How do AI tools see the world? This question led to maps made by asking AI to generated an image of specific areas. This reveals how AI training data shapes its view of the world. With sometimes surprising results, like Puglia only showing dogs or North-West England showing Kanye West's daughter.
How do new AI tools (DALL-E, Midjourney, Stable Diffusion) see the physical world? This question led to large geographical maps made up of tens of thousands of AI-generated images, where the AI was asked to generating an image of a specific area.
This process resulted in some remarkable spots on the maps. For example, Normandy looks like D-Day because “Normandy” in the AI models' training data mainly consists of images from D-Day. In certain regions, you predominantly see images of football stars, election campaigns, or dog breeds native to that area. And North-West England: in this area, you see pictures of Kanye West and Kim Kardashian's daughter, because her name is “North West”, and she appears more frequently than actual photos from that region.
These locations together form an expedition to learn about how AI sees the world. Each location reveals something about how AI depends on training data and how this data is processed and collected. It offers visitors to Generated.Earth a critical perspective on the images AI generates, which is crucial to remain clear-headed in a future where AI dominates our media.
