Artificial Intelligence and Digital Tools for Assisting Low-Carbon Architectural Design: Merging the Use of Machine Learning, Large Language Models, and Building Information Modeling for Life Cycle Assessment Tool Development

Abstract

The construction sector is a significant contributor to global carbon emissions and a major consumer of non-renewable resources. Architectural design decisions play a critical role in a building’s carbon footprint, making it essential to incorporate environmental analyses at various design stages. Integrating artificial intelligence (AI) and building information modeling (BIM) can support designers in achieving low-carbon architectural design. The proposed solution involves the development of a Life Cycle Assessment (LCA) tool. This study presents a novel approach to optimizing the environmental impact of architectural projects. It combines machine learning (ML), large language models (LLMs), and building information modeling (BIM) technologies. The first case studies present specific examples of tools developed for this purpose. The first case study details a machine learning-assisted tool used for estimating carbon footprints during the design phase and shows numerical carbon footprint optimization results. The second case study explores the use of LLMs, specifically ChatGPT, as virtual assistants to suggest optimizations in architectural design and shows tests on the suggestions made by the LLM. The third case study discusses integrating BIM in the form of an IFC file, carbon footprint analysis, and AI into a comprehensive 3D application, emphasizing the importance of AI in enhancing decision-making processes in architectural design.

1. Introduction

The construction sector, including the decisions made during the design and construction of buildings, accounts for a significant portion of global carbon dioxide emissions, thereby contributing to anthropogenic climate change [1], and up to 38% of global greenhouse gas emissions [2]. In response, the European Green Deal aims for a net-zero emission economy by 2050 [3], introducing measures like the Fit for 55 package to cut greenhouse gas emissions by 55% by 2030 compared to 1990 levels [4]. On 12 March 2024, the EU Parliament approved a revised Energy Performance of Buildings Directive (EPBD), marking a significant move in enhancing building climate action, following an agreement in December 2023 after negotiations starting from the Commission’s initial 2021 proposal. The updated EPBD now requires the assessment and disclosure of buildings’ carbon footprints according to the EN 15978 [5] and Level(s) standards [6], initially targeting buildings over 1000 square meters from 2028 and expanding to all buildings by 2030, broadening its scope from focusing merely on operational energy consumption [7]. When considering architectural design, one of the leading approaches is to design for reduced environmental impact, which should be supported by appropriate analyses conducted at various stages of the design process [8]. Architects, engineers, and technical consultants have a significant impact on the carbon footprint of buildings through their decisions, designs, and material selection [9]. Designers, therefore, need more tools to assist them in the decision-making process, especially at the early stage of concept sustainable design. The proposed solution involves the development of a Life Cycle Assessment (LCA) tool. The main tools that are presently available can be divided into three categories: Life Cycle Assessments (LCAs), building information modeling (BIM), and artificial intelligence (AI).

The carbon footprint of a designed building (the total greenhouse gas emissions over the life cycle of the building) is one of the tools that can be used to check the impact of human activities on the environment. The carbon footprint includes the estimated total, direct, and indirect emission of greenhouse gases caused by a specific activity or action measured as the carbon dioxide equivalent [10]. It is typically calculated per reference unit, such as per building or per square meter of the building floor area. The total carbon footprint is divided into the operational carbon footprint (related to energy consumption) and the embodied carbon footprint (e.g., materials used for construction, renovation, replacement, modernization, or end-of-life processes). The embodied carbon footprint is often overlooked despite accounting for over 10% of global emissions, as the emphasis is mostly placed on reducing the operational carbon footprint [11]. From a 2050 perspective, the embodied carbon footprint is particularly significant because most of its emissions occur immediately. In contrast, operational emissions are spread evenly over the life of the building.

This article focuses on optimizing the carbon footprint of buildings in the architectural design process. One of the fundamental methodologies used for calculating the carbon footprint is the Life Cycle Assessment (LCA). This is a detailed analytical tool used to analyze the impact of products, buildings, and processes on the environment. The procedure for applying the LCA method in architectural design and construction is described in standard EN 15978:2012, “Sustainable buildings—Assessment of environmental performance of buildings—Calculation method” [5]. A LCA is more frequently used for the final evaluation of completed projects and less commonly as a tool to support designers during the design process [12,13,14]. In the modern architectural design process, there is a growing list of various environmental analyses being performed [15]. Studies show that optimizing the building form at the conceptual stage can reduce a building’s carbon footprint [16], even if LCA research is still incomplete due to various guidelines and their different interpretations [17].

