Initiatives developed at USP’s Faculty of Architecture and Urbanism and Design (FAU) seek new approaches to cataloging images in digital collections and aim to bring scientific knowledge closer to society
.
By: Ana Beatriz Tuma*
“Artificial intelligence [AI] is everywhere. It is profoundly transforming the economy and the very relationships within society.” This is how Glauco Arbix, professor at USP’s Faculty of Philosophy, Languages and Human Sciences (FFLCH), begins his lecture for the course Scientific Dissemination for Communicators and Journalists, jointly organized by the School of Communications and Arts (ECA), the Superintendence of Social Communication (SCS), and the Institute of Advanced Studies (IEA).
According to him, society is entering a new cycle centered on digital technologies—particularly AI—which is only just beginning. In this context, USP has become a hub for experimentation across multiple fields of knowledge, exemplified by the Arquigrafia 4.0 Experience project, a thematic initiative supported by the São Paulo Research Foundation (FAPESP) at the School of Architecture and Urbanism and Design (FAU).
In Arquigrafia, experiments have focused on two main areas: producing science communication content to bring architectural technology closer to the public, and developing new methods for cataloging images in digital collections.
Bringing science closer to society
artificial intelligence for science communication in the project emerged from recognizing the potential of the Synthesia video production platform for this purpose.
In the experiments conducted at Arquigrafia, AI is applied throughout the video production process – from conception to editing. According to the professor, the platform can quickly organize content using avatars or voice-over narration and generate an initial video script that the team can later refine, creating a dynamic and interactive workflow.
According to Gustavo Alves Machado, designer and Level II Science Journalism fellow at Arquigrafia, this is one of AI’s main advantages. “The great potential of AI is that it allows researchers to create their own science communication videos without spending too much time editing or learning how to do it,” he says. For him, this potential is particularly valuable for projects that do not have a dedicated science communication team.
.
However, the designer notes that the videos generated by artificial intelligence can be considered of medium but acceptable quality. Therefore, one of the main challenges in producing an excellent product is continuously refining the AI output. “If you want to create a high-quality video, you need to dedicate a lot of time – it takes almost as much time, if not more, than producing it from scratch without AI assistance.” In the project he is involved in, the team supervises everything generated by artificial intelligence, making the necessary adjustments to ensure high-quality results.
In addition, Rozestraten emphasizes that the use of AI in Arquigrafia is carried out “always with a purposeful sense and a critical perspective, positioning ourselves in a judicious manner regarding what is presented.”
The AI-generated videos developed in the project are being published on Instagram and YouTube.
New formats for image cataloging
The thematic Fapesp project team coordinated by Artur Rozestraten has joined the Digital Archives and Research Group, led by FAU professor Giselle Beiguelman, with the goal of developing new methods for image cataloging through the use of AI in digital archives. For this purpose, the initiative has used the collaborative platform Arquigrafia as an experimental environment. The platform brings together more than 14,000 photographs of buildings and urban spaces shared by the Portuguese-speaking community.
According to Beiguelman, artificial intelligence enables computer vision models to “excavate” images in search of new cataloging formats and to generate metadata from stored visual information.
“In the case of Arquigrafia, a platform with over 14,000 images under a Creative Commons license, we use computer vision to identify architectural elements such as doors, windows, and staircases; materials such as concrete and glass; furniture such as chairs and tables; and artistic pieces such as paintings and sculptures,” she explains.
.
According to the professor, this exponentially expands the research possibilities and analytical potential of the archives, going far beyond what is defined in traditional cataloging processes – which remain relevant.
Artur Rozestraten notes that this AI-driven approach has opened up “the possibility of engaging in dialogue with the image arrangements proposed by the systems and of using these arrangements to refine our research, directing this ability to quickly gather images toward more precise, specific, and relevant interests.”
To explore the prototypes developed through this partnership, visit this link. To learn more about the Arquigrafia platform, click here.
JC-IV fellow with the Fapesp Thematic Project Arquigrafia Experience 4.0 and collaborating researcher at FAU
Article developed as the final paper for the course Scientific Dissemination for Communicators and Journalists, offered by USP in the second semester of 2025. The text was inspired by the lecture Challenges of Artificial Intelligence in the 21st Century, delivered by professor Glauco Arbix.
English version: Nexus Traduções, edited by Denis Pacheco
.
