Putting the Art into ARTificial Intelligence

Recent developments have brought a discussion on whether machines can make art. Surely your social media has been flooded with Midjourney creations in recent weeks. The text-to-image software takes an input from a user and delivers an image generated by machine-learning algorithms drawing from a large corpus of images online. You may have also come across the debate about Theatre d’Opera Spatial, which won at Colorado state fair last month. Other AI content generators include Dall-E (text-to-image), Stable Diffusion (t2i), Runway (text-to-video), Plotagon (text-to-video) and many others.

This begs many questions: Who should be considered as creator? Will machines replace artists? And… can you compete in art?

But first – how new is this debate? South-African author JM Coetzee – a Nobel-prize winner – tells a fascinating story in his autobiographical Youth: Scenes From Provincial Life II (Secker and Warburg, 2002). Working as a programmer in the UK in the 1960s, Coetzee programmed the big mainframe computers of the time. He describes how the computer was programmed to have a sort of intelligence, allowing it to prioritise and plan the various computation tasks that the programmers gave it. In his free time, the author programs the computer to write poetry. He writes that some of those poems were published by a news-paper in Johannesburg. (I have tried to find them, but no success.)

The discussion on AI and art is at least six decades old. Who should be considered creator in this case? Coetzee who programmed the poetry-AI and selected the works that were brought to the newspaper (perhaps also making edits)? The owner of the mainframe computer? The original programmer of the operating system that could prioritise its tasks?

The same questions come with the recent AI art discussion, but with one more layer: the text-to-image systems mentioned above are machine learning systems looking at big corpuses (or is that corpi?) of images online. Someone made the images that the AI was trained on. Several people have made the point that the AI would be useless without those works and the creators and rights-owners ought to be duly compensated.

The games industry has used what is there called procedural content generation systems at least since 1990s. Remember Sid Meier’s Civilization? IMHO the most captivating game of all time. You start with an all-blacked out map, except a small spot where you see what is around you. As you explore the map, you find mountains, rivers, seas and all kinds of landscapes. The map is new every time. No one made those maps by hand, the software generated them. It would be a very time-consuming (and tedious!) task to make hundreds of thousands of maps by hand. 2021 hit Valheim is a modern-day example using the same technology. Does this take away jobs from map-makers? No, the games would not have been feasible without this technology. (The copyright questions are not concerning as this is in-house technology.)

Surely, we will see the more technologies like these. Perhaps a text-to-game AI will be next? The point of license for the material that goes into training the machine-learning will need to be addressed.

Should Jason Allen’s Colorado art prize be recalled? He was transparent about the creative process. He made edits to the AI-generated image. He made a number of versions. If someone would use the artwork without permission, it is likely that he would win the lawsuit. The protection is perhaps weaker than some other works, but this case is rather clear. Copyright is not always binary. AI is.