Generative AI: The New Geotechnical Assistant?
Publication: Journal of Geotechnical and Geoenvironmental Engineering
Volume 149, Issue 10
Artificial intelligence (AI) is the science and engineering of making intelligent machines (McCarthy et al. 2006). Generative AI is a subcategory of AI that generate new content such as text, images, or sounds. Some examples of generative AI platforms include ChatGPT, Google Bard, DALL-E, and Musico. Researchers have explored the intersection of AI and civil engineering for several decades, and some early implementations include neural network–based material constitutive models (Ghaboussi et al. 1991). Generative AI is different from these earlier AI applications in that it produces new content, such as written reports. In March of 2023, ChatGPT-4 was introduced as OpenAI’s most advanced chatbot, capable of reasoning and problem-solving.
In a technical release, OpenAI showed that ChatGPT-4 performed well on several academic and professional exams, but civil engineering exams were not included. The authors prompted ChatGPT-4 with National Council of Examiners for Engineering and Surveying (NCEES) civil and geotechnical practice exam questions. ChatGPT-4 correctly answered six of nine provided questions, a score of about 67%. Questions with images were not included because the authors did not have access to ChatGPT-4’s vision capabilities. This score encouraged the authors to investigate ChatGPT-4’s usefulness for performing common geotechnical engineering tasks.
First, ChatGPT-4 was prompted to write a geotechnical data report for a sandy silt embankment with standard penetration tests of 6–12 blows/m (). The chatbot drafted an acceptable report template with introduction, site description, subsurface investigation, soil description, laboratory testing, geotechnical recommendations, and conclusion sections. ChatGPT-4 identified that the soil is a medium dense to dense sandy silt with a USCS symbol of ML or SM, which is reasonable. The report included placeholders for commonly documented information, such as the depth and number of borings. The report also included a placeholder for foundation design and slope stability considerations.
Second, ChatGPT-4 was prompted to write a function using Python version 3.9 to calculate the ultimate bearing capacity of a spread footing using Terzaghi’s bearing capacity equation. The chatbot identified the required input parameters and correctly wrote the bearing capacity equation; however, the unit weight bearing capacity factor, , was calculated incorrectly. ChatGPT-4 then was prompted to write a Python function to calculate an infinite slope factor of safety for cohesive and cohesionless soils. The chatbot correctly implemented the cohesionless soil calculation; however, it omitted one trigonometric function in the calculation of cohesive soil stability.
Lastly, ChatGPT-4 was prompted to evaluate whether liquefaction would occur in a sandy soil layer given values for cone penetrometer test (CPT) tip resistance, peak ground acceleration (PGA), layer depth, and groundwater depth. The chatbot recommended the use of the “OpenAI (2023)” simplified empirical method. The chatbot estimated a sand unit weight of and calculated the critical stress ratio (CSR); however, total and effective stresses were switched in the chatbot’s calculations. The chatbot wrote that the cyclic resistance ratio (CRR) cannot be calculated without a complete set of CPT data. The authors asked ChatGPT-4 whether liquefaction would occur if the CRR was 1.0. The chatbot responded that a CRR of 1.0 is unreasonably high, but for the sake of argument, liquefaction would not occur because the CRR would be greater than the CSR.
Implications
ChatGPT currently is prone to errors, as engineers sometimes are, and quality control is needed. Although ChatGPT may be capable of passing professional licensure exams, AI likely will not provide the final signature of approval for engineering designs for the foreseeable future. The authors asked ChatGPT-4 if the chatbot could replace the job of a civil engineer. ChatGPT-4 responded, “AI models like ChatGPT can be used to automate some tasks or provide information more efficiently, but they cannot replace the human expertise, judgment, and creativity required in a field like civil engineering.” ChatGPT has the potential to substantially increase productivity within geotechnical engineering. This technology remains in its infancy and is expected to improve significantly with more-advanced algorithms and additional focused geotechnical engineering training.
The US soon may experience a shortage of civil engineers as infrastructure investment increases and civil engineering college enrollment remains flat. This labor gap presents an opportunity for generative AI tools, such as ChatGPT, to serve as a new generation of virtual geotechnical assistants at a time when civil engineers are in high demand.
References
Ghaboussi, J., J. H. Garrett Jr., and X. Wu. 1991. “Knowledge-based modeling of material behavior with neural networks.” J. Eng. Mech. 117 (1): 132–153. https://doi.org/10.1061/(ASCE)0733-9399(1991)117:1(132).
McCarthy, J., M. L. Minsky, N. Rochester, and C. E. Shannon. 2006. “A proposal for the Dartmouth summer research project on artificial intelligence, August 31, 1955.” AI Mag. 27 (4): 12. https://doi.org/10.1609/aimag.v27i4.1904.
OpenAI. 2023. “GPT-4 technical report.” Accessed July 1, 2023. https://cdn.openai.com/papers99/gpt-4.pdf.
Information & Authors
Information
Published In
Journal of Geotechnical and Geoenvironmental Engineering
Volume 149 • Issue 10 • October 2023
Copyright
© 2023 American Society of Civil Engineers.
History
Received: Apr 15, 2023
Accepted: May 3, 2023
Published online: Aug 4, 2023
Published in print: Oct 1, 2023
Discussion open until: Jan 4, 2024
Authors
Metrics & Citations
Metrics
Citations
Download citation
If you have the appropriate software installed, you can download article citation data to the citation manager of your choice. Simply select your manager software from the list below and click Download.