2 OCTOBER 2022 / THOUGHTS

Art by AI

Dall-E 2, Disco Diffusion, Midjourny, Craiyon.ai and now Stable Diffusion. The development of generative diffusion models is moving incredibly fast. And all this in an area that was considered untouchable just a few years ago: art.

What are diffusion models anyway?

Very simply explained and never doing justice to the matter, diffusion models are generative A.I. models that can create images from random noise using a text (prompt).

Emotional Discussions

The discussions about the images and the use of such models have become very emotional. There seems to be no middle ground any more between artists who use A.I. as a tool (e.g. in conceptualisation or to generate textures for 3D projects) and those who demonise the images and see the models as the enemy. But the fact remains: such models are available. They can be used easily, and they will no longer disappear. In fact they will only become more powerful and more accessible. None the less I think it is impossible that such models will ever replace artists. Art is an expression of human sensations, feelings and observations. Art can criticise and shake. Works of art can be made tangible and palpable with writing, pen, brush or a musical instrument. Without the human factor, without this emotion, art has no meaning. To be almost philosophical, I see the future most likely in a symbiosis. A.I. can support artists, but it can just as well be left out, it will never be a prerequisite for artistic creation. A work can have an effect on the viewer regardless of the tools used to create it.

Uncharted legal territory?

Of course, for me as a lawyer, the legal issues surrounding models and the generated images are also of particular interest. Models learn concepts from millions of images. They learn the connection between the word "dog" and what is in training images that have that label: Dogs. These are training images, which are probably in most cases copyrighted works. Based on these training results and the learned concepts, an A.I. under the guidance of a human (author of the text prompt) now generates an image. But what is this image? Is it a copyrighted work within the meaning of Art. 2 para. 1 URG? Are the images intellectual creations of art that have an individual character? Can a machine create intellectually at all, or is the human being entering a text prompt the creator? And can the generated images have an individual character at all if the A.I. only learns from existing works? But don't we humans also only learn from what already exists?
Just a small outline of the many questions that are buzzing through my head.

The cover image has been created by AI by the way. It is a Dall-E 2 image.