Artificial Intelligence as a Tool for Architectural Design: Comparison
View Latest Version
Please note this is a comparison between Version 2 by Catherine Yang and Version 1 by ZHIHUI ZHANG.

Artificial intelligence (AI)-generated designs demonstrate significant potential, exhibiting competitive results in the categories of Attractiveness and Creativity. AI faces challenges in replicating the distinctive aspects of human design styles, pointing to the innate subjectivity inherent to design evaluations. These findings shed light on the role AI could play as a tool in architectural design, offering diverse design solutions and driving innovation.

  • AI-generated image
  • architectural design
  • stable diffusion
  • perception

1. Introduction

The advent of artificial intelligence (AI) has triggered significant transformations across various disciplines, thanks to its computational capabilities that enhance traditional methodologies and foster innovative approaches. Architecture stands out as one of the fields where the transformative potential of AI is being leveraged, ranging from the conceptualization to the execution of designs [1]. The integration of advanced AI techniques such as Generative Adversarial Networks (GANs) [2], Latent Diffusion Models (LDMs) [3], and Segment Anything Models (SAMs) [4] into architectural software underscores the scope of AI’s application in this domain. These techniques enable AI to generate a diverse array of design alternatives, optimize structural components, and even emulate the stylistic subtleties of esteemed architects [5].
In addition to its role in design generation, AI is progressively recognized as a transformative educational tool in architecture. Beyond traditional classroom methods, AI offers an enriched learning experience, with its capacity to visualize unbuilt designs and understand design principles [6,7][6][7].

2. AI as a Creative Design Tool for Generating Alternatives

Artificial intelligence (AI) has progressively become a fundamental part of architectural design, pushing the envelope of what is possible and transforming traditional design methodologies into innovative, future-facing ones [11][8]. As the computational power continues to advance, a paradigm shift where machine learning has become a pivotal tool in architecture [12,13][9][10]. Previously, the integration of machine learning into architectural design tools was limited due to the complex and creative nature of design tasks. However, with the rise and integration of more advanced machine learning models, such as transformer models, into design workflows, this barrier is gradually being overcome [14][11]. Artificial neural networks, drawing inspiration from biological neural networks [15[12][13],16], have been key players in transforming the design space. By training these networks on specific examples, known as the training set, in the form of input parameters and corresponding output values, they learn and iterate on design solutions [17][14]. Some cases, which employed neural networks to generate innovative design alternatives in architectural planning, demonstrate how AI can broaden the creative possibilities [18][15]. Moreover, the concept of swarm intelligence, as elucidated by Bonabeau, Dorigo, and Theraulaz [19][16], has seen a transition from being a phenomenon observed in natural systems to a technique applied in artificial systems, specifically architectural design. It emulates collective behaviors observed in nature, such as bird flocking or insect swarming, and has been harnessed to produce complex spatial forms [20,21][17][18]. Swarm intelligence can be leveraged to optimize energy usage in building design, offering an approach that harmonizes design aesthetics with sustainability. In the more recent years, AI models such as Generative Adversarial Networks (GANs) and Latent Diffusion Models (LDMs) have further expanded the horizon of creative possibilities [22][19]. These generative AI models, which leverage vast databases for initial learning, have an inherent level of decoding uncertainty. This can lead to the generation of diverse (see Figure 1), unconventional patterns and solutions, pushing the boundaries of conventional design thinking.
Figure 1.
Due to the characteristics of Latent Diffusion Models, it is possible to generate diverse results.
The advent of AI has laid the groundwork for a revolution in architectural design by offering an entirely new realm of possibilities and paths for exploration. The integration of automated design systems not only streamlines the design process but also bolsters the role of conceptual thinking in crafting solutions [11][8]. Yet, the human element in the process—the role of the architect—remains indispensable. Architects bring a critical perspective in selecting the most suitable solution from a multitude of scenarios generated by AI, ensuring the blend of creativity and functionality in the final design [11][8].

