These images, while very intricate and pretty, are not fractals, and actually show a very interesting limitation with AI nowadays. Image generation AI tools such as Stable Diffusion or Dall-E don’t actually know the meaning of the words you’re using to prompt them, they just have a pretty good idea of what sorts of things pop up if you search for those words.

A fractal is, by mathematical definition, self-similar. You can zoom into part of the smaller detail of a fractal and find the original image, and do the same with the details in the zoomed image, and so on and so forth ad nauseum. Computers are pretty good at making these, once they’re given the rules.

What the image generation bot has given you is an image that looks like a fractal, and that’s what it’s supposed to do. In the same way that large language models like chat-GPT will be very confidently wrong about the information it tells you, and for the same reasons, image generation AI should not be used for important topics that the prompter doesn’t already have some background information about, such as generating a map of some place the prompter has never been in preparation for a road trip.

Thanks for coming to my TED Talk.