From issue: #25 Multiplication
Carolina Semprucci, 06 December 2024
General outrage regarding Meta’s small print enabling AI training from public posts on Instagram and Facebook has recently reignited debate about the lack of transparency surrounding data collection. Although Meta confirmed that re-sharing the viral post objecting to the training was never a valid opt-out method, the ease with which many users participated in the mass re-share reflects the widespread carelessness that alienates most (of us) from the potential consequences of their online choices – agreeing to Terms and Conditions they did not read, allowing various forms of tracking through their mobile devices, accepting all cookies instead of selecting only essential ones, and so forth.
This episode has proved that, despite expressing their dissent toward the increasing implementation of online AI features – such as Google’s AI Overview, Meta’s AI assistant, and Instagram’s AI Info – most users put little effort into understanding how to effectively counter these technologies, perhaps discouraged by the complexity of opting out processes. Today, the average internet user seems to have come to terms with their data being collected and fed back to them through targeted ads. This acquiescence, fostered by the incessant rhythm at which one is required to exist on social media, risks paving the way toward a general acceptance of data collection being used in increasingly intrusive ways.
Though social media surveillance is not a new phenomenon, its legislative implementation is fairly recent, and continuous advancements in the technologies it employs have left it largely unregulated. Sara Hodges, a multimedia project by Belgian artist Marcel Top, explores how social media surveillance impacted American social movements in 2020, when, due to the rise in civil unrest, law enforcement was permitted to use facial recognition to track demonstrators through social media. Following instances of people being identified and arrested due to what they had written online, Top crafted Sara Hodges, a symbolic countertype to “suspicious” online behaviour. The project shows the potential of reclaiming surveillance practices by gaining knowledge of available tools to protect the rights to freedom of expression and assembly from oppressive state surveillance.
Sara Hodges is an AI-generated prototype of the ideal American citizen in the eyes of surveillance. In the online realm, her persona (@sarahodges1973) is based on the digital presence of people who expressed love for America on social media and are therefore less likely to be flagged as threats.
Using code available online, Marcel Top collected 50,000 Instagram posts using the hashtag #iloveamerica, resulting in a folder containing the profile pictures and usernames of 17,000 Instagram accounts. To ensure that only the data of actual American citizens would be collected, Top manually reviewed these accounts. Once the selection was accurate, he used code again to collect all the available data, including profile pictures, posts, and biographies to generate new assets through machine learning.
Initially, a new face was generated from the database of collected profile pictures. During the process, the features of a middle-aged white woman were the most recurring result. From all the variations, Top selected the version he thought best displayed the balance between realistic and unsettling features. To simulate her online presence, he categorised the posts, which amounted to over 1.5 million images, to generate new pictures. Common subjects included trucks, sunsets, flowers, the American flag, dogs, guns and babies. Once he had a face and a profile picture, Top used a LLM (Large Language Model) with the text he collected to generate a new name, a username and an Instagram bio. Similarly to the process for the face picture, Top selected the results he believed best aligned with the other accounts he reviewed. The generated name chosen was Sara Hodges and the Instagram bio was:
“💍 Married to my HS sweet 💕
Mom of 5 👩🏻🎨🇺🇸believer in the American Dream✝️
Matthew28 Psalm23:4
Faith is strength is determination is practice”
On Instagram, this generated self-description frames the images that have populated Sara’s feed since 2020: her own portrait, the picture of a strange red car, a distorted sunset, a warped American flag that appears to float in a field, a deformed baby-like figure, an object resembling a gun and the odd-looking face of a man.
After establishing her online presence, Top brought Sara Hodges into the real world. By rendering a 3D model of her face and providing an instruction manual to turn it into a mask, Top created a scenario in which demonstrators can protect themselves from surveillance using the very tools with which they are surveilled. If the mask appeared in a crowd, facial recognition would redirect law enforcement to the social media profile of this unsuspecting citizen. Borrowing her face becomes a way to reclaim the fundamental right to freedom of expression on social media and in real life, without the fear of being identified and tracked online.
Although an answer to specific events, the project goes beyond the context it originates from, as deceiving facial recognition technologies is not its only purpose. Sara Hodges is an average product, and the product of machine learning. Due to the paradox she stems from – that in machine learning a finite database can generate an infinite output – there can potentially be infinite versions of this persona, all different yet the same. While for humans an average result is typically a single parameter, averageness for AI is a fragmented concept, made of infinite possible averages. The only way Sara Hodges can be a universal average is by transcending all the possible visual-text combinations that form her digital footprint, leaving behind all her multiples, as she is ultimately an idea. As such, she embodies the prospect of a society that only tolerates standardised thinking, beyond which any divergence from government ideals risks being flagged as a threat.
Marcel Top is a visual artist living and working between Belgium and London. He researches the topics of mass surveillance, privacy and data collection.
Carolina Semprucci is a London-based writer and editorial designer. Semprucci’s practice revolves around the different ways print can enhance artistic work.