There is a considerable number of different LCA tools on the market. Some tools are more specialized in performing very detailed LCA analyses not connected to the building sector, like SimaPro, GaBi, OpenLCA, and Athena [18]. Still, there are many tools dedicated to building LCA, such as OneClick LCA, LCAByg, and Tally. The extraction of high-quality data from building models and the development of inventories frequently encounter significant challenges due to the absence of precise methods for integrating building data effectively [19,20].

The concept of BIM has existed since the 1970s. The first software tools developed for modeling buildings emerged in the late 1970s and early 1980s. Currently, BIM technology is crucial in facilitating a thorough LCA of buildings, which is essential for enhancing their environmental performance [21]. BIM models offer an integrated approach to design, facilitating information management and collaboration among all stakeholders throughout the life cycle of a building project. This enables the exploration of various design alternatives, considering all potential variations and parameters during the initial phases of construction projects, and allows for automating the process of LCA study early in the design process [22]. In one study, Durdyev et al. highlighted that merging BIM and LCA can improve results in constructing energy-efficient buildings [23].

AI is an emerging technology that has become part of the digital shift and is receiving increased attention in sustainable building design [24]. AI encompasses a broad range of technologies that simulate human intelligence. At the same time, machine learning (ML) is a specific subset of AI that focuses on enabling machines to improve at performing tasks by learning from data. Various approaches are being applied in finances, the medical field, engineering, and internet solutions. In building design, the application of AI was very limited a few years ago [25]; however, many new examples have appeared in recent years [26].

Currently, AI is increasingly influencing architecture, with novel AI-related methods emerging each year. For instance, Rhee and Veloso demonstrated the use of deep learning to create urban layouts that mirror the style of those in the training dataset, showcasing a novel approach to urban design [27], while Zhao et al. demonstrated an approach on a smaller scale, utilizing ML for facade design [28]. Similarly, Theodore Galanos showed how machine learning can drastically reduce the time needed for environmental analysis [29], while Zargar and Brown utilized neural networks for the static evaluation of architectural designs [30]. The application of AI extends to optimizing buildings’ energy efficiency [31] or environmental footprint, as explored by Pomponi and others [32]. Furthermore, the technology was applied to predict the energy consumption of buildings in varying climates in work by Işeri et al., who leveraged neural networks in their predictions [33]. In the realm of smart cities, de Las Heras and colleagues detailed numerous ML applications for smarter urban planning [34]. Machine learning has also been employed to forecast the microclimate within agricultural buildings, according to a study conducted by Arulmozhi and his team [35]. Moreover, AI techniques, particularly convolutional neural networks (CNNs), have been used to analyze images, such as satellite or street photos, for various purposes. These applications include classifying different types of buildings [36], estimating green spaces and other land uses [37], detecting unauthorized structures [38], and evaluating roof efficiency for solar panel installation [39]. Lastly, the exploration of solution spaces, as discussed by del Campo, highlights the broad potential of AI in navigating complex architectural challenges [40]. A subset of solutions is also based on evolutionary algorithms, such as generating functional layouts of single-family houses [41]. There is a growing list of commercial applications that use AI in architectural design. This advancement has led to the rise of interdisciplinary research within the field, broadening the scope and depth of architectural studies [26].

Large language models (LLMs) have emerged as a groundbreaking development in the realm of AI, particularly in natural language processing. There are numerous applications of LLMs in research, including chemistry, surveying, and other disciplines [42,43,44], as well as architectural design [45], including in the context of BIM [46]. LLMs’ potential for assisting in low-carbon architectural design by reducing the carbon footprint is an area worth exploring. One noteworthy LLM example is ChatGPT, an AI model built on the GPT-3.5 architecture developed by OpenAI.

