We use AI to pair phrases and images. Then given a random string, we look for the closest image encoded near the string. (We actually modify an image to make it as close as possible to the string).
Here, I put strings related to "pi" or "mathematician" and "math people". So the images were created by an artificial intelligence (VQGAN+CLIP using code by Johnowhitaker_Art).