Artificial Intelligence (AI) is taking the world by storm and it is impossible to not see any AI related content. With investment forecasts sitting at $500 billion for 2026 alone, we can be sure that where AI is today, it won’t be at the end of the year. It must be said that its’ reception and effect has differed significantly across industries. Software development probably had the dimmest forecast, with many people speculating that many software developers, if not most, would lose their jobs to AI. That hasn’t been the case at all though and just goes to show how uncertain even the most certain of predictions are regarding AI and its current roles, and future roles.
With that said, it is impossible for any industry that makes use of software as some part of its workflow, to not be affected, and that puts Structural Engineering square in the AI crosshairs.
State of AI in the Industry
Let’s firstly look at where we currently stand. The formal development of AI based tools, specifically focused on the Structural Engineering industry, does not seem to have reached any significant heights. Even the latest versions of well known software packages such as Prokon, Autodesk’s Robot Structural Analysis (Robot), IDEA StatiCa, SCIA Engineer, Dlubal’s RFEM and so forth, do not seem to make use of any new AI based tools. These packages, with the exception of Robot, have improved significantly over the years, but no major, if any, strides made regarding the incorporation of AI in any workflows. But after giving this some thought, I realized that there is most definitely some form of AI functionality in all of these programs, it just depends on what you would define as being attributeable to AI. In my mind, things like the Group Verification and Group Design of Robot, or the Select Lightest Sections functionality of Prokon (to name a few) can be put in the AI bracket; although we might call it dumb AI. But let us consider it, with some setup requirements we can, with the click of a button (maybe more than once), get a software package to look at masses of data, process that data in line with a specified design code and then specify a particular profile which meets all of the relevant requirements.
Again, comparing this to what we see out there, it might seem super simplistic, but it still completes a task which can take an engineer many hours, in a few seconds. It is taking a repetitive task and running through all of those repitions, with their corresponding variations, in a matter of seconds. That to me, falls in the AI category. We often exclude certain types of functionality from the AI pool because it doesn’t work with the typical prompt-based approach of tools like ChatGPT, Gemini, Grok, etc. but if we think about it, all it is is a fixed prompt: Optimize these members, as an example. This is but one example of how, if you are willing to consider it in a specific way, AI has indeed become part, and been a part of, the Structural Engineering industry.
But I get it, we are specifically thinking of the role of non-restricted prompt based AI workflows. We want to get an Architectural model, upload it and tell the AI “agent” to load, analyze, design AND detail the structure for us. We want it to be intelligent enough to know that there is more to consider in the design of a structural element than purely load resistance. We want it to understand transport, availability, cost, constructability and all of the other variables that go into the workflow. Designs that took months, if not years, and required massive (and diverse) engineering teams could then be completed by one person, possibly in a few days, giving room for the computational time even an advanced AI “agent” would need. Human experience will essentially become negligable as these “agents” will retain an unlimited number of years of experience, if it has access to it of course. (Most experience in the industry is shared verbally and not necessarilly recorded in such a way that an online based entity would have access to it). AI would be able to optimize at unconventional levels, designing non-standard/compound members. And it would’ve run the numbers. Cost of manufacturing versus overall structural benefits and so forth.
The reality is that a form of this type of workflow already exists. . . . .sort of. For example, I gave Google’s Gemini the following prompt: “Desgin a 5m I-beam with a torsional force of 115kNm in the middle. Design it using SA standards. If there aren’t SA standards, use the next best thing, making sure that you provide references at every step.” Below is an extract of the response.
The complete response was lengthy but in the end it designed a boxed I-Beam, indicating steps along the way. The prompts can be significantly more complex and you will get a comprehensive answer. Something that would’ve taken an experienced engineer, probably and hour or two to do, done in a few seconds. But, I can already hear the murmurs: “How do we know it has done this correctly? Yes, it provides references, but are those checks sufficient? Can I verify this by hand? I need to go and read through the code for myself.”
These murmurs are underpinned by the very thing that should (I use “should” very intentionally) keep AI in its place in the Structural Engineering industry. Structural Engineering is obsessed (and rightfully so) with safety. It deals with a skillset developed to effectively model, load, analyze and design structures in such a way that they are safe for use by humans and fit for purpose. Behind all of this sits a person which, because our industry deals with safety of people, needs to take legal responsiblity for the given output. Someone, always, needs to poke their neck out and say: “I deem this safe for use and fit for purpose.” That someone understands the legal ramifications of failure and should have a strong value framework of human life. And that is exactly it. The human aspect of Structural Engineering is of the utmost importance. AI cannot have the foresight of possible guilt and deep-seated regret, and that is incredibly important to realize. Also, simply on the legal front, a signature from a human being will always be required. You cannot take AI to court. Sure, you can hold the relevant company responsible, but you took the risk in using the tool and didn’t verify the results. I genuinly believe that this is at the forefront of any experienced engineer’s mind when we think of “AI in the Structural Engineering Industry”.
So, what should we watch out for? Well, I think that an issue has been flagged for many years, and the extensive incorporation of AI into the industry is an extension of it. I have heard many seasoned engineers complain of engineers using software but not understanding the output. When asked to assess and verify the data, they are unable to do so, but the availablity of software an ease of use, emboldenss some to “design” structural elements without the necessary foundational knowledge required to make critical decisions. There are many scenarios that make this more nuanced, but the core point is that simplicity of use and functionality should not replace the necessity for the foundational knowledge needed to correctly justify the outputs. I believe that AI can only further aggravate the issue. Although I could prompt an AI “agent” to design a section in torsion, it should never replace the need for me to understand the process for myself. If I do not know how to design a section in torsion, I should consult more than just ChatGPT or Gemini, but spend the required time in verified engineering resources (design codes, textbooks, etc.). AI should never be seen as a replacement for experience and sound judgement, but rather as a incredibly handy tool.
Conclusion
The adoption of AI in the Structural Engineering industry has been around for quite some time, although that would depend on your definition of AI. Advancements have been minimal on the unrestricted prompt-based approach as with models such as ChatGPT or Gemini. That being said, AI could very well become a significant contributer to efficiency and optimization in the industry. It cannot, however, replace the human component, both morally and legally.
Engineers should not rely on AI as a source of engineering knowledge but should have access to non-AI-based resources which can help and guide them in the structural engineering workflow, and should see AI as a tool, in line with that of software.