It should be noted that AI tools are altering the architectural industry’s planning, production, and building processes. Using these resources supports efficiency, speeds the design development, and saves time and resources for other design-driven issues such as cost analysis and green building solutions. These new commercial applications should include programs such as Adobe Firefly version 3 software, initially aimed at creating visual and textual effects, or other digital solutions using AI for complex architectural tasks, like Digital Blue Foam [47], InfraRed [48], Autodesk Forma [49], Kolega [50], and PlanFinder [51]. These and similar solutions are unparalleled in their ease of use, data mining, and web collaboration, helping building designers to implement projects more quickly and analyze them more quickly at an early stage. Outside their high potential, the presented tools are subject to discussions about whether algorithm-based solutions lack human intuition and emotional understanding, even though they are very supportive tools. The 2016 report “Technology at Work v2.0” [52] suggests that technology will have a significant impact on jobs in the developing world, creating new roles while requiring less human intervention in areas where emotional, creative, and interpersonal skills are crucial. However, certain jobs, like those in architecture and service design, will still heavily rely on human expertise due to the need for complex problem-solving and personal interaction, highlighting the importance of adapting to AI advances for continued relevance in the design field [52]. Thus, it is important to explore forms of collaboration between designers and AI models.

2. Methodology

Conducting a comprehensive analysis requires accurate data from an almost completed design. To perform LCA at earlier stages of the design process, a simplified version of the analysis should be used. This can be achieved, for example, by using statistical averages for processes and building components. Another solution could be the application of AI, which can assist designers in performing the analysis, automating the process, or finding optimal building parameters that result in the lowest carbon footprint. AI could also be used as an assistant to the designer in suggesting better choices. This study focuses on integrating artificial intelligence in architecture with the environmental analysis of buildings.

In order to develop a prototype carbon footprint estimator, several studies were first conducted on how different types of solutions could support architectural design work. Subsequent research focused on developing a prototype tool, and its earlier stages have already been published in scientific articles. Previous experiments consisted of creating a parametric model of a multifamily building and optimizing its carbon footprint via evolutionary algorithms [53]; applying machine learning for design space exploration [54]; and developing a prototype machine learning-based tool, including convolutional neural networks [55], with further development in a Ph.D. thesis [56]. As part of this research, analyses of the carbon footprint of similar buildings using various tools were conducted. This article presents the final version of the MLCO₂ Tool, as well as a subsequent tool developed from it.

3. Case Study 1. Machine Learning-Assisted Carbon Footprint Estimation Tool

3.1. Materials and Methods

AI has been identified as one of the answers to the problem of data accessibility and decision making with LCA early in the design process, allowing designers to perform analyses without a finished 3D model [32]. Machine learning-based solutions are based on either access to a pretrained model or to a dataset that can be used for training the model. In the first case study, an experiment was conducted by training the model using a simulated dataset. The creation of a simulated dataset was proposed by Galanos [29]. In this case study, the authors investigated a tool that can predict building carbon footprints based on a trained machine learning model. The authors chose Rhinoceros 6 and Grasshopper due to the possibility of using automation by applying visual scripting in Grasshopper. The authors developed a script that uses machine learning to estimate the carbon footprint during the architectural design phase. This tool can be utilized during the initial design stages, improving the understanding of the connection between a building’s form and its overall carbon footprint.

The development process has been divided into consecutive steps. As no dataset of multifamily buildings in Poland has been found, the authors generated their own dataset [55]. To achieve this, a parametric model of a multifamily building was created. This dataset encompassed buildings ranging from 1200 to 3200 square meters (Figure 1). The assumed height of the building ranged from 3 to 6 stories. Ten different parameters were recorded into a csv file for each generated example (

Table 1. Parameters recorded from each of the generated buildings [56].

Parameter Name
Wall Area
Ground Floor Area
Roof Area
Height
Window Area—South
Window Area—North
Window Area—West
Window Area—East
Climate Location
Construction Technology

). The values of the ten parameters were randomized. The surrounding urban area was also randomized to generate different shading scenarios. It was assumed that the building was located in Polish climatic conditions, and the climate data were used in the form of an EPW file (EnergyPlus weather format) [57]. The generated building shapes were then analyzed in terms of embodied and operational carbon [56] utilizing EnergyPlus for energy simulation [57] and the Oekobaudat database for embodied carbon data [58]. The training set consisted of 3300 examples divided into a training and test set at a 75% to 25% ratio. In the next step, these data were used to train the model (consisting of 10 previously mentioned parameters and png grayscale images that recorded the heights of the buildings in the surrounding area, as shown in Figure 2). The climate location parameter had 3 possible values based on 3 typical Polish climate locations. The construction technology parameter had 4 possible values based on 4 preselected construction technologies prevalent in Poland.

