Information technology infrastructure, or IT infrastructure, is everywhere. It has expanded across our rural and urban environments as a mix of visible and invisible networks, supporting the flows of data, the functioning of information systems, and the communication between devices, computers, humans and non-human beings. Ultimatelly, this infrastructure mediates many relationships among people, machines, animals, and nature.
In our cities, IT infrastructure configures landscapes that vary according to the local cultures, economies, politics, and the particular processes of technology appropriation and development. Despite the global forces that drive the digital transformation, the way in which IT infrastructures are built and maintained in particular cities presents variations. The differences on the materiality of the IT infrastructure cityscapes, its design, aesthetics, and layout, reveals some of the uneven processes of digital transformation, and shows particular future imaginaries.
During the past months I have been working on the LatinAmerican Infrastructure Cityscapes Project (LITCP), a visual exploration of IT infraestructures of three Latinamerican megalopolis (Bogota, Mexico City, and São Paulo) using Runway ML (latent diffusion models), image synthesis and digital photographs. Last May, I shared some of the results of this work-in-progress project at the Co-Designing Generative Futures Workshop that took place at Harvard University. Led by the Berkman Klein Center for Internet & Society, the NCIS at BI Norwegian Business School, ITS Rio, the TUM Think Tank, in partnership with the Global Network of Internet & Society Centers, this workshop brought toghether a network of experts, learners, and decision makers to leverage diverse viewpoints, topical knowledge, and practical experiences about the possible broader societal impacts of generative AI.
Introducing the LatinAmerican Infrastructure Cityscapes Project (LITCP)
In this post, I will share some reflections about the co-creation process with Generative AI (GAI) tools I have developed in the LatinAmerican Infrastructure Cityscapes Project, and showcase some of the imagery I presented at the workshop. This project, is the first one I develop using the commercial Generative AI tools that have become available for a broad spectrum of consumers and users, specifically, the Runway tools.
The recent boom of free and low-cost GAI web and mobile apps has sparked a public debate and discussion about the impact of AI on society. Commercial AI generators are now being used for creating a variety of content, including texts, images, videos, 3d models, and music, and have disrupted the work flows and processes that take place in many economic, cultural and educational sectors. Furthermore, their commercialization and broad adoption in an environment that lacks clear norms and guidances for the design, deployment and use of AI systems, has raised several concerns about their risks. From questions about the intelectual property of the databases used for training the AI models, to concerns about the biases on the outputs of AI systems, to questions about the authenticity of the generated content (e.g. replacement of human creativity, undermining of artistic works) and worries about the possibility of weaponizing these new tools (e.g. disinformation, misinformation, profiling minority populations), the worries about the impact of GAI have grown fast. Although such warnings and worries have been triggered by critical scholars and ethicists before (see for instance the paper clip problem posed by Bostrom in 2014), they have gained visibility while governments, institutions, and citizens directly interact with the tools and with the content that is generated with AI.
With the LITC project I try to explore, through a co-creation and experimental process, some of the challenges and opportunies that emerge with the raise of GAI. Using the RunwayML’s GAI tools for image synthesis (Diffusion Models), and my own digital photographs and textual prompts, in this project I explore and speculate on the present and future IT infrastructures cityscapes of three Latin American megalopolis (Bogota, Mexico City, and Sao Paulo). My approach is based on an ethical relationship with the AI tool that considers it an entity capable of generating content (images) that can be questioned and criticized. This relationship with AI acknowledges the capacity of the GAI systems for creating new content (almost ad infinitum), but treats their outputs as materials that need to be re-worked and questioned.
The theme of urban IT infrastructures is well suited for this kind of speculative, creative and critical exploration. As mentioned before, the materiality of IT infrastructure varies across contexts. Despite the global scale of IT, their implementation and maintenance is different around the world. Mapping such infrastructure networks, particularly elements that stand out in the cityscapes (e.g. antennas that transmit and receive electromagnetic waves and data) can be useful for revealing some of the unique characteristics of the process of technological development and implementations. Its materiality, furthermore, provides clues about particular technological imaginaries, policies, and aesthetics.
For the project, I assumed that by asking the GAI tool to generate synthetic images about the IT infrastructure cityscapes of Bogota, Mexico City and Sao Paulo, one could obtain biased outputs. Or more precisely, one could obtain images that do not accurately represent the 21st century cityscapes (both at the present and future) of the three megalopolis.
