Structural Engineers to AI, a Friendly Handover, or a Hostile Takeover?

Will bots take over our jobs? Or should I say: when will bots take over our jobs? These are the pressing questions of the day, urgently talked about in everyday conversations, especially after the recent hype of OpenAI and ChatGPT. The structural engineering profession is not exempt from this concern or opportunity, depending on whose point of view you seek. An architect may yearn for a day when her creative concepts are realized at the push of a button with no push back or arguments from the structural engineer. Developers may dream of having a structural engineering bot, and an architectural one for that matter, that would translate their investment ideas into concepts and construction drawings, all while being offered endless design options and cost estimates, without the need to involve human professionals. Some AI tools are already on the market. Take for example Midjourney, which generates images and architectural concepts by simply responding to a textural message. This article’s image was in fact generated by Midjourney simply using the gist of its title. While these tools are still early versions of the AI agents needed to fulfill the above aspirations; they do illustrate possibilities, which were unimaginable a few years ago.

?Evolution in the Structural Engineering Practice

For a practicing professional observing this evolution, the above scenarios are not farfetched and are ultimately inevitable. Changes in structural engineering have been so profound in the past few decades, thanks to the exponential growth of computing power and the tapping of this digital torrent by the software industry. These changes have already radically transformed how structural analysis is performed, design calculations are automated and Building Integrated Models (BIM) are generated. This is a far cry from engineering offices in the late seventies when I started my practice; back then, we worked in rooms full of drawing boards with engineers stooped over their desks with slide rules in hand.

?While tasks were transformed, professionals and technicians did not become obsolete. Instead, the skillsets of those who survived the change evolved. Not only did the design environment change by the newfound digital power, but expectations of developers and architects did as well. Buildings went up higher, they defied gravity with daring articulations, design efficiencies became the rule as uncertainty in analysis and design diminished through calculations made by accessible and powerful computing devices. A rigorous three-dimensional analysis of a complex structure was inconceivable forty years ago and would have theoretically taken a lifetime to complete manually; today, it is routinely performed using relatively inexpensive hardware and software. Reliance on Excel sheets, Mathcad and a slew of analysis and design software packages became mandatory for any structural engineer. This also led to changes in building codes, as they adapted to the latest computational capabilities. They stipulated requirements that could not have been fulfilled through hand calculations.

?What will AI Change, then?

So how will artificial intelligence take this already automated structural engineering scene to the next level?? Although AI as a discipline is more than half a century old, its power has only recently been widely realized in view of the maturity of the internet and the tremendous evolution of computing power. Currently the term AI has become overused and confused in common conversations with computing power at large. So before delving into how AI can transform structural engineering, we need to agree on what we mean when we say AI.

A simple definition of Artificial Intelligence is that it is the engine that allows computers to behave in ways that mimic and possibly go beyond human capabilities. It includes tools such as machine learning, deep learning, neural networks, computer vision and natural language processing. Let’s not confuse this with Artificial General Intelligence (AGI) which would become a reality when an intelligent agent becomes capable of accomplishing any intellectual task that a human can perform.? At that time such an agent would hypothetically not only replace a structural engineer but would singularly replace every professional or scientist. There would not be a need to classify such a bot as a structural engineer or a physician because it will be able to replace all the above and exceed their capabilities by leaps and bounds.

?This however is not an imminent or even a forgone eventuality. There are many opinions among specialists as to the timeline of reaching AGI; the most optimistic predict that it will still take us a few more years, so this is not the focus of this article. Instead, attention will be paid to the applications conceivably in the pipeline that will tap OpenAI and other AI platforms currently competing for dominance. I will argue below that these apps will not soon, replace a structural engineer entirely but will overtake and/or transform aspects of such an engineer’s duties.

?Potential AI Usage in Structural Engineering

Machine learning algorithms can use supplied data or otherwise scrape the internet for endless data to recognize patterns and learn insights. Such insights can offer intelligent advice to users, with varying degrees of accuracy depending on the amount of relevant data learned. It is readily conceivable that an AI foundation model can be developed to design, say, common wood houses through training on data about the structural components of thousands of houses, say from municipal databases, without the need to analyze and design in the traditional way engineers and builders are accustomed to doing. This approach would not be a game changer as far as the housing industry is concerned in view of the ample available tools for readily sizing wood members and the automation of such design but, if it becomes reliable, it will signal a profound shift from how structural engineering has been performed for decades.

?Applying machine learning to the design of more complex and unique buildings or bridges is limited by the size and diversity of the database. Perhaps what a current AI may be set up to do is to replace a structural expert who can make a judgement on the feasibility of say an iconic stadium concept. The AI will rely on interpretations and extrapolations from previously constructed stadia, and perhaps other man made or even naturally occurring forms. By contrast a human expert will tap his own experience and what he knows of others’ experiences, follow that with rudimentary calculations and finally build a 3D finite element model for the stadium to carry out a rigorous analysis. This analysis would confirm and refine the initial understanding which was based on expertise. This structural expert/automation dialogue may become an AI/automation exchange. At this time, it would be expected that AI bots, if configured accordingly, would access structural packages such as ETABS, SAP2000 or RAM to perform the analysis. It perhaps can also be configured to tap other software to perform design, obtain material take-off and produce a cost estimate. It is, however, conceivable that with deep learning AI may ultimately figure out other approaches that incorporate the whole workflow including potential generation and use of code to replace the above-described automation tools.

