Igor Svetel, Slađana Marković, Markus Hudert
https://doi.org/10.60152/i80enc6z
Abstract: The field of artificial intelligence (AI) has recently attracted the attention of many disciplines due to the exceptional accomplishments and rapidly emerging possibilities. Architecture, a discipline traditionally skeptical of new technologies, is intrigued by the results achieved with generative neural networks. The emergence of new technologies in the field of generative artificial intelligence has led to numerous scenarios among architects about possible applications of these technologies in their practice. A large number of these scenarios are based on the functionality of commercially available Large Language Models and their application to text to image and image to image transformations. A smaller number of scenarios are based on specific applications of AI technologies through the development of real software solutions. The paper critically analyzes existing scenarios through a review of existing research and a description of software solutions. The focus is on the application of Generative Neural Networks and the paper provides an overview of the functioning of the basic classes of neural networks that make up this technology. An analysis of how artificial neural networks work shows that they all depend on the data set used to train them. Today’s neural networks use an enormous amount of data, but even that is limited to what is available in electronic format. The main source of general knowledge is online encyclopedias, while expert knowledge is provided by published articles that have a certain level of reliability. Architects who are mistrustful of this technology need to understand that better results can only be achieved through higher-quality training data sets and that their protectively guarded knowledge also comes from studying the work of other architects.
Keywords: artificial intelligence; AI; architecture; Generative Neural Networks; training sets
How to cite this Paper (Harvard referencing style):
Svetel, I., Marković, S. and Hudert, M. (2025) ‘The role of AI in architecture: fantasies and reality‘, in R. Bogdanović (ed.) On Architecture — Crosscutting and Fusion of Disciplines, Proceedings. Belgrade, Serbia: STRAND, pp. 96–103.
See publication On Architecture (2025) Conference Proceedings