To view the figures and tables associated with this article, please refer to the flipbook above.
Artificial Intelligence (AI) has really taken off since the release of the initial version of ChatGPT in 2022 by OpenAI, and its use has accelerated with the many subsequent versions, having a significant impact in so many different areas of life. The newest open-to-public version from OpenAI is GPT-5, with more sophisticated capabilities for solving problems using the deeper reasoning model (GPT‑5 Thinking) and GPT-5 Pro with research-grade intelligence. AI is rapidly growing and affecting our daily lives, both professionally and personally. So it is important to understand its benefits and limitations applied to structural engineering and how it can be used by the profession going forward. Until about 1980, structural analysis and structural design of buildings, bridges, and other structures, were still typically performed using hand calculations. However, since about 1980, most structures have been analyzed and designed using computers and dedicated, commercially available structural analysis software, from beam elements to sophisticated finite element solutions, including shell and solid elements. While hand calculations are still used by structural engineers, this is typically relegated to spot-checking the computer results.
In 2022, another potential option suddenly presented itself to solve structural analysis problems, including statically indeterminate structures, with the release of the first version of ChatGPT. Engineers now have three possible ways to solve structural engineering problems: (1) hand calculations, (2) developing a computer model of the structure and solving it on a computer with dedicated structural analysis software, and (3) using generative AI.
Since the first author has, for decades, been steeped in hand calculation methods and deriving unique closed-form solutions to complicated structural engineering problems, allowing these to be solved by hand, and the second author is an expert of AI applications in structural engineering, this article presents several examples of statically indeterminate structures and compares the AI results to those found from the relatively simple hand calculations. When ChatGPT and hand-calculated solutions did not match, another hand calculation method was used to verify the hand results. In the following, five different statically indeterminate problems are given, asking for either the numeric or symbolic solutions from ChatGPT, depending on the problem statement. Interestingly, a computer model is not required to solve these problems with ChatGPT, at least not a model developed by the user. For each problem, three independent trials were conducted. ChatGPT was provided with a screenshot of the problem statement and the corresponding structural image, along with the following prompt: “Think thoroughly, satisfy governing physics and compatibility requirements, and solve this.” Notably, even a simple hand sketch of the structure proved sufficient in place of a detailed engineering drawing, as ChatGPT is trained to interpret visual inputs through advanced visual-understanding capabilities.
View or download the file below to see the 5 problem examples.
Conclusions
AI is being used in all sorts of ways, including in structural engineering. This article considered state-of-the-art AI for public use, focusing on the most recent version from OpenAI, GPT-5, including the more advanced GPT-5-Thinking and GPT-5-Pro versions for their higher problem-solving capabilities. Five statically indeterminate problems were solved by hand calculations and then compared to the ChatGPT solution. In all cases, the input to ChatGPT was just a simple drawing of the problem and the statement that it needed to solve it by thinking. From the drawing, ChatGPT correctly interpreted the member lengths, point loads and distributed loads, boundary conditions, and the different moment of inertia values for the various members. It also understood when the point loads were applied at mid-span, based on the geometry of the drawing. To solve the problem, ChatGPT wrote its own computer program in Python and then ran it. Interestingly, it relied on classical methods of structural engineering, rather than inventing its own new technique, and then wrote the program to solve the problem once it had decided on a method to use - typically Moment Distribution, the Stiffness Method or the Slope Deflection Method. For one problem, it solved it three different ways in the three different attempts. When there was a difference between the ChatGPT and hand calculation results, another hand calculation method was used to verify the initial hand calculations. In addition, for all five problems the first author’s Closed-Form Method was applied by hand to verify the results.
The AI approach is extremely simple and sometimes provides the correct results, but it often gives the wrong answers, even on repeat attempts for the same problem that it got correct on another attempt. And sometimes the results were off by a lot. This variation in results is expected because large language models (LLMs) such as ChatGPT rely on probabilistic reasoning rather than deterministic computation. Each attempt follows a slightly different logical path, depending on how the model interprets the problem, leading to different intermediate steps and final answers. Note that for each attempt of the same problem, no information was given from one try to the other; they were all completely independent efforts. Also, after providing ChatGPT with the initial information for a given problem, there was no human involvement whatsoever. Clearly, had we guided ChatGPT along its path to a solution, better results could have been found and, perhaps, the correct solutions obtained more consistently. It seems that for structural engineering applications, especially for statically indeterminate structures, AI in its current state should be used as an assistant to the structural engineer, helping with given tasks, but not allowed to just move along on its own, from start to finish. Results from the five examples in this article clearly show that AI is powerful, but can be wrong, and for various reasons. To use AI, the structural engineer needs to verify key elements of the analysis and prod it to change direction if it gets lost.
While the examples in this article demonstrate the remarkable capability of ChatGPT to interpret sketches and write its own computer code and reason to solve structural problems, this may not be the most effective use of AI for engineering work. The proper role of AI should be to assist in developing, validating, or automating components within established engineering workflows, rather than replacing them. At this stage, a more practical approach would be to use AI to generate or translate model inputs for established matrix-based analysis programs where materials, boundary conditions, and load combinations can be defined and verified by the engineer. In this way, AI could act as an intelligent interface—converting sketches or descriptions into preliminary models, plotting and visualizing the structure, or writing custom code to perform specific tasks—while leaving final analysis and design decisions under direct human control. Future implementations may evolve toward an agentic AI workflow which involve existing software/tool use, where AI systems interpret and help reason with the problem to complete the solution process collaboratively. These examples therefore highlight both the power and the current limitations of AI, emphasizing that its greatest potential lies in assisting the engineer to work more efficiently, not in performing complete tasks autonomously. Ultimately, the structural engineer, the human, is still responsible for the analysis and design of structures, and the proper use of AI, computer modeling and hand calculations. ■
Try It
Want to experiment yourself? Download the Chat GPT prompts and hand calculations below and compare the results.
About the Authors
Robert K. Dowell received his B.S. degree in Civil Engineering from San Diego State Univeristy (SDSU), and his M.S. and Ph.D degrees in Structural Engineering from the University of California at San Diego (UCSD). He is a licensed Civil Engineer (PE) and a Professor of Structural Engineering at SDSU.
Dr. Althaf Shajihan, Assistant Professor at San Diego State University, received his Ph.D. in Civil Engineering and M.S. in Computer Science from the University of Illinois Urbana-Champaign. His research bridges structural engineering and artificial intelligence to advance the structural assessment of civil infrastructure.