?Often, as in fact I have done in the above discussion, emphasis regarding the role of AI is primarily paid to design of new structures (greenfield developments), which is only one of many responsibilities entrusted to a structural engineer. This may be because it is the activity that lends itself the most to machine overtaking the role of the structural engineer. Remember that automation has already taken over many of the analysis and design tasks. This is as engineers keep busy feeding and extricating data and results, respectively, from specialized software. Other structural engineering activities may yet be taken up by intelligent bots, both automation and AI based. These agents would, naturally, have different capabilities than those needed for greenfield design. This will occupy the latter part of this article, but I will start with some elaboration on the greenfield design scenario.

?Greenfield Structural Design

As discussed above, automation has been deeply rooted for decades in the structural design process, from analysis to design, detailing and drafting. For repetitive structural systems such as steel warehouses the process is sometimes completely automated including, at the end of the workflow, the generation of instructions to Computer Numerical Control (CNC) machines to cut, mill and drill steel members to be ready for erection. Such processes may be enhanced through AI, but the real gain would be with the one-off unique design where the role of the structural engineer is currently crucial and where no fully automated process is likely to succeed without AI. Aside from convenience, the objective of an AI driven process would be to improve efficiency, avoid errors and omissions and create opportunities for enhanced innovations.

?While the structural engineer may oversee the process, at least initially, ultimately the AI agents will inevitably take over. Large Language Models (LLM) would allow architects and developers to speak or textualize in plain language to express their wishes and goals for the planned development, as is done with Midjourney. However, in the next generation the AI will go beyond the spatial configuration options to produce a buildable, functional, and feasible reality. Some early examples of generative design are already on the market. Testfit, and Hypar explore instantaneous site development potentials and achieve efficiencies through AI generated iterations. The latter’s website claims automated conceptual design, estimation, building product configuration, and mechanical & structural engineering. In discussions with the provider, it became clear that it is early days of incorporation of structural engineering in the planning let alone ultimate design. But it is their objective to do so. In fact, the path forward does not require a lot of imagination to foresee, even if the realization is not imminent.

?With ever-growing computer power and potential link between AI tools and structural analysis software, the accurate capturing of behavior of structures with multiple permutations thereof is predictably achievable. As the output is linked to BIM, the properties of the structure can be captured by the AI, including geometry, cost and embodied carbon. An AI can then invoke tools such as Life Cycle Analysis (LCA) for the Global Warming Potential (GWP) of each design permutation and recommend the structural option that best meets multiple, often conflicting, project objectives, such as cost and sustainability. Those would be declared at the outset, in plain language, by the property owner. Going through such cycles without AI has the potential pitfalls of massive effort, probable bias of team members based on their previous experience, limited access to ever-expanding data on materials on the market including their Environmental Product Declarations (EPD), etc. In a nutshell, AI would be the ever-learning expert that steadily guides the process objectively and swiftly.

?Before we turn the page on the subject, it is worth reflecting on the question of liability. Like the conundrum associated with self-driving cars, as to who is responsible in case of an accident, the question of liability for errors and omissions becomes one that needs to be settled before the structural engineer hands over professional seals to the machine.

?Expansions and Retrofit

Maintaining the built-up fabric, as opposed to demolishing and building anew, is a current trend that is only likely to increase for social, heritage, sustainability, regulatory and even economic reasons. One is already seeing more retrofit and adaptation projects than greenfield ones, particularly in older urban areas in North America and Europe. The role of the structural engineer on such projects includes condition assessment, structural evaluation, seismic and/or structural retrofit and other activities related to such rehabilitation or expansion. AI could replace the engineer’s judgement and expertise if supplied with all necessary data. This data is however multifaceted and includes visual information from site, historic information from reports and interviews, architectural and user aspirations, testing requirements and results, data used in construction, and other less tangible parameters that are specific to the project at hand. The workflow is not as linear as that of a new design, thus the training of the AI to repeat and extrapolate from past cases is less evident here.

?Let’s first reflect on project specific data acquisition. Digitizing tools that allow access by AI agents to real world information are becoming quite common. For example, capturing 3D models of the building fabric through LiDAR technology has become feasible enough for some A/E practices to keep the hardware in-house. Even basic iPhone versions are now available. The next technology that is gaining momentum is that of digital twins which exceeds the 3D models by the ability to integrate data from various systems and to support different aspects of the facility, from concept to operations.