Convolutional neural networks were built using Keras version 2.3.1, an open-source neural network library for Python. The training process was performed in the Python Environment.

The result of this study was a trained neural network capable of predicting the total carbon footprint of a design proposal when using limited information. This information includes parameters such as the wall area, ground floor area, roof area, height, window area in different orientations (south, north, west, and east), and the urban layout, all automatically captured from the model created by the user inside Rhinoceros 6. The results are presented to the user using a user interface (UI) created using the Human UI plugin for Rhinoceros 6 and Grasshopper [59] (Figure 3). The tool was released as open source [60].

3.2. Results

The tool was tested on a conceptual project aimed at optimizing the design of a building to minimize its carbon footprint. The building in question had a total floor area of 2250 square meters, falling within the average range of floor areas in the dataset used to train the machine learning model. The test location selection was based on several criteria. Firstly, the plot had to be surrounded by buildings that could overshadow the analyzed plot within a radius of 15 m in order to include the effect of shading on the operational carbon footprint. The site had to be at least 2000 square meters to allow for different building layouts to be considered. In addition, a site with an orthogonal urban layout, such as the layouts on which the machine learning model was trained, was preferred. After evaluating several locations in Warsaw, a plot in the Mokotów district, specifically in the Ksawerów area near the streets of Domaniewska and Modzelewskiego, was chosen (52.182629, 21.013666). The location was selected based on a few premises: the site had to be in one of the possible climate locations used in training the model; the site had to be an empty plot, with some space to move the design around; and there had to be some tall surrounding buildings to cause overshadowing. There was a car park on the site. The urban context was modeled using Rhinoceros 6 software, leveraging data from the Geoportal.gov.pl web portal [61]. To evaluate the tool, various simplified versions of a multifamily building were modeled. Each time after modeling an updated version of the building or changing its parameters, carbon footprint analysis was performed automatically using the tool. Based on information on the expected carbon footprint, decisions were made regarding changes in the form and parameters of the building. In total, 15 different alternatives were created to find the best solution [56]. Figure 4, Figure 5 and Figure 6 show some of these variants.

The program estimated the carbon footprint of the initial building alternative at 15.18 kgCO₂eq/m² (Figure 4). The different alternatives for the building related to different building shapes, locations on the plot, and construction technology systems [56]. The 3 selected variants that have been presented in this article were based on 3 different construction technologies: brick and concrete frame (Figure 4), a prefabricated concrete system (Figure 5), and a cross-laminated timber system (Figure 6). The final variant had the lowest carbon footprint of all the analyzed alternatives at 11.29 kgCO₂eq/m² (Figure 6), which represents a reduction of approximately 25% in terms of the carbon footprint compared to the initial variant. The design process was supported by analyses performed using a prototype tool to find the shape and parameters of a building with a much lower carbon footprint than the initial assumptions.

3.3. Conclusions of Case Study 1

This tool helps to lower the building’s carbon footprint early in the design process when information is scarce. Carbon footprint values are less accurate, but the benefit is that one can make the analysis early in the design process, as verified in the previous study performed by the authors [55]. However, some current limitations of the tool were observed: it is limited to only the Rhinoceros 6 program, and there are some limitations regarding building typology and size (only multifamily buildings in a 1200 to 3200 m² area). The climate locations and construction technology used in the training set were also limited to three and four variants, respectively [56]. Another limitation is the use of a simplified building model with zero element thickness for energy simulations. This approach cannot take into consideration more detailed decisions. Training one’s own machine learning model requires vast amounts of data, and enlarging the scope into more variable sizes, typologies, and locations would require a larger dataset. This would be impossible to achieve by self-generating the training examples. There are two possible further routes: either using datasets from different users and generating bigger datasets with various examples or using pretrained models. The authors decided that the best approach would be a mixed one: an experiment using a pretrained model while allowing tool users to submit their calculations to create a larger dataset. The data users submit may vary significantly in accuracy, precision, and format. This inconsistency can introduce noise into the dataset, complicating the training and fine-tuning processes of the machine learning model.

