The rise of OpenAI’s Dall•E 2
Addendum to article,"Will AI Ever Create Beautiful Art?," originally published October 8, 2019.
The past several years have led to some real innovation in art generated by deep learning algorithms, a form of artificial intelligence (AI). Natural language processing and tagged image inputs, from pretty much the entire internet, have been used to train such algorithms. One AI in particular now stands out: OpenAI’s Dall•E 2. This product is now available to the general public, but does have a waiting list.
Dall•E 2 has two chief functions: create artwork from processed text, and modify existing images provided to it. The natural language algorithms used by Dall•E 2 are based on the advanced GPT-3 engine. For Dall•E 2, one can enter a description and have a unique art product output. For example, if you wanted to say, “Mona Lisa made out of jam,” Dall•E 2 would most likely output a strange image of the Mona Lisa made out of jam. One could also upload an image of the Mona Lisa (or anything else), and ask it to make a variation. It may output the Mona Lisa as if it were done by Van Gogh, or it may put the Mona Lisa in the middle of a subway.
Dall•E 2 is very interesting and fun to play with, but is it truly original art from a sentient AI? Alas, no it is not. It is merely using linear algebra and chance to produce unique artwork. It is not truly an artist or truly creative. It does not actually feel anything from the art it produces. It does not feel love, hate, jealousy, happiness, or fear. It cannot produce art like a conscious human being… yet. However, it does demonstrate a move forward in an AI’s ability to create art. It’s just not authentic because all authentic art comes from human consciousness and emotions. AI has no emotions, it only acts like it does.