3. Potential of AI in Architectural Education

The utilization of artificial intelligence (AI) in architectural design and education has made substantial progress, transitioning from rudimentary design tools to powerful instruments capable of generating innovative design solutions, optimizing existing designs, and playing a pivotal role in education [23,24][20][21]. These advancements have pervaded all facets of architectural education, including technical, theoretical, representation, and design studio modules. Technically, the impact of AI is manifested in the use of building information modeling (BIM) and parametric design software, tools proficient in generating 3D spatial data to enhance the design and construction process [25][22]. Meanwhile, the application of machine learning (ML) requires customization for each project, with data collection, preprocessing, and computational power being crucial elements [23][20]. In theory-based courses, the employment of AI, particularly in collecting, storing, and analyzing massive amounts of textual data, significantly alleviates the students’ workload [26][23]. This aligns with Negroponte’s research suggesting that machines can learn architectural design via sampling and evaluation, bypassing the need for pre-encoding rules, offering unique opportunities for architectural education [27][24]. Moreover, deep learning (DL) models, with the support of big data, have demonstrated the capacity to tackle architectural design problems [28][25]. Remarkably, Generative Adversarial Networks (GANs) have been applied to create architectural layouts, generating an abundance of architectural floor plans even without a vast quantity of image data [18,29][15][26]. In representation modules, digital technologies such as Virtual Reality (VR), Augmented Reality (AR), and 3D printing have revolutionized spatial perception and design presentations [7]. AI, employed as a creative design tool, is starting to transform the communication process between architects and clients. For instance, AI image generation tools such as DALL-E can swiftly articulate architects’ design intentions, reducing the architects’ workload, and stimulating new creative thinking. In addition, novel AI technologies such as Generative Adversarial Networks (GANs), Latent Diffusion Models (LDMs), and Any Segment Models (SAM) are beginning to be incorporated into specific software tools. Tools such as Stable Diffusion V.5, Midjourney V5.1, and Photoshop 2023 (Beta) are being integrated into design studio workflows, with some studios even starting to train their algorithm models [30,31,32][27][28][29]. This rapid visual communication bears immense potential in architectural education as it fosters more effective communication between students and teachers during the design stages. Design studios, viewed as the core of architectural education, serve as a confluence of theoretical and technical knowledge. Here, AI assumes a crucial role in data processing, research object indexing, environmental analysis, and suggesting design proposals through building performance analysis tools [7]. Innovative tools such as the Nuncias chatbot have been integrated into architectural education to aid in enhancing the verbal definition of designs [33][30]. However, the application of AI in architectural design and education is not devoid of challenges. Deep learning models and extensive datasets cannot mimic human ways of thinking, such as “common sense”, i.e., the ability to generalize, create, and simulate abstract information [28][25]. Thus, the progression of AI needs to coincide with enhancements in educational models and strategies to adapt to the rapid technological advancements and cultivate designers capable of effectively utilizing these tools [24][21]. The integration of AI in architectural design and education equips students and designers with potent tools to explore novel design methods and optimize existing designs. Nonetheless, to effectively incorporate AI, appropriate educational models and strategies need to be developed to adjust to the rapid technological advancements.

4. Assessment of Aesthetic Value in AI-Generated Architectural Design

The rapid application of artificial intelligence in architectural design is catalyzing significant changes, including the rise of many innovative design approaches and practices.  Evaluating the aesthetic value of architectural design is a complex task encompassing multiple dimensions. It is also important to acknowledge that architecture, although primarily captivating through visual expression, is a multisensory experience incorporating elements such as touch, sound, and even smell. This awareness of the multisensory nature of architecture aligns with the Enlightenment period’s emphasis on prioritizing the senses of sight and hearing over the senses of smell, touch, and taste [35,36][31][32]. Contemporary theories, as proposed by Bille and Sørensen, suggest a further extension of this awareness, highlighting the atmospheric elements, processes, and practices in architectural design [37][33]. Mehaffy, Gorichanaz, and Lavdas offer comprehensive discussions on architectural form, user experience, and sociocultural aspects, which furnish invaluable insights into understanding and evaluating the function of AI in architectural design [38,39][34][35]. Simultaneously, aesthetic evaluation straddles both subjectivity and objectivity. Fechner’s endeavor to quantify aesthetics with numerical scales provides a potential approach to aesthetic assessment. This concurs with Christopher Alexander’s notion of “Quality Without A Name” and Buras’s “beauty scale” theory, both stressing the importance of quick perception in aesthetic evaluation [41,42][36][37]. The efficacy of using a beauty scale for evaluation can reach 80–90% in rating consistency, suggesting that it can serve as an efficient instrument to predict and comprehend people’s aesthetic experience of architectural design, and achieve consensus in the design and execution process [43][38]. Assessing the aesthetic value of AI-generated architectural design requires contemplating human perceptual experience, subjective feelings, memory, social and cultural demands, and aesthetic responses from a neuroscientific perspective. Among these, human intuitive responses and perceptual experiences play a pivotal role in the evaluation. In assessing the architectural image generation ability of AI, subjective scales can be employed for experimentation.

