Cities of Ibero-America as seen by Artificial Intelligence

Nadia McGowan (Universidad Internacional de la Rioja)

This study explores how Generative Artificial Intelligence envisions Ibero-American cities through the creation of synthetic content (García-Peñalvo & Vázquez-Ingelmo, 2023). Using the Midjourney AI tool, images of cities from 23 countries in Ibero-America were generated. The AI was prompted with “/imagine <city name> –ar 16:9” to create these images. This leaves the generation of the city in the hands of the Midjourney AI, without further modification. The only change is the aspect ratio, to better adapt to current 16:9 standards instead of the default 4:3 it uses. Music was also AI generated using Loudly for ambience.

The images obtained reveal several common aspects. Firstly, the AI-generated images often look strange and uncanny due to the mix of architectural elements. The AI combines different styles without consistency and does not always follow real-world rules, resulting in images that do not follow the laws of physics (Hanafy, 2023; Wang et al., 2023).

Secondly, while visually appealing, the AI’s images don’t capture cities as humans see them (Rico Carranza, Huang & Besems, 2023). The images seem familiar yet distant, not always representative of how we conceive these cities and our experience of the urban space. This is because the AI lacks understanding of human experiences and cultural context that shape our perception of cities. 

Thirdly, certain features like graffiti, domes, and apocalyptic scenes are overrepresented. This happens because the AI learns from its training data and they are likely to be overrepresented in it, giving excessive focus to these elements. AI models tend to prioritize features more easily distinguishable and visually appealing (Anicin & Stojmenovic, 2023). Domes are often associated with iconic architectural features, which makes them more likely to be included to create a visually interesting scene.

While relevant landmarks appear, they are often unrealistic in their portrayal. Interestingly, the AI blurs the lines between cities, sometimes creating confusion. This challenges the distinctiveness of cities as AI intermingles elements from different places (Ye, 2023). The AI’s images also share traits with modern photo editing, featuring exaggerated lighting, vibrant colors, and fantastical elements. The prominence of pedestrian areas and green spaces align with AI’s focus on human-centric spaces.

This study examines how technology can be used to create representations of urban space. These are alternative sights, beyond those created by people. It showcases the AI’s ability to produce unexpected images from a dataset. These images are visually striking but lack the cultural context, which makes them familiar but distant.

Bibliograhpy

• Aničin, L., & Stojmenović, M. (2022, December). Bias Analysis in Stable Diffusion and MidJourney Models. In International Conference on Intelligent Systems and Machine Learning (pp. 378-388). Springer Nature Switzerland.
• García-Peñalvo, F.J., & Vázquez-Ingelmo, A. (2023). What Do We Mean by GenAI? A Systematic Mapping of The Evolution, Trends, and Techniques Involved in Generative AI, International Journal of Interactive Multimedia and Artificial Intelligence. http://dx.doi.org/10.9781/ijimai.2023.07.006
• Hanafy, N. O. (2023). Artificial intelligence’s effects on design process creativity:” A study on used AI Text-to-Image in architecture”. Journal of Building Engineering, 8(1). https://doi.org/10.1016/j.jobe.2023.107999.
• Rico Carranza, E., Huang, S. Y., & Besems, J. (2023). (In) visible Cities: What Generative Algorithms Tell Us About Our Collective Memory Schema. In CAADRIA proceedings (pp. 463-472). CAADRIA.
• Wang, B., Zhang, S., Zhang, J., & Cai, Z. (2023). Architectural style classification based on CNN and channel–spatial attention. Signal, Image and Video Processing, 17(1), 99-107.
• Ye, S. (2023). Generative AI May Prefer to Present National-level Characteristics of Cities Based on Stereotypical Geographic Impressions at the Continental Level. arXiv preprint arXiv:2310.04897.