Nina Isabelle Onia
Professor Doris Huber
ENGL 109W-03
12 October 2022
Annotated Bibliography Summaries
Elshafei, Ghada, and Abdelazim Negm. “AI Technologies in Green Architecture Field: Statistical Comparative Analysis.” Procedia Engineering, vol. 181, 2017, pp. 480–88, https://doi.org/10.1016/j.proeng.2017.02.419.
Elshafei and Abdelazim introduce different forms of artificial intelligence (AI) and how they can be applied to green design and ensure that seemingly unrelated variables in a project will not conflict with each other. These forms include Genetic Algorithms (GA) Analytical Hierarchy Process (AHP), and Fuzzy. Variables include energy efficiency and types of renewable energy. A GA analyzes the fitness and wellness of each member of a population and predicts their odds of reproducing. This tool has been used to determine a project’s optimal lighting and thermal system. The AHP is known for finding the most efficient solutions to problems within a project based on facts and math. It does so by compiling a database, analyzing data, and ranking the attributes based on priority. The Fuzzy-Delphi method has been used in tandem with AHP, using a variety of tools and an evaluation model to determine the best course of action for a project. Fuzzy is used mainly for sustainability (carbon footprint, materials, etc.)
Elshafei and Abdelazim then compare and contrast the three AI. GA is a flexible tool that is most often applied to construction but relies solely on the input of population data to function properly. AHP and Fuzzy are both most often applied to environmental, logistical, and energy-based problems. However, AHP is better used for making simpler decisions while Fuzzy can account for unknown factors of a problem.
Elshafei and Abdelazim's paper is important to the topic of AI in architecture because they discuss various AIs that have been applied to architectural work. Each type of AI has its own section in the text that defines how the AI functions and what jobs it's been used for most often; specific examples of the latter are only mentioned briefly which means further research is needed. This is important because it clearly demonstrates that "AI" come with algorithms that are unique to each variety and can be programmed for specific jobs. This answers my research question “what tasks can artificial intelligence carry out to help architects create pre-design, development, and construction documents?” Through this evaluation of AI, Elshafei and Abdelazim prove that AI helps architects plan ahead so they know what to design. This source clearly states that various AIs help architects by providing information and solutions for the architect to apply to their work.
Florian, Maria-Cristina. “‘This House Does Not Exist’ Uses AI to Generate Images Inspired by ArchDaily's Modern Architecture Projects.” ArchDaily, ArchDaily, 7 Sept. 2022, www.archdaily.com/988606/this-house-does-not-exist-uses-ai-to-generate-archdaily-style-images-of-modern-architecture.
Florian’s article discusses how AI—a resource accessible to the public via image-generating websites—is a tool that professionals should use with caution. The titular site “This House Does Not Exist” is automated; with only a button push, the AI uses a database of keywords and images of architectural projects from the architectural publishing company ArchDaily to create images of modern houses. Florian states that architecture should not be automated by replacing human architects with AI. Instead, she states that professional architects and engineers should use AI as augmentation: improvement upon human workers as they and their tools learn to design more efficiently. Florian also provides insight from architect Michael Hansmeyer who claims that AI could also generate a process of designing itself. This would allow architects to better understand the compiled data and brainstorm alternative ways to develop their projects.
Florian’s article is important to the topic of AI in architecture because it covers how AI is used in architecture in simple terms. She defines terms that may be hard to understand (augmentation, automation, etc.) and smoothly transitions from a casual perspective to a hypothetical professional one to highlight the usefulness of AI as an architectural tool. This article also answers both of my research questions: “what tasks can artificial intelligence carry out to help architects create pre-design, development, and construction documents?” and “is there a significant difference in workflow and output between an AI-aided project and a typical project?” Although she does not mention specific examples of AI used by professional architects, Florian and Hansmeyer claim that AI is a tool that can speed up an architect’s workflow through problem-solving and fast data management. The beginning of the article about AI-operated websites might not be covered in my literary review because those examples are not applied to professional architectural works.
Ploszaj-Mazurek, Mateusz, et al. “Methods to Optimize Carbon Footprint of Buildings in Regenerative Architectural Design with the Use of Machine Learning, Convolutional Neural Network, and Parametric Design.” Energies (Basel), vol. 13, no. 20, 2020, p. 5289–, https://doi.org/10.3390/en13205289
Ploszaj-Mazuerk claims that certain strategies can remedy design practices that result in a project with high carbon emissions (production, construction, use stage, and end of life). Regenerative Design has been used in urban design to reuse resources within a project based on the conditions of the site and city. An architectural design process prompts architects to plan ahead to reduce their project’s carbon footprint based on its materials and parameters. Architects have also used AI as a means to gather and analyze information. However, Ploszaj-Mazuerk notes that Machine Learning (ML) is a work in progress when dealing with carbon emission issues. The focus shifts to three studies of ML using a parametric model of a multi-family building. For the first study, researchers put the model into thousands of simulations to randomly generate variations of the building, focusing on the designs that would result in the lowest carbon emissions. This was based on the dimensions of the building, the materials of the components (walls, floor, roof, etc), and the environmental factors (wind, sunlight, etc). The resulting ML-generated building was a close match to the design established by the researchers. For the second study, the model was provided with more specific parameters but a wider variety of shapes the building could take. The ML predictions for embodied and operational carbon footprints were also close to the researchers’ results. The final study uses a neural network—a form of ML—to analyze various sites in which the model would reside alongside the established parameters. This network predicted the total carbon footprint of the project with a small marginal error. These studies exhibit that ML can be a valuable tool for architects. This AI will not work properly if the user does not provide proper data. Ploszaj-Mazuerk concludes that an AI needs data from the real world to properly simulate a project.
