It all started with a curious coincidence. In 2018, German car manufacturers, taken aback by the advents of Google and Tesla on autonomous vehicles, invested a lot of money in research. They gathered an immense amount of data from their cars sensors that they then had to process. The same year, Venezuela was hit by a terrible economic crisis which threw thousands of workers in unemployment and poverty. A lot of them turned to online micro-working platforms and ended up annotating images that were coming by the thousand from Germany.
Unknown Label is a multi-channel video installation exploring the daily experience of online micro-workers from the Global South who annotate images for self-driving cars. It investigates the power asymmetries and neocolonialist exploitation involved in the manual labor necessary to train AI systems. Unknown Label reveals the hidden crowd that works with these operational images and help shape how machines see the world. A digital collage, layering together fragments of annotated images and a virtual map of the world as seen by self-driving cars, creates a space where the viewers can share the vision of otherwise invisible data workers. The video installation thus raises the question of how to reappropriate these operational images which depict and categorize the world we live in without our prior knowledge or agreement.
########################
In Dialogue: Artist Talk with werkleitz Center for Media Art, Halle (Saale)
werkleitz: Could you please give us a short introduction to the EMAP project you are developing at werkleitz?
Nicolas Gourault: The project I’m working on during my residency in werkleitz is called Unknown Label. It is a video installation that deals with the human labour involved in training AI systems, and more specifically with the training of self-driving vehicles in order to "read" their environment, that is recognizing elements around a car when it's driving itself. The process is called "image segmentation" and it consists of drawing an outline around every object in a given image of a street and put a label on it. It is basically translating our world to be machine-readable. It’s a very labour intensive process, it takes roughly one hour per image and there are many thousands images needed, it is thus outsourced through online platforms to countries from the Global South where labour is cheaper. I talked to people who do this job and who are based in Venezuela, Kenya and the Philippines. Unknown Label is a deep dive into the experience of these online micro-workers.
werkleitz: Why were you interested in spending your residency here at werkleitz Center for Media Art in Halle (Saale), Germany?
Nicolas Gourault: Unkown Label is a loose continuation of my previous film V.O., which was already about the hidden labour involved in training self-driving cars. I continued my research and I discovered the work of image segmentation. At the time, I read an article by Florian A. Schmidt about the images sent to online platforms by German car manufacturers like BMW or Mercedes to catch up with Google or Tesla in the field of self-driving cars. There were thousands and thousands of images sent from Germany to the online platforms to be segmented and annotated. Thus my first interest was going to Germany and I immediately thought about Werkleitz for the residency place because it's connected to this complex local context and has a strong focus on media art and documentary filmmaking.
werkleitz: All these micro workers you involve in Unknown Label are based in the Global South. How did you get in touch with them and how did you communicate?
Nicolas Gourault: I got in touch with Florian A. Schmidt who pointed me to a Venezuelan journalist, Andrea Paola Hernandez, who wrote a follow-up piece and agreed to help me with the research. Later I also got in touch with Leonard Simala, a researcher based in Nairobi, Kenya. I relied on their expertise and knowledge of the local situation in their country. It was through them that I could get in touch with micro-workers who chose to remain anonymous. It was important to build some trust with these people because everything is taking place online, there is a lot of fear about scam or abuse because the workers on this platform don't know their clients, they don't know who they're exactly working for. Everything is very opaque, and there is always this fear of not getting the money at the end of the week. Because of this I needed to reassure them of my intentions for the project. For the interviews it was always the three of us – the worker, the local journalist and me.
werkleitz: What was the most interesting aspect of what you discovered in this process?
Nicolas Gourault: It was uncanny to discover this whole process of human labour which is hidden behind a cover of so-called intelligent algorithms. Far from the narrative of the technological breakthrough, there is still until now a very old school system that relies on the inequalities of the global economy in order to enable the AI systems that we benefit from in our daily lives. Through the circulation of images that need to be worked on, it creates an ad hoc relationship between two remote worlds. It was unsettling to hear the reflection of the workers upon images that to me seemed very familiar. I could relate to them quite easily, and yet when I talked to them, I had this feeling of estrangement. It was very special to feel that through their gaze I gained a different perspective on our own world and its own artificiality.