4. Case Study 2. LLM-Assisted Web Carbon Calculator Tool

4.1. Materials and Methods

Web-based applications for architects have become increasingly popular, improving user productivity [62]. There are numerous examples of web tools that allow designers to move their workflow from desktop to the cloud (fx. Speckle, Shapediver, Viktor) [63,64,65]. This allows for greater flexibility, easier data sharing, and faster collaboration in complex AEC industry problems.

Leveraging the power of deep learning techniques, ChatGPT processes and responds to user queries in a coherent and contextually appropriate manner. The model is a multi-layer neural network that uses sophisticated algorithms to analyze and learn patterns from input data. By predicting the next word in a sentence based on the preceding context, ChatGPT becomes proficient in grammar, syntax, and semantic relationships between words. This proficiency allows ChatGPT to generate responses that are not only linguistically accurate but also contextually relevant. LLMs can be used as virtual assistants, suggesting optimization possibilities and changes to architectural design. Currently, the input to LLM needs to be in the form of text. By inputting a text that describes a part of a building, the designer can expect to receive recommendations on how to reduce the carbon footprint of the design solution. However, it is important to recognize the limitations of LLMs like ChatGPT. While they excel at generating language-based responses, their knowledge is constrained by the training data they have been exposed to. As such, it is essential to critically evaluate and verify the information provided by ChatGPT. Furthermore, low-carbon architectural design requires a holistic understanding of numerous factors beyond textual knowledge, such as material science and engineering principles, which may not be fully captured by ChatGPT alone.

This case study served as a proof of concept, and it is recommended that LLMs be developed specifically for AEC to fully realize its potential and reduce mistakes. The application of LLMs should be complemented with expert knowledge and careful consideration of other multidisciplinary aspects involved in low-carbon architectural design.

To utilize the power of LLMs in architectural design, an online carbon calculator tool was created (Figure 7). It was developed in JavaScript and consists of basic functionalities required to calculate carbon footprint. The tool uses the Ökobaudat database of emission factors of different building materials [58]. A user can create an account by visiting the web page www.slad.ai (accessed on 1 April 2024), and then create a building model [66]. In the next step, the designer should place all the building components by creating their material composition. The user has access to the Oeakobau.dat library or can add information about the material from an EPD (Environmental Product Declaration). Finally, the quantity of the given component is specified. By inputting information on building elements and selecting the correct corresponding Oeakobau.dat materials, a full building model can be created (Figure 8 presents values of example materials, and GWP/unit presents the sum of A1–A3, C3, and C4 lifecycle phases). The tool then calculates the carbon footprint of the building in the A1–A3 and C3–C4 phases—the embodied carbon footprint (Figure 9). If the designer enters the results of an energy simulation from another application, the operational carbon footprint can also be calculated.

4.2. Implementing AI into the Carbon Calculator Tool

Using ChatGPT’s OpenAI API (Application Programming Interface), the SLAD.AI website was connected to the ChatGPT LLM gpt-3.5-turbo-instruct [67]. After specifying the building model in the application, the user can ask ChatGPT for suggestions. The application sends the building description to the ChatGPT server and waits for a response. The response is then displayed to the user.

4.3. Results

The tool was evaluated in numerous workshops and classes with university students (university course entitled “Case Study”, December 2023–January 2024; workshops “Carbon Footprint for Architects”, October–November 2023; Workshops “ESG Academy” May 2023, March 2024). It proved useful in clarifying the carbon footprint estimation process. Users commented that the app has helped them to understand the process of carbon footprint calculation and that the UI of the tool has been developed in a logical manner that is easy to understand. Using the LLM model also helped users choose better solutions. In some cases, the AI found a better material and suggested it to the user (Figure 10 and

Table 2. A comparison of correct and incorrect suggestions provided by ChatGPT (own elaboration).

Component	ChatGPT Suggestion (Correct)
Plaster	Clay-based plaster
Mineral Wool	Straw insulation
Light Concrete Blocks	Energy-efficient (insulated) Light concrete blocks
Fiber Cement Slate	Slate tiles
Component	ChatGPT Suggestion (incorrect)
Plaster	Plaster
Light Concrete Blocks	Light concrete blocks
XPS	Plasterboard
Fiber Cement Slate	Fiber cement slate

). In the example, the AI suggests fx. using straw insulation instead of mineral wool material, resulting in lowering the carbon footprint of the component (mineral wool: 33.58 kgCO₂eq/m³; straw insulation: 12.31 kgCO₂eq/m³) [58]. In other cases, the AI model did not understand the architectural design well enough and suggested replacements with a lower carbon footprint but that were incorrect in terms of building technology (Figure 11 and Table 2).

