Find out more about the project here.
The image on the left is from a database called ImageNet-1k, which is used for training AI image models. This image is among over one thousands of images labeled with "ice lolly, lollypop, popsicle"—The database told the algorithm how popsicles look like.
On the right page is the interpretation of this image from an AI model that is trained by the database. Obviously the AI model reads the images more than just a popsicle.
Find out more about the project here.