References

  1. As, I.; Basu, P. The Routledge Companion to Artificial Intelligence in Architecture; Routledge: Abingdon, UK, 2021.
  2. Pan, Z.; Yu, W.; Yi, X.; Khan, A.; Yuan, F.; Zheng, Y. Recent Progress on Generative Adversarial Networks (GANs): A Survey. IEEE Access 2019, 7, 36322–36333.
  3. Pinaya, W.H.; Tudosiu, P.D.; Dafflon, J.; Costa, P.F.D.; Fernandez, V.; Nachev, P.; Ourselin, S.; Cardoso, M.J. Brain Imaging Generation with Latent Diffusion Models. In Lecture Notes in Computer Science (Including Subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Springer: Cham, Switerland, 2022; Volume 13609, pp. 117–126.
  4. Kirillov, A.; Mintun, E.; Ravi, N.; Mao, H.; Rolland, C.; Gustafson, L.; Xiao, T.; Whitehead, S.; Berg, A.C.; Lo, W.Y.; et al. Segment Anything. arXiv 2023, arXiv:2304.02643.
  5. McCullough, M. Digital Ground: Architecture, Pervasive Computing, and Environmental Knowing. Digit. Ground 2004, 3, 290.
  6. Maher, M.L.; Bilda, Z.; Gül, L.F. Impact of collaborative virtual environments on design behaviour. In Design Computing and Cognition ’06; Springer: Dordrecht, The Netherlands, 2006; pp. 305–321.
  7. Ceylan, S. Artificial Intelligence in Architecture: An Educational Perspective. In Proceedings of the 13th International Conference on Computer Supported Education—Volume 1: CSEDU, INSTICC, SciTePress, Online, 23–25 April 2021; pp. 100–107.
  8. Cudzik, J.; Radziszewski, K. Artificial Intelligence Aided Architectural Design. In Proceedings of the eCAADe 2018: Computing for a Better Tomorrow, Lodz, Poland, 19–21 September 2018.
  9. Samuel, A.L. Some Studies in Machine Learning Using the Game of Checkers. IBM J. Res. Dev. 1959, 3, 210–229.
  10. Smetannikov, I.; Isaev, I.; Filchenkov, A. New Approaches to Parallelization in Filters Aggregation Based Feature Selection Algorithms. Procedia Comput. Sci. 2016, 101, 45–52.
  11. Han, K.; Wang, Y.; Chen, H.; Chen, X.; Guo, J.; Liu, Z.; Tang, Y.; Xiao, A.; Xu, C.; Xu, Y.; et al. A Survey on Vision Transformer. IEEE Trans. Pattern Anal. Mach. Intell. 2023, 45, 87–110.
  12. Stanley-Price, N.; Talley, M.K.; Vaccaro, A.M.; Institute, G.C. Historical and Philosophical Issues in the Conservation of Cultural Heritage; Getty Publications: Los Angeles, CA, USA, 1996; p. 500.
  13. Thomas, M.S.C.; McClelland, J.L. Connectionist Models of Cognition. In The Cambridge Handbook of Computational Psychology; Cambridge University Press: Cambridge, UK, 2012; pp. 23–58.
  14. Mohri, M.; Rostamizadeh, A.; Talwalkar, A. Foundations of Machine Learning; MIT Press: Cambridge, MA, USA, 2012; p. 414.
  15. Chaillou, S. Architecture & Styl |A New Frontier for AI in Architecture; Harvard University: Cambridge, MA, USA, 2019.
  16. Bonabeau, E.; Dorigo, M.; Theraulaz, G. Swarm Intelligence: From Natural to Artificial Systems; Oxford University Press: Oxford, UK, 1999.
  17. Chazelle, B. The Convergence of Bird Flocking. J. ACM (JACM) 2014, 61, 1–35.
  18. Reynolds, C.W. Flocks, herds, and schools: A distributed behavioral model. In Proceedings of the 14th Annual Conference on Computer Graphics and Interactive Techniques, SIGGRAPH 1987, Anaheim, CA, USA, 27–31 July 1987; Volume 21, pp. 25–34.
  19. Goodfellow, I.J.; Pouget-Abadie, J.; Mirza, M.; Xu, B.; Warde-Farley, D.; Ozair, S.; Courville, A.; Bengio, Y. Generative Adversarial Nets. Adv. Neural Inf. Process. Syst. 2014, 2, 2672–2680.
