to real-world infrastructure projects safely and effectively is a careful balancing act.
AI offers unique benefits
AI is being applied to many different use cases, but where it really shines is when it comes to handling big data, said Norma Jean Mattei, Ph.D., P.E., F.SEI, Pres.17.ASCE, emeritus professor at the University of New Orleans.
The city of Memphis, Tennessee, found a place for this capability in its efforts to address potholes. Using cameras on buses to capture a digital feed of the roadway and identify potholes, the city trained an AI to differentiate high-quality road from damaged road.
“That’s a tool that you can use as part of your asset management work – find where the potholes are so you have a map,” said Mattei. “You may even be able to develop a tool which then autogenerates a schedule of repair.”
As long as you have plenty of reliable data, she continued, there are “so many ways in which these tools can be used to better do your work, no matter what it is.”
Michael Miller, P.E., P.Eng, F.ASCE, F.SEI, vice president of engineering services at Exo, also sees a future for AI in remote infrastructure inspection and sensing activities, such as dynamic line rating. This practice “is the next evolutionary step in grid operation to get more power of the existing lines without sacrificing reliability,” he said.
Using this technique, civil engineers can determine real-time wire temperatures, which informs the adjustment of power transfers for increased efficiency. With this data, AI could “help make the complex decisions needed to optimize the power flow across the grid.”
Although AI is useful for solving a variety of civil engineering problems, Mattei doesn’t see it replacing civil engineers. The verification of data, infrastructure design, and the actual construction process is still dependent on people.
“There will be engineers using that very powerful tool to do that work, but it’ll still need someone human who understands what’s going on,” she said.
Can the energy grid keep up?
AI is a great tool for data analysis, but the pressure it puts on the energy grid shows a different side of the AI boom.
Generative AI tools require huge amounts of energy to respond to prompts from users – up to 10 times as much as a typical online search.
Information technology companies are building more and more data centers across the country to power AI tools. But these structures are putting an “unprecedented demand” on the energy grid, said Miller.
“Over the past 30 years, the nationwide demand increases for electric connection requests have averaged at about 2% year over year,” he noted.
More recently, this figure has increased dramatically. In 2023 and 2024, much of the U.S. saw this demand rise from 10 or even 30% in many, said Miller. Experts forecast a continuation of this trend.
“Meeting this demand takes longer to permit and build than data centers, so a backlog of interconnection requests has developed in many regions,” he said. “The advent of cheaper renewable energy has also contributed to this bottleneck as many of the older, less-clean technologies such as coal generation are being phased out and replaced with interruptible renewable generation, which changes the stability needs of the grid as a system.”
Even data centers that use water cooling are “energy hogs,” said Mattei.
“Depending on where the data center is, you may be stressing your water supply, whether it's surface water or groundwater,” she said. “So, if you're in the arid west, the question is: Where is the water being sourced? Does it need to be potable quality? In these big centers, can you use wastewater from industrial processes or the municipality and get it to the quality they need?”
There is also the question of data center placement. If they build them further from communities, they will be further from power sources, and there will be a new demand for transmission lines or small nuclear plants so they can function.
Information: a digital source of fuel for AI tools
AI requires more than what we traditionally think of as energy – it needs accurate information to power successful, “low hallucinatory” results.
“The progress made by the AI tech industry in the past several years is truly game-changing,” said Eva Lerner-Lam NAE, Dist.M.ASCE, president of Palisades Consulting Group. “But the ‘high octane’ fuel for those machines – that is, the source data from ‘verified and validated’ knowledge bases – has not been organized and tagged by civil engineering domain experts in ways for AI tools to access easily.”
Agentic AI and reasoning models gather information in various forms from all over the internet to construct their responses to user prompts. Much of that information is translated by AI machines into “intelligence tokens” that have not been verified by experts in their respective subjects.
Trusting machine models to synthesize knowledge based on sources and modeled inferences that have not been thoroughly verified and validated risks producing results commonly referred to as “garbage in, garbage out,” where incorrect or irrelevant information not only shapes the response but further propagates incorrect “knowledge” for responses to future prompts.
Lerner-Lam wants civil engineers – and experts in every domain – to rethink the way AI tools gain access to source information and start packaging it in a way that can be “efficiently and effectively” used by these models.
Generative AI models gather information in the form of tokens, or bits and pieces of words, concepts, and formulas that are easily and robustly digestible for these tools. Lerner-Lam hopes to see civil engineering knowledge packaged into tokens so AI tools will not only pick up accurate information but will also compensate the creator of the tokens according to copyright licensing agreements.
“The only people we should trust to create the content for civil engineering intelligence tokens are our domain experts,” she said. “The Achilles’ heel in the AI ecosystem for civil engineering today is the near-total lack of verified and validated knowledge bases. Instead of engaging with domain experts, the AI world has built ‘scale-up’ models to take small data sets to synthesize larger data sets.”
She emphasized that without intelligence tokens that are verified and validated by civil engineers, any AI response would be likely to “hallucinate” and produce inaccurate responses to prompts.
AI may shine with large datasets, but a small piece of bad information in a single intelligence token can send a model and its users in the wrong direction.
Lerner-Lam shared an example of an AI company that built a medical AI model using a dataset containing 5,000 pieces of data from a participating hospital on a specific ailment, which produced AI-generated responses with an over 90% accuracy rate.
When the company added data from another hospital, the accuracy rate of the AI model plummeted to 54% because the original data from both sources was not defined in a consistent, verifiable, and validated manner.
Lerner-Lam pointed out that leaving the verification and validation of source data up to data scale-up models generated by AI tech programmers, without active participation by domain experts, puts knowledge “truth,” and therefore AI technology as a whole, at risk.
What’s next for AI in civil engineering?
Despite the flaws of AI, “the cat’s out of the bag, and you know it’s a tool that’s going to be used,” said Mattei.
Like with any technology, she continued, there are early adopters and those who are hesitant to embrace new tools.
Among civil engineers, Miller has seen “a reluctance and almost an intimidation factor” toward “technologies that are so all-encompassing” like AI.
“Civil engineering is a seasoned science, and I mean it in a good way,” he said. “Those of us practicing engineers who have spent decades learning and refining our design processes are encouraged to see new technologies being added, but only if those new technologies are transparent and not disruptive.”
As technology progresses, later adopters might “find themselves handicapped,” said Mattei. But there are risks with early adopters who may not “know what they don’t know” and put too much faith in the technology.
“The challenge with AI is to make useful technology readily available to aid engineers to become more efficient without providing more uncertainty or a black box of not knowing what’s behind the curtain,” said Miller.
Using these technologies with caution can help, as seeing them in action “encourages more use and acceptance of AI,” he added.
As client needs and engineering technology evolve, what matters most is that “we never lose that emphasis we now place on protecting and enhancing the public’s health, safety, and welfare,” Mattei said.