Ploszaj-Mazuerk’s paper is important to the topic of AI in architecture because it recounts studies specifically to test the proficiency of AI in green design. This paper answers my research question “what tasks can artificial intelligence carry out to help architects create pre-design, development, and construction documents?” By running multiple studies, the researchers determined that an AI can produce results and reliable solutions as long as it receives the appropriate input from architects and engineers on the development and construction of the project. This paper does not necessarily answer if AI is used for pre-design because the studies involved a model building rather than an actual project that an architect conceptualized and provided rough drawings for. This paper also answers my other research question— “is there a significant difference in workflow and output between an AI-aided project and a typical project?”—to an extent. By comparing the AI’s calculations to a human worker’s calculations, the paper implies that an AI can think like a human but produce results faster.
Rochd, Abdelilah, et al. “Design and Implementation of an AI-Based & IoT-Enabled Home Energy Management System: A Case Study in Benguerir — Morocco.” Energy Reports, vol. 7, 2021, pp. 699–719, https://doi.org/10.1016/j.egyr.2021.07.084.
Rochd begins by establishing that high carbon emissions are an international problem that can be remedied. Solving this problem is difficult considering the limits set by a country’s economy and available technology. A decentralized solution should allow workers of a project (architects, construction workers, engineers, etc.) to better determine a means to solve problems and save energy. AI specifically does so using neural networks and variations of Fuzzy.
Rochd specifically covers the use of a Home Energy Management System (HEMS) in the Smart Campus of the Green & Smart Building Park (GSBP) in Benguerir, Morocco. The HEMS the GSBP researchers used features a database made by a Monitoring Module and converted into a Human-Machine Interface. Data includes flexible and critical loads of household appliances, the former capable of adjustments that will not strongly affect the comfort level of the people in the household (air conditioning, washing machine, water heater, etc). The HEMS’ objective is to plan its architecture to establish comfort while keeping costs low. The supply side of the strategy focuses on costs while the demand side focuses on comfort. The AI has to apply the data to an algorithm with several formulas to balance supply and costs with demand and comfort. This led to the establishment of Real Time Pricing (RTP), Time of Use (ToU) pricing, and fixed tariff. RTP and ToU could save users money depending on how often they use their appliances throughout their life.
The study in Rochd’s paper is important to the topic of AI in architecture because it focuses on AI solving problems in green design. Rochd’s claim that AI is a helpful tool is justified to be flexible for architects using discussions, figures, and graphs throughout the paper. This paper answers both of my research questions—“what tasks can artificial intelligence carry out to help architects create pre-design, development, and construction documents?” and “is there a significant difference in workflow and output between an AI-aided project and a typical project?”—somewhat. Based on the information in the paper, an AI is more geared toward the engineers and users of a project if programmed to focus on the electrical system. However, an architect should be aware of what engineers and clients need from them so their design can accommodate electronic equipment.
Tsigkari, Martha, et al. “Towards Artificial Intelligence in Architecture: How Machine Learning Can Change the Way We Approach Design.” Foster + Partners, Hennick Company, 29 Mar. 2021, www.fosterandpartners.com/plus/towards-artificial-intelligence-in-architecture/.
Tsigkari covers how Foster + Partners’ Applied Research and Development (ARD) team analyzed the capabilities of AI. In architecture, AI should be programmed to resemble the human thought process in a way that saves architects time and money on design. The ARD team focused their research on surrogate models and design-assistance modeling.
Tsigkari first notes how programming an artificial neural network into an AI is similar to teaching a child: a cycle of giving data and feedback to gradually improve its work until it can work by itself. To achieve this in architecture, programmers compile a database, allowing the AI to establish rules to replicate a specific style and understand how the popularity of a project equates to a subjectively appealing design. Original data can include the works of an architecture firm while synthesized data is a result of AI creating its own imagery based on provided original data.
Tsigkari then recounts the research projects of the ARD team involving parametric models. The first was a collaboration with Autodesk in which they studied thermo-active laminates. They used the software Hydra to take a database of parametric models that resemble those laminates and predict how the material will respond to varying levels of heat over the course of its life. The ARD team also experimented with parametric models of workspace floor plans. They programmed an AI with this database to quickly generate sample floor plans and analyze how the spaces of the plans connect. The ARD team plans to continue tinkering with AI to provide Foster + Partners with quick, precise tools.
Tsigkari’s article is important to the topic of AI in architecture because Tsigkari discusses how researchers that work at the same architecture firm as she does want to implement AI as a tool in their firm’s future works. This research demonstrates that a firm can easily program an AI because the firm has several resources that can be compiled into a database. This clearly answers both of my research questions: “what tasks can artificial intelligence carry out to help architects create pre-design, development, and construction documents?” and “is there a significant difference in workflow and output between an AI-aided project and a typical project?” According to Tsigkari, AI can be programmed to produce outputs for various topics (material, style, weather, etc.). This flexible tool can reduce the time an architect could spend planning, designing, and polishing.