  20. Tamke, M.; Nicholas, P.; Zwierzycki, M. Machine learning for architectural design: Practices and infrastructure. Int. J. Archit. Comput. 2018, 16, 123–143.
  21. Basarir, L. Modelling AI in Architectural Education. Gazi Univ. J. Sci. 2022, 35, 1260–1278.
  22. Ceccon, L.; Villa, D. AI-BIM Interdisciplinary Spill-Overs: Prospected Interplay of AI and BIM Development Paradigms. In From Building Information Modelling to Mixed Reality; Springer Tracts in Civil Engineering; Springer: Berlin/Heidelberg, Germany, 2021; pp. 195–217.
  23. Longo, L. Empowering Qualitative Research Methods in Education with Artificial Intelligence. Conf. Pap. 2020, 1068, 1–21.
  24. Negroponte, N. The architecture machine. Comput.-Aided Des. 1975, 7, 190–195.
  25. Mustoe, J.E. Artificial Intelligence and Its Application in Architectural Design. Ph.D. Thesis, University of Strathclyde, Glasgow, UK, 1990.
  26. Huang, W.; Zheng, H. Architectural drawings recognition and generation through machine learning. In Recalibration on Imprecision and Infidelity, Proceedings of the 38th Annual Conference of the Association for Computer Aided Design in Architecture, ACADIA 2018, Mexico City, Mexico, 18–20 October 2018; Association for Computer Aided Design in Architecture: Fargo, ND, USA, 2018; pp. 156–165.
  27. Reviriego, P.; Merino-Gómez, E. Text to Image Generation: Leaving no Language Behind. arXiv 2022, arXiv:2208.09333.
  28. Lee, S.; Hoover, B.; Strobelt, H.; Wang, Z.J.; Peng, S.; Wright, A.; Li, K.; Park, H.; Yang, H.; Chau, D.H. Diffusion Explainer: Visual Explanation for Text-to-image Stable Diffusion. arXiv 2023, arXiv:2305.03509.
  29. Borji, A.; Ai, Q.; Francisco, S. Generated Faces in the Wild: Quantitative Comparison of Stable Diffusion, Midjourney and DALL-E 2. arXiv 2022, arXiv:2210.00586.
  30. Başarır, L.; Erol, K. Briefing AI: From architectural design brief texts to architectural design sketches. In Proceedings of the 9th ASCAAD International Conference, Virtual, 2–4 March 2021; Volume 1, pp. 23–31.
  31. Henshaw, V. Urban Smellscapes: Understanding and Designing City Smell Environments; Routledge: Abingdon, UK, 2013; p. 256.
  32. Zhang, Z.; Fort Mir, J.M.; Mateu, L.G. The Effects of White versus Coloured Light in Waiting Rooms on People’s Emotions. Buildings 2022, 12, 1356.
  33. Bille, M.; Sørensen, T.F. Atmospheric Architecture: Elements, Processes and Practices. In Senses and Sensation: Critical and Primary Sources; Howes, D., Ed.; Bloomsbury Academic: London, UK, 2018; Volume 4, pp. 137–154. ISBN 9781474274050.
  34. Salingaros, N.; Mehaffy, M. A Theory of Architecture; Intercollegiate Studies Institute: Wilmington, DE, USA, 2006.
  35. Lavdas, A.A.; Salingaros, N.A. Architectural Beauty: Developing a Measurable and Objective Scale. Challenges 2022, 13, 56.
  36. Alexander, C. The Timeless Way of Building; Oxford University Press: New York, NY, USA, 1979; p. 552.
  37. Buras, N.H. The Art of Classic Planning; Harvard University Press: Cambridge, MA, USA, 2022.
  38. Berridge, K.C.; Kringelbach, M.L. Building a neuroscience of pleasure and well-being. Psychol. -Well-Being 2011, 1, 3.
More
© Text is available under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/)
Video Production Service
Feedback