With Runway, the synthetic images can be generated in two ways: by using textual prompts (text-to-image), and combining existing photographs and textual prompts (image-to-image). While for the text-to-image process, I created inputs related to 21st century IT infrastructure of the specific cities, for the image-to-image I used digital photographs I have taken during the last two decades and wrote prompts requesting the GAI to create futuristic style cityscapes for each city.(More information about the image generation process and the outputs can be found on the project website).
The results of the co-creation process allow us to compare and contrast the synthetic images and the photographs, and reflect on the ways that GAI is representing the current and future cityscapes of Bogota, Ciudad de Mexico, and Sao Paulo. This kind of comparison, I believe, can inspire critical reflections and dialogues. That is precisely what I wanted to create during the exhibit of the LITCP at the Berkman Klein Center for Internet and Society. Besides setting up a website for the project and delivering a short talk at the workshop, I had the opportunity, with the help of Sandra Cortesi and Patrick Goulart Soares, to print 11 of the images and physically display them at the space of the center.
Selecting the images for the website and the exhibit was a challenge given the amount synthetic images I generated with Runway and the number of photographs I had on my personal archive. This was a problem I confronted during the co-creation process, and that I solved by focusing on the goal of the project. That is, on the possibilities for reflection and critical thinking that the comparison of different images about IT infrastructure cityscapes can spark on both the audience, and the human co-creator.
Three images of each city were displayed on the walls of a lounge area: one generated through text-to-text, one generated through image-to-image, and one digital photograph.The images were displayed vertically, one below the other, with captions about their characteristics, and a QR code for entering the project website.
Furthermore, I also displayed two synthetic images that I co-created by training the Runway AI model with a dataset of 30 photographs from Bogota. (For more examples of the images generated using a custom trained model see here)
During the exhibit workshop participants and other members of the BKC community that attended the social events were able to see the printed images, compare them, and reflect on their own about the different representations of cityscapes. The feedback I received was positive. For some of the workshop participants coming from Latin American countries, seeing the images of the cities they know, resonated more, and made them think about the existing differences about local infrastructure cityscapes, and the way they were rendered by the GAI with sci-fi and futuristic styles. Some of the colleagues from Brasil, agreed that the images of Sao Paulo, both synthetic, and digital photographs, revealed some of the futuristic imaginary and aesthetics that exist in their country, particularly with the architectural works of Niemeyer.
Moreover, some of the conversations that I have during the workshop, allowed me to reflect on how the project can continue to evolve. One interesting area of creative inquiry that I discover through the co-creation process, is the importance of antennas for the infrastructure cityscapes. Despite the ubiquity of telecommunication antennas in the Latinamerican cities, the imagery generated by GAI cannot represent them with accuracy. As it happens with the representation of human fingers and hands (a limitation that has been discussed by the artistic community), GAI tools cannot fully create accurate images of antennas, one of the basic elements of contemporary IT infrastructure cityscapes. Another important discussion I had was related to the usefulness of displaying the original photographs along with the synthetic image generated through image-to-image prompt. Having the possibility of revealing the input source that the GAI leverage to generate new content can be useful not only for the users of the tools, but also for the audience, and for the authors of photographs and works of art that are being used for training AI models.
Finally, this project allow us to think on the importance of developing ethical relations with GAI tools that foster human creativity and critical thinking, and, while acknowledging the biases and limitations of AI systems, opens spaces for dialogue and collaboration. Since the capacity of GAI tools to generate images, videos, music, and texts, is practically unlimited, it is easy to become overwhelmed by the speed and volume of outputs that can be created with these technologies. When using these tools, time is accelerated in ways that affect our human creativity and imagination. The outputs of GAI can easily distract us, detour us from imagining and thinking our own futures, cultures and diversity. As we interact and collaborate with these tools our human cognitive and creative process can easily become overwhelmed by the GAI capabilities. As users of these tools we need to grasp with such power, and explore forms of collaborating and co-creating with GAI in healthy and ethical ways that do not diminish our humanity and diversity. This problem is critical, and needs to be urgently addressed in our societies and communities while AI technologies continue to be used by artists, journalists, students, writers, economists, policy makers across all sectors and institutions.