Just how creative can an algorithm get?
The image an astronaut relaxing poolside is not a photograph. But it’s not a drawing either, or at least, it wasn’t drawn by any specific person. In fact, the image comes from DALL·E 2, an algorithm recently presented by OpenAI that is capable of creating all kinds of images in a range of styles based on a written description.
You simply type what you want created: “a duck in a scarf swimming in a bowl of soup”, define the style, photorealistic, pixel art, ’70s, steampunk, pencil drawing, etc., and the algorithm generates it for you with surprising precision, even proposing alternatives.
It is another dimension, but with the same procedure, of the texts that simulate human writing generated by GPT-3 or the images of people that are increasingly being created for all kinds of uses without the need for a professional model: do you want an article on a particular topic, or an image of a smiling person to illustrate an advertisement? Don’t write it or look for a model: have an algorithm generate it. Want a picture? Describe it and ask Ai-Da. A music video?
With DALL-E 2 — the name is a mix of Pixar’s famous robot and Salvador Dalí’s surname — the principle can be extended to illustrations, and I can imagine myself every morning, when I have to choose an image to illustrate my daily article, describing something for an algorithm…
