CONSTRUCTION SEMANTICS & GENERATIVE PRE-TRAINED TRANSFORMER (GPT) LANGUAGE MODELS
CONSTRUCTION SEMANTICS & GENERATIVE PRE-TRAINED TRANSFORMER (GPT) LANGUAGE MODELS
Ankur Mitra & Ahmed Hassan - 01.07.2023
Generative Pre-trained Transformer language models, also known as ChatGPT, have taken the world of artificial intelligence and future technology by storm. Created by OpenAI, a non-profit research laboratory formed in 2015 to foster collaboration among leading AI researchers across organisations and global locations, ChatGPT is an advanced language model developed by fine-tuning OpenAI’s foundational GPT 3.5 and GPT 4 models. ChatGPT uses supervised and reinforcement learning techniques during its development and training processes to continually improve its conversational abilities and provide more accurate and contextually appropriate responses. Since its launch in November 2022, this AI-powered chat-bot has quickly amassed 100 million active monthly users, a feat that took Instagram two and a half years to accomplish after its release.
What makes ChatGPT so popular is its ability to provide answers to virtually any prompt. It can present factual information, write an essay, story, poem, or even a programming code. Unlike traditional search engines that rely on keywords and phrases, ChatGPT comprehends and interprets the given prompt, formulates an answer based on past data, and presents the answer in a concise and conversational manner.
ChatGPT accomplishes this impressive feat by utilizing a massive corpus of data written by humans, including books, documents, articles, and content from the open internet, which has been fed into the algorithm. The model is trained to forecast the optimal responses for any given prompt. This training involves employing a technique known as “tokenisation,” which splits the prompt into strings of words and predicts the best answer based on the training data.
So, what is in it for the construction professionals? Asking this question to ChatGPT provides a concise reply:
While the response is impressive, it remains necessary to approach it with cautious optimism. Experienced construction professionals are familiar with the notion that new digital solutions often claim to address all issues but may not fully deliver satisfactory results in practice. Hence, it is crucial to conduct comprehensive testing of the chat-bot to ascertain its effectiveness in addressing specific industry needs.
Q: “I am an Irish contractor in the residential buildings sector - what fire regulations do I need to look out for developing a 5-storey residential building?”
The above question yielded the response mentioned below:
This information could be valuable to engineers with little practical experience. Likewise, it could remind experienced professionals of the codes and regulations they must consider before starting a new project. Nevertheless, the trial progressed to demand more detailed answers by providing the chat-box with the following question;
Q: Could you please describe the way to design the escape route for five storey building?
And finally, we asked about a specific detail;
Q: What should be the minimum dimensions for windows used as escape routes in residential buildings?
As the inquiries became precise, ChatGPT’s responses kept pace with remarkable accuracy. Also, It is worth noting that each answer’s conclusion was delivered with a notable degree of humility, underscoring the importance of seeking specialized advice or human review.
The academic exploration of ChatGPT’s capabilities in construction has yielded intriguing insights. In one study, Jamil Uddin et al. (2023) investigated the chat-bot’s potential to enhance safety in construction scenarios by asking a group of people to review images of construction sites to identify potential hazards with the help of ChatGPT. The promising results made the researchers assert that the chatbot could be a valuable asset for the future of construction. However, for a cautious realist, it is difficult to become overly enthusiastic about this prospect, as evaluating construction safety from comfortable surroundings is a far cry from the harsh and unpredictable conditions of the field.
Prieto et al. (2023) focused on a distinct construction challenge: developing a project schedule for building a partition wall. While study findings were not necessarily infallible, the digitally produced work breakdown and timeline estimates demonstrated high comprehension. With further refinement and real-world training, ChatGPT could become capable of handling more complex scheduling tasks.
Undoubtedly, utilizing a customized chat-bot constructed upon the ChatGPT technology will furnish exceedingly precise answers. Producing technical knowledge in construction areas such as design, planning, and compliance can be partially, if not entirely, digitized. This notion should not be misconstrued as a forewarning of the dreaded “AI takeover” since human oversight and expertise remain vital. However, the primary benefit lies in the significant increase in the output of construction professionals with intelligent digital assistance. The increasing need for enhancing the productivity of construction personnel has become an imperative national necessity requirement in the present times due to skills scarcity and the sustained expansion of global construction activity. Hence, the prospect of having a knowledge repository that can supply historical data and anticipate future iterations based on past data is inspiring.
References:
Uddin, S.M.J., Albert, A., Ovid, A. and Alsharef, A. (2023) “Leveraging ChatGPT to Aid Construction Hazard Recognition and Support Safety Education and Training,” Sustainability, 15(9), p. 7121.
Prieto, S.A., Mengiste, E.T. and De Soto, B.G. (2023) “Investigating the Use of ChatGPT for the Scheduling of Construction Projects,” Buildings, 13(4), p. 857.
ChatGPT was used in the process of writing the article.