What role do images play in globalization? This role is under-studied, especially on a large scale. What are the most shared images? Has the often criticized homogenization truly occurred? Are there dominant centers and cultures in the global circulation of images? 

The Visual Contagions project (FNS 2021-2024), led by the Chair of Digital Humanities at the Faculty of Arts of the University of Geneva, aims to answer these questions. Some visual sources, mass-digitized, are available globally for the entire 20th century: illustrated periodicals, movie posters, social media images. The challenge now is to conduct their historical study, utilizing algorithms as well as our own knowledge.

Team: Marie Barras, Nicola Carboni, Adrien Jeanrenaud, Adélaïde Quenson, Thomas Gauffroy-Naudin, Rui-Long Monico, Bokar N'Diaye, under the direction of Béatrice Joyeux-Prunel. With the collaboration of Guillaume Aebi, Céline Bélina, Raoul Bikel, Aurora Madsen, Tara Lazovic, Elodie Sierro, Esther Solé i Marti, and Barbara Topalov, and the students in digital humanities.

 

Des algorithmes pour étudier la mondialisation visuelle 

When quantity overwhelms us, machines can assist us. Yet, it is essential to gather relevant, representative, and well-described data for its study to be useful.

For periodicals, millions of magazines and newspapers are accessible online worldwide.  Although European press and North American periodicals remain the most represented, the team is gradually discovering magazines from Latin America, Asia, and North Africa. Nearly 15 million images have been extracted from periodicals published primarily from the 1890s to the 1960s. 

To study movie globalization, tens of thousands of film posters are available, compiled from specialized cinema and poster websites. Created post-1945, these film posters are accompanied by details regarding their production, distribution, and reception, and can also be analyzed as visual artifacts. Regarding social networks, they represent an ephemeral repository of global preferences.

Images shared on these platforms can serve as valuable resources for investigating phenomena such as the globalization of Street Art and contemporary art. Grouping our images by visual similarity helps identify patterns, reproductions of objects, 'icons' of the 20th century, or even styles, whose circulation is then reconstructed.

 

Visual Epidemics - Origins of Our Time?

What were the most viewed images of the 20th and 21st centuries? What do they say about past and present generations? Are there centers and peripheries in the global deluge of printed images? Is there a pattern behind the apparent chaos of visual globalization?

The early 20th century appears as a still religious century: images of the Virgin Mary circulate in large quantities, especially in the European press before the 1950s. Since the 1950s, the illustrated press reflects a movement of disenchantment and distancing from European legacies.

An image of the Virgin among images of the Virgin

The 20th century witnessed the rise of a culture that embraced automobiles, which became the most widely reproduced consumer product of the era. We are aware of the influence of images on consumption habits and the shaping of desires—a phenomenon Deleuze and Guattari referred to as "libidinal capitalism." Could there come a time when the reproduction of such images is prohibited in order to break free from the collective addictions that anchor us to the Anthropocene?

The golden age of Hollywood cinema saw its last hours at the end of World War II, but it did not know it yet. Soon, television and video would offer faster and more personal means to see images. National cinematographies gradually open up. Then, films and their images circulate more globally. More recently, digital technology has accelerated the circulation of all kinds of images.

A colorimetric visualization of Venus images on Instagram

Weaving, Playing, and Being Played: AI in Humanities Research

Visual globalization, big data, images in bulk... How can we narrate and visualize globalization through so many  images? How do we uncover patterns within chaos without imposing our expectations onto our datasets? 

And how do we account for the method's effects on the results? From global infographics to case studies, viewpoints can vary widely. The challenge is to maintain coherence without overly objectifying results, and to make room for the intelligence and intuition of the experts we are.

Each step of our research influences the others; like a complicated weaving where the outcome depends on our ability, precisely, not to forget any thread. The unconventional methods of research-creation was a good way to gain perspective on this approach, and better understand what is at stake. We decided to confront algorithms directly, in the very exercise where historians may think themselves a bit too independent: storytelling.

We also wanted to embrace our project in a larger and longer history. Every study of visual globalization in the contemporary era is charged with the history of images, the history of techniques, as well as the history of fears and hopes that arise from encounters between cultures.

A recurring theme surfaced: weaving. Spanning from the Bayeux Tapestry, crafted in the 11th century to depict the Norman conquest of England, to the textile-inspired terminology of the internet (Web, Threads, Networks...), and including the inception of punched cards in the textile industry, textiles intricately link together the diverse facets of our historical exploration.

 

B-AI-Yeux. A Tapestry Generated from a Project, a Story, and Images

B-AI-Yeux – a strange name for a fictional tapestry born from the generation of images by an algorithm trained on reproductions of the original Bayeux Tapestry, based on a very contemporary story. Amid the controversies surrounding the introduction of machines into the textile industry (up to the destruction of the Luddites), and those that concern our era about artificial intelligence, taking control of these image-generating tools was like an attempt to tame these machines.

Thus, B-AI-Yeux. Like an algorithm's name; "AI" obviously refers to its origin; "Yeux" (eyes, in French) clearly indicates that we seek to see what has been seen, reviewed, forgotten, made invisible, or highlighted in the history of globalization; and the sound of “B-AI-Yeux” reminds us of our tapestry’s medieval ancestor.

Two examples of images from the Bayeux Tapestry that were used to construct an image generation model

B-AI-Yeux represents a departure from the conventional historical methodology. We initiated the process with a narrative— that of our project and its outcomes. Subsequently, we generated data from the ground up by feeding prompts to a machine, which in turn produced images. Some prompts consisted of textual descriptions, articulating in a language conducive to machine comprehension the methodology of the Visual Contagions project, while others were simply images extracted from the project's archive. Leveraging an open-source algorithm pre-trained on the Bayeux Tapestry, we embarked on a lengthy journey of image generation, characterized by numerous failed attempts and incremental adjustments.

An image prompt used to generate a tapestry sequence

The images produced underwent a process of concatenation and were subsequently upscaled multiple times. They transcended their status as "poor images," characterized by low definition and originally intended for online use—the very kind of images that artist Hito Steyerl refers to as the Lumpenproletariat of images, often overlooked and ignored. Instead, they transformed into "rich images," boasting high definition and capable of being materialized. In an age dominated by Open Access and digital mediums, where only a handful of our findings will manifest in tangible print, these images may be the sole remnants of our project once the computers are powered down.

True, False, Scientific, or Not – Nevertheless, a Reflection on Images Yesterday and Today

In this B-AI-Yeux tapestry, truth and falsehood intertwine seamlessly. It unfolds as a whimsical tale in images, simultaneously coherent and surreal. It encapsulates our project on globalization through images, encapsulating our data, methods, tools, and initial findings. It chronicles our trials, errors, improbable discoveries, dead ends, doubts, and sudden revelations.

The tapestry both resembles and diverges from our intended message. From a distance, the characters seem to live and breathe, each action imbued with meaning; up close, one notices their disjointed forms, unfinished features, awkward movements, and distorted symbols. Thread by thread, the connections may seem humorous, nonsensical, yet profoundly relevant.

Contrary to the prevailing discourse on AI, algorithms, and science, this is the essence of research. It is a process marked by fluidity, where choices and non-choices, serendipity, and effort interweave. It involves extensive groundwork, often involving Herculean efforts to mitigate the biases that surround us, and the realization that acknowledging these limitations is essential. If everything were too polished, too perfect, or too obvious, there would be no room for discussion. If everything were imposed, how could one feel truly free to interpret and engage with these visual legacies?