Ghosts@Work, Lilly Lulay
Artist Meets Archive, Fotoszene Köln / Rheinisches Bildarchiv 2023
Ghosts@Work
Artist Meets Archive#3 exhibition during Photoszene Festival 2023
The following projects are the outcome of a research residency at RBA Rheinisches Bildarchiv Cologne. The starting point of my investigation were approximately 1m3 of analogue travel photographs and stereoscopic images, produced between 1961-1999 by Karl-Heinz Hatlé. These analogue images had just been included in the RBA archive and now had to be digitized in order to make them publicly accessible online. Assisting this very labour intensive process made me raise many questions:
- how do we connect with the world on an analogue VS digital level?
- visual impression/ 3D images VS physical experience
- how many hands have touched the analogue objects and digital products we use every day?
- automated vs handmade labeling of large datasets
Thinking of the image archives that digital companies like Google and Meta hold today made me wonder about the difference between public and private archives:
- purpose of / access to / lifespan of the contents?
- which categories / hierarchies / censorship are at work?
Masks
A series of 15 seeing aids that transform the visual appearance of the visitors in the exhibition as well as their view on the objects. The visitors can use these stereoscopic masks to explore 3D effects inside the two bodies of work:
The Original Taste / Performing for the Camera
The masks are the result of a workshop with students from Gesamtschule Holweide regarding public image archives like the RBA in comparison to Instagram, as todays most popular but privately owned image archive. How does this social network inform our collective behavior off- and online? How does it attract our attention and shape our perception? We observed how this platform seduces and influences our decisions in a playful manner by suggesting filters, hashtags, locations and social connections. We also discussed how our personal photographs are turned from medias of memory and communication into instruments of social control and surveillance in the framework of such apps. We touched on topics like data mining, location tracking, face recognition and the question if we as "Prosumers" should be payed for the labor we unconsciously perform while scrolling through, posting and liking images. The masks may instigate the users to take self portraits but protect them from being recognized by face recognition algorithms.
The Original Taste, a Tasty Takeover
16 folders with self designed archive cards that trace down the accident appearance and the percentage of space that American consumer products like Coca Cola, Pepsi and Marlboro occupy in the travel photographs of Karl-Heinz Hatlé and equivalently in the public space all around the globe.
--> supranational American brands shape our environments:
analogue world = Coca Cola / Pepsi / Marlboro...
digital world = Google / Meta / Amazon...
Performing for the Camera, Dias VS Instagram
A popular technique in the analogue era for presenting photographs to a wider audience was via diapositives. Instagram, currently the most popular photo sharing app, is clearly inspired by the aesthetic and format of this earlier medium. In a series of curtains, I juxtapose stereoscopic images from Hatlé, that show people posing or performing for the camera, with images that I found in Instagrams database, using Haltè's typewritten caption as #.
Launched in 2010 Instagram has 2 billion users today = 25.31% of the world’s population use the platform actively. This means:
1. this platform is significantly influencing the way people look at and present themselves to the world.
2. users are collectively building an image archive of the present whose data volume far exceeds that of the RBA or any other publicly owned image archive.
The Rheinisches Bildarchiv (RBA) is one of the largest public art-historical picture archives in Germany with approximately 5.5 million analogue and digital images. The number of daily uploaded pictures / videos on Instagram is estimated at 95 million.
While the RBA’s mission is to ensure the conservation of knowledge and history for future generations to come and to serve the public good no one knows under which conditions and if at all Instagrams database will still be accessible in 20 or 50 years. Deeply rooted in a commercial logic Instagram has to remain profitable in order to persist. So how will the platform explore its collectively built visual archive? How does it already capitalise on the behavioral patterns we unintentionally exhibit when we perform everyday gestures such as scrolling, liking, posting and tagging images in this digital archive.
Ghosts@Work, Framing Your View
Nowadays, we are more connected to the rest of the world than ever before. On the one hand, via electronic devices and digital services on the internet. On the other hand, through food, clothing and other products of daily use. Yet we hardly think about those who are on the production line of a physical or digital product. The increasing use of AI applications is contributing to this tendency - things now seem to work without any human intervention. In reality, however, behind every seemingly autonomous AI application stands an army of click workers who train and constantly correct an AI. Just as industrially manufactured products erase the gestures of the individual worker, AI applications willingly hide the workers in the background.
For this series, I searched Hatlé's travel photos for images of people working with their hands. I then fed these images to Google Vision AI - an AI system for recognising content and labelling images - to see what this AI would recognizee in such old-fashioned scenes of manual labour. The result is a series of wall sculptures of partially absent workers which show both the AI's focal points as well as its blind spots.
--> green frames = objects detected /highlighted by Google Vision AI
--> gestures of manual labor = the section of the picture that I allow the viewer to see
--> black panels = reference the 3:4 display format that became a standard with the rise of digital photographs and screens / the "back-box" character of AI / the tools used to create this series
Whereas AI intentionally hides the human labor required for its training and functioning, the black panels in the background instead, make the tools of their production explicitly visible. As soon as the viewer steps aside, photogram like imprints of sand paper, tooth blades, paint rollers, pliers, screws and other objects manipulated to create the series, become visible as shadow figures.
Ghosts@Work, Text as Key
An image that has no annotations is dead - meaning that there is no way to find, organize, link and compare it with other contents in an archive. In this series of semi-transparent fabric prints I contrast 3 types of image description. The typewritten labels that Hatlé stuck on his slides by hand many years ago. The labels that Google Vision AI automatically generates to describe Hatlé's images, and the textual metadata that the RBA adds to them once they are digitized.
While Google deliberately obscures the rules its AI follows to select and label image content, the parameters of the RBA are transparent. Following the premise to make information available to the public, the RBA assigns labels very carefully, using acclaimed systems such as "Iconclass", which allow the interchange of knowledge between institutions worldwide. Comparing the text data assigned by RBA with the labels attributed by Google Vision AI, the commercial orientation of the AI system becomes clear. With striking frequency it subsumes all sorts of objects under terms like packaged goods, hats, t-shirts, kitchen appliances and other trendy topics. Every object recognized with a confidence value higher than 50 % is labelled, even if this leads to contradictory outcomes. The AI labels the same object as a table 67% and is also 23% sure that it is a bed. While information is only included in the RBA database if it can be verified by trustworthy sources, Google seems to follow the premise that a false label is better than no label at all, thereby creating a lot of misinformation.
The image of a stonemason's workshop in Egypt, with workers in white clothing that sit in front of a turquoise wall, furthermore reveals the cultural biases of Google's AI. Apparently trained primarily on images from western cultures, the color combination of white and turquoise is recognized but misunderstood as a medical setting. Subsequently only false labels are associated with this image from a less known time and culture.