This clearly shows the potential of using LLMs in the optimization process, as well as the importance of building AI models with domain knowledge.

4.4. Conclusions of Case Study 2

The web application allowed users to use the tool from any desktop computer or mobile device, ensuring greater flexibility and allowing greater adoption. The app’s logic flow was changed, as the tool is now based on user input in the form of descriptions of all building components.

AI was implemented to help users improve building component design. The numerical results show that AI material suggestions can lower the carbon footprint of the components of the building.

Importantly, this tool version can analyze any building of any size. The previous limitations of size or typology are gone. However, inputting the information about the building takes more time than in the previous version. Therefore, there should be a more automated manner of inputting the building information so that it would be possible to leverage the benefits from case studies 1 and 2 together.

5. Case Study 3. Integrating BIM, Carbon Footprint Analysis, and AI into a 3D Application

5.1. Materials and Methods

BIM is becoming a standard way of approaching the complex process of architectural design. However, proprietary data formats associated with specific applications often cater to particular use cases and may not fully accommodate the diverse needs for data exchange and detailed information required in BIM scenarios. IFC (Industry Foundation Classes) is a common exchange format for BIM applications [68]. IFC serves as a standardized digital representation of the construction and infrastructure sectors. Operating as an open and globally recognized standard (ISO 16739-1:2018) [69], it advocates for vendor-neutrality and versatile functionality across various hardware, software, and interfaces, catering to diverse applications. BIM and IFC have already been proven useful as tools that help in faster material quantity estimation, environmental optimization, and thermal analyses [70,71,72]. Zhang et al. proposed an automated tool that helps designers estimate the carbon footprint of their designs based on the BIM IFC model [73]. A number of similar solutions exist on the market. The idea of creating web apps with BIM technology is not new [74]. There is a growing list of examples of implementing AI within BIM technology; however, there is still a need for more research in this area [75,76].

The authors developed a web application (published on GitHub) that integrates BIM, IFC, AI, and carbon footprint calculation in a single tool (Figure 12). The tool loads an IFC model and creates a material takeoff. A user can then apply material combinations to the building components. Finally, the user can calculate the carbon footprint of the building and ask AI for suggestions on how to improve the design.

The application was developed in an open-source way—all the code is available for download online from the GitHub repository [77]—allowing users to create their own versions of the application by modifying components inside. Users can also collaborate in the app’s design by submitting “pull requests”—items that need to be changed. The app uses the Open BIM Components library [78], an open-source repository for parsing 3D IFC files in the browser based on the Three.js library [79]. The application was programmed using Typescript [80] and the React framework [81]. Like the previous case study, material data were imported from the Ökobaudat database [58].

5.2. Typical Workflow for Users

A user would typically upload a building model in the form of a file of the IFC 2 × 3 type [82]. The IFC file is then saved to the database, and a bill of quantities is automatically generated. The next step includes the designer predefining building components by selecting the material and quantity for each layer of the component (Figure 13). In the example, the AI suggests replacing a few materials, for example, using cellulose insulation instead of EPS material to lower the carbon footprint—EPS: 118.4 kgCO₂eq/m³; cellulose fiber insulation: 12.31 kgCO₂eq/m³ [58].

The predefined components can then be assigned to the actual building elements in either the 3D view or the bill of quantities view. Finally, the user can check the result of the carbon footprint calculation based on the input data. The user can also request help from AI, where the description of the selected building components is sent to the AI agent via API, a similar solution as in case study 2. The application uses the ChatGPT LLM gpt-3.5-turbo-instruct model [67]. In the last step, a suggestion is shown to the user regarding the materials that could be changed to reduce the carbon footprint of the design.

5.3. Results

The final version of the tool was evaluated using an example project [77]. The project file was modeled in Autodesk Revit 2020 and then exported to the IFC 2 × 3 format (Figure 13). The components were defined inside the application using data from

Table 3. Technical details of the building components (own elaboration).

