During the last two years generative AI became good enough to pass for human creation in text, image, video and audio. I was wondering how that would influence creation of those media and if there would be some kind of noticeable fundamental shift. I was looking for radically new things in those areas, for example completely autonomously created art by AI without any human input.

What I observed is that art and creation were liberated in some form via natural language interfaces. It was now possible to generate something by using basic instructions, like the style or era or some qualifiers. This lead to a lot of new media, but most of it in the expected range of what was popular today or before. With new interpretations or surprises usually part of the prompt, so actually triggered and thought of by humans.

In general it feels like generative AI became more like an additional tool with a new instruction set to learn. People quickly learned the trigger words, the qualifiers that where tagged and curated manually by people, leading to constraints of what can be requested or created. In a sense it was just some interactive catalog that contained a truly vast collection to be perused.

From experience, it looks like most generated media is low effort, a quick joke or an illustration, hardly the game changer that I expected from actual artificial intelligence. Though it is a decent ghostcreator delivering expected, but derivative results on a professionell level.

Sure there were some new things that came out if, due to people - not AI - being creative. AI hallucinations and other artifacts are used as an effect in video, audio and images. Text generation is useful in creating drafts and iterating quickly to try stuff out and AI tools in general increased productivity for creators.

In my opinion this just points to the fact that generative AI is really like any other tool in the hand of a skilled creator. Another tool to create art, but not a creator in itself. Or is there a way for it to be an actual artist? So here we are, asking the fundamental questions what is art and who is an artist?

For example, in the context of an image generator, the input prompt is the influence and direction a creator gives to actually create something with a deeper meaning, or something that resonates with the audience, or some new perspective. Without the guidance of what a person wants to see or evoke there is nothing the AI can deliver from itself.

And that brings me to the thing that AI actually adds to the equation. There is a random component in how a prompt actually leads to a piece of media. No two generations are identical and this noise is exactly the only place for generative AI to add something. Usually it is ignored, or seen as annoyance if the result is not exactly as imagined.

Current "AI" models are generated using large datasets with human annotation, which on datum by datum is nothing interesting but taken into the whole picture, similar labels and correlations emerge between different parts of the dataset. These hidden connections might become visible after prompting and result in unexpected outcomes. Exactly these kind of connections are a fundamental part of making art and being creative.

The question now is, can this be seen as a true creative process? Or is it just a glorified remix engine? Is it truly new and unique or just a mashup? Is there a way to find groundbreaking new media doing this? I do not know, but investigating the correlations and allowing the model to make and show connections easier would be worthwhile to explore in my opinion. Currently I am not aware of a piece of generated media that is considered conceptionally new, meaning not its generation process, but its result.

Individually, it comes down to what a piece of media evokes in the beholder to be considered art. I am sure there is media that is convincing enough to be thought of as a creation by a human and is looked at as outstanding, especially if no creation context is given or revealed.

Maybe thinking about it differently, this kind of art can be seen as coincidental art, kind of like pressing a random generator and hoping for a nice piece to emerges from the centuries of human data and careful training. Then, the only difference to human art, which also heavily relies on history and careful training, would be the intent of creation, the spark, the desire to show something or evoke something particular or just base expression.

I think, we will see how people will judge generative AI in the future, currently though, I think it's kind of like clipart or stock pictures. And that is fine for what it is. Asking it to emulate human creativity is like asking for a definition of art.