?At a more basic level, it is not apparent that AI can at this time interpret blueprints, Revit models or point clouds, even if it is conceivable that it will ultimately be able to do so. Just as AI foundation models were trained through deep learning to acquire and interpret human knowledge, they will be able to understand and interpret such engineering media. Various software providers claim such early signs. Kreo advertises their system’s capabilities in reading blueprints and performing material takeoff. CrowdANALYTIX for Blueprint’s site has an AI solution that semi-automates the reading of drawings by use of object recognition and semantic segmentation. Automated solutions are available which convert PDFs, to AutoCAD DWGs and DWGs to Revit models, and those in turn to structural models such as ETABS; all depending on quality of original and intermediate data, and often with significant manual intervention. Such automated processes, while still imperfect, are a step in the right direction of data acquisition and ultimately interpretation by AI.

?Getting back to the process currently followed by a competent structural engineer in such rehabilitations, reading drawings is only one of the steps and is an intuitive one for a human engineer. Once read the engineer may determine that they are lacking key information or are inconsistent with existing conditions. Sometimes an intervention requires discussion with a heritage representative or an official of authorities having jurisdiction. These are but a few scenarios that make the replacement of the human structural engineer less foreseeable for such responsibilities than for those involving greenfield projects.

?Assessments and Forensics

From personal experience and that of my colleagues, gathering ample information from a site under evaluation or that of a recent failure that one is forensically investigating is key to a successful study. Matterport claims the use of a combination of AI and LiDAR-enhanced 3D cameras to transform physical buildings and spaces into what it describes as photorealistic digital twins. Whether it is this generation or a coming one, with such models a structural engineer can endlessly view, with acceptable precision the condition of the visible structure including cracks, spalls and other visually obtainable parameters which inform of the condition of the structure or provide clues as to the cause of failure. Such interpretations can plausibly be done by AI, and with higher precision and efficiency. What this would still be lacking, however, is the tactile part of the inspection, whether it is by tapping for delamination or removing a ceiling tile to view a concealed soffit. Of course, that is not to mention the requisition of destructive or non-destructive testing, the performance of such tests and the interpretation of the results thereof. With all data in hand an AI agent may ultimately be a better, more precise, and less subjective analyst. Before getting there though, the role of the field structural engineer would need to be taken over by robots walking the site and perhaps drones hovering into the nooks and crannies of the investigated building, a proposition that seems far from feasible for now.

?Construction Administration

This article does not address any of the processes typically performed by the constructor. Some require individuals with structural engineering training. Reference in this section is therefore only to the activities typically carried out by the consulting engineer during the construction stage. The level of involvement of such consultant varies from practice to another. In my experience, consulting engineers in the US have somewhat limited on-site presence, while in the Middle East, for example, full-time presence is common. Canada falls in between but leans more towards the US approach. Their activities are quite varied and include field reviews, clarifications on documents and sometimes adaptation of design to found conditions. While automation using software such as Procore helps streamline the process, the role of the CA structural engineer seems harder to replace. Ultimately AI may potentially fulfill the expert role when all data and requirements are fed from boots on the ground, if not robots that is.

?Other aspects, such as the contractual and negotiation aspects, fundamental to any construction project, also make the role of AI harder to predict. While, for example, performing a cost estimate of additional work can be done accurately by AI, enforcing that on the contractor would require a paradigm shift from how construction contracts are currently drawn up.

?Final Notes

As I write these lines, I am taken back to a meeting in 1981 with my PhD advisor, Professor Ed Wilson at UC Berkeley, the creator of the original CSI suite (ETABS, SAP2000, etc.) which he first dubbed SAP 80 to mark the year 1980 and the Intel microchip 8080, which powered the newly invented Personal Computers running his novel software.? In that meeting, he challenged me, as he did with other students, to find ways to simplify user data input, which was done at the time through line editors and text data files. In just a few years after that encounter, some of my talented students in a Computers and Structures course studiously created replicas of the likes of STAAD and produced others of their own design with sophisticated graphic user interfaces, all within a semester duration.

?The pace of progress has been remarkable and anyone predicting the future should not be jittery about being off-the-mark. What is certain is that computer power will continue to expand, and that AI will be mobilized progressively and rapidly. This can lead to more creativity, reliability, and efficiency. AI will inevitably engender leaps in structural engineering, as it likely will in all other disciplines. These advancements may be in the materials we use, the systems our hitherto training is accustomed to employing and the empirical or probabilistic concepts that are embedded in our codes and standards.

?How will this affect the people in the profession is far from certain, but what this article predicts is that there are still many years when some duties will continue to be performed by technicians and professionals because it is hard to replace their type of work by AI. There are other activities that will be done more effectively by AI, especially automated greenfield design. There may also be new human chores, some sophisticated and others less so, that will be generated in the service of the AI for engineers and technicians with upgraded skillsets.

?Another factor that has not been discussed is how unrushed the change in building codes, standards, regulations, laws, etc. are compared to the urgent progress now set at the pace of the machine. Some brakes are being discussed and likely will be enacted in the context of avoiding missteps resulting from the expected rush to AI. Perhaps structural engineering would not be immune and perhaps there is still room for a generation or two of structural engineers.

?