Exterior Wall	Exterior Finish—Plaster [2 cm]
	Thermal insulation—mineral wool [20 cm]
	Structure—precast reinforced concrete [12 cm]
	Interior finish—plaster [2 cm]
Roof	Roof cover—EPDM [1 layer]
	Substrate—screed [4 cm]
	Thermal insulation—EPS [25 cm]
	Structure—precast reinforced concrete [12 cm]
	Interior finish—plaster [2 cm]
Ground Floor	Finish—wood [2 cm]
	Substrate—screed [5 cm]
	Insulation—EPS [15 cm]
	Structure—reinforced concrete [15 cm]
Interior Wall	Structure—precast concrete wall [5 cm]
	Insulation—mineral wool [7 cm]
	Structure—precast concrete wall [5 cm]
Exterior slab	Structure—reinforced concrete [15 cm]

. Then, in the 3D view, the predefined components were applied to the 3D elements of the building based on the automatic quantity takeoff created by the application. The application automatically calculates the carbon footprint of the building based on the selected component and the automated quantity takeoff (Figure 14). The results are presented in a chart (Figure 15). In the test, the embodied carbon footprint was calculated for the external wall, roof, internal wall, ground floor, external slab, windows, and doors based on the IFC model. The calculated carbon footprint was 64,120 kgCO₂eq.

5.4. Conclusions of Case Study 3

The integration of BIM, LCA, and AI in a web-based application using IFC files significantly advances sustainable architectural design. This study resulted in a prototype calculator that merges BIM and AI technology to estimate buildings’ carbon footprints. This tool improves the accuracy of environmental impact assessments and simplifies the design process by leveraging 3D models for comprehensive analyses. Its internet-based platform fosters global collaboration, allowing real-time interaction and decision making among project team members.

Key to its innovation is the automated processing of building information, which reduces manual data entry and focuses efforts on sustainability and design creativity.

The AI-driven suggestions for material and design optimizations further bridge the gap between complex data analysis and actionable design strategies. The assistance of the AI is also shown in the numerical results, which prove that this addition can lower the carbon footprint of the building’s components.

Despite its advancements, this application highlights the need for further automation in data extraction from BIM models to improve efficiency and the user experience. In summary, this tool represents a significant leap toward accessible, efficient, and sustainable design practices by integrating innovative technologies in a user-friendly, collaborative environment.

6. Discussion

While AI and other technological tools offer substantial support to architects and designers, enhancing the efficiency of their designs, they still fall short of replicating human intuition and emotional intelligence. The “Technology At Work v2.0” report [52] highlights how technology, despite its advances, will reshape job landscapes in the developing world by automating tasks that require less emotional, creative, and interpersonal input, yet it also emphasizes the irreplaceable value of human expertise in roles demanding intricate problem-solving and personal engagement, such as in architecture and service design (Hoque, 2024) [51]. This underscores the necessity for designers to evolve alongside AI technologies to maintain their relevance in the field. Exploring synergies between human designers and AI models becomes crucial in this context.

The discussion around incorporating AI, BIM, and LCA tools into architectural design underlines the potential benefits and existing challenges. While AI can aid in the decision-making process, for example, in LCA [32], improving design efficiency, the journey toward seamlessly integrating these technologies faces hurdles like data accessibility issues in the construction industry and the need for increased model accuracy to enhance the reliability of LCA studies derived from BIM models [18,21,22]. Developing industry-wide standards for data formatting and interoperability between different BIM and LCA tools could also help support data accessibility. The promotion of open-source and open-data practices, along with the development of domain-specific AI models trained on sustainability, BIM, and LCA datasets, is essential for overcoming these challenges. Additionally, open-source platforms could gather trained models and help to distribute them.

As the architectural field moves toward a more sustainable future, the enhancement and integration of AI-driven tools with BIM and LCA methods appear paramount. This evolution not only aims to streamline the design and engineering workflow but also seeks to minimize the workload associated with LCA studies, thereby allowing professionals to devote more time to creative endeavors. The path forward involves refining these AI tools for better integration into architectural practices, highlighting the collaborative future between human creativity and machine intelligence [83].

Looking forward, several areas of further work have been identified to enhance the integration of AI, BIM, and LCA tools in architectural design for sustainability. The development of larger datasets related to sustainability, BIM, and the carbon footprint can support further research work. The expansion and refinement of these datasets are crucial for improving the accuracy and applicability of AI models in the field of architecture and sustainability [19,20]. In addition, there is a need for more interdisciplinary research to understand the full potential and limitations of integrating these technologies.

