ACAPULCO – Seeing (through) images from Large Scale Vision Datasets

Application summary

For Acapulco, people were invited to live with images contained in Large Scale Vision Datasets (LSVDs), used to train machines. The rise of AI/computer vision has gained prominence, especially as it is increasingly used in facial recognition tools, autonomous vehicles, the military industry, etc. This emergence coincides with the dissemination of LSVDs: millions of low-resolution images taken without consent from social media, organized through precarious labor carried out by microworkers on platforms such as Amazon Mechanical Turk. Challenged by this opacity, I decided to tackle the inaccessibility of dataset black boxes by creating a long-term practice for a deeper understanding of them. During the pandemic, sets of 3 postcards with LSVD images were sent (by mail) to art historians, engineers, museum guards, artists, programmers, activists, designers and microworkers. With more individualized images than in the LSVDs, I established countless exchanges with these “seers”. My proposal is to work with their feedback and create an experimental film about images that train machines. I also intend to carry out collective outdoor activities with these images (many related to fauna and flora), juxtaposing them with the natural surroundings of the Saari residency. Acapulco, the project's name, comes from a conversation with a microworker. They told me that for a week they performed a task on Amazon Mechanical Turk in which they had to choose which beach was displayed on the screen from 5 alternatives. One of the options was always Acapulco beach (probably a bug). The recurrence spurred they to reflect on the power of image systematization and how Acapulco can be a metaphor for places that are not necessarily physical, but that exist from the relationship between humans and images. “I’ve never been to Acapulco, but I like it there,” they told me.