7. Conclusions

The integration of AI, BIM, and LCA tools in architectural design has been shown to significantly advance the pursuit of low-carbon design solutions. AI-driven tools, as discussed in the case studies, provide valuable insights that can guide architects in optimizing design parameters for sustainability, for example, providing information about carbon footprint without a finished architectural model or guiding the designer by suggesting better variants.

The three case studies presented explore various aspects and applications of integrating AI and machine learning into architectural design processes, specifically focusing on carbon footprint estimation and optimization.

Case study 1 developed a tool within Rhinoceros 6 and Grasshopper using machine learning to predict building carbon footprints during the architectural design phase. This tool demonstrated the potential to enhance early design decisions by linking building shapes to their carbon footprints. The development involved generating a simulated dataset of multifamily buildings in Poland due to the absence of such data. Although the tool showed potential in reducing the carbon footprint (achieving a 25% total carbon reduction in the example case), limitations were noted in its dependence on specific software, building types, locations, and building sizes. Future enhancements could involve using larger, varied datasets or pretrained models to increase versatility and accuracy.

Case study 2 involved a web-based tool that utilizes large language models (LLMs) such as ChatGPT to assist in architectural design optimizations for reduced carbon footprints. The tool, developed using JavaScript, accesses material emission factors and provides users with AI-driven suggestions to modify building designs for better environmental performance. The suggestions delivered by the AI lowered the carbon footprint of the building components. While the LLM facilitated better material choices, it became clear that one of the limitations is the fact that, sometimes, the model fails to comprehend architectural contexts properly. The freedom to use the tool on any device highlighted its flexibility but also showed the need for more streamlined data input methods.

Case study 3 presented a web application integrating building information modeling (BIM), Life Cycle Assessment (LCA), and AI to create a comprehensive tool for calculating and reducing the carbon footprint of buildings. The recommendations provided by the AI reduced the carbon footprint of the building material selection. This tool leverages IFC files for detailed modeling and automated data processing, enhancing the accuracy and efficiency of environmental impact assessments. It emphasizes the importance of automation in extracting data from BIM models to improve user experience and sustainability outcomes. Future work should focus on developing custom LLMs that better understand the architectural context and carbon footprint information.

The automation of complex analyses and calculations frees up architects and designers to focus on innovation and creativity. With AI-driven tools managing the technical aspects of sustainability, designers can explore new materials, forms, and construction techniques that might have been too resource-intensive to consider previously.

While the benefits are clear, challenges such as data accessibility, the need for domain-specific AI models, and the integration of these tools into existing workflows must be addressed. Overcoming these challenges presents an opportunity to not only enhance the sustainability of individual projects but also to drive systemic change within the architecture and construction industries.

By incorporating these tools early in the design phase—at the concept stage—architects and designers can make informed decisions that lead to more sustainable outcomes. Early in the design process, the possibility to modify the design is usually very high, while it decreases in the later stages [54]. AI-driven tools, as discussed in the case studies, provide valuable insights that can guide architects in optimizing design parameters to support sustainability. These tools facilitate a more data-driven approach to design, enabling the exploration of various scenarios and their environmental impacts before final decisions are made. It is important to note that the use of AI and digital tools not only speeds up the design process by automating complex calculations but also encourages innovation through the exploration of new materials, forms, and construction techniques. This approach allows for a more creative yet environmentally conscious architectural design exploration.

The case studies also highlight the role of these tools in educating designers about the environmental implications of their decisions. By making sustainability metrics easily accessible and understandable, these tools promote greater awareness and commitment to sustainable design principles among the next generation of architects.

In conclusion, the path to sustainable architecture is complex and multifaceted, requiring a holistic approach that integrates AI, BIM, and LCA tools into every stage of the design process. Integrating these technologies facilitates greater collaboration among architects, engineers, environmental scientists, and other stakeholders. By working together from a common technological platform, teams can synchronize their efforts toward achieving the most sustainable outcomes. By harnessing these technologies, architects and designers can lead the way in reducing the environmental impact of buildings, contributing to a more sustainable and resilient built environment.