Rodolfo De Nadai

Does AI generate art?

Posted: 18 days ago

  • ai
  • machine learning
  • knowledge
  • human creativity
  • art

These are just random thoughts about art, its role in human creativity, and how AI is impacting this field.

A long time ago (though it now seems closer than before), no one would have imagined that algorithms could generate art - at least not the kind of art we’re discussing today. Sure, there has been a lot of "art" generated by computers, but for the most part, it has always been algorithmic art: random pixels drawn on a canvas. An example of this is Genuary, a challenge where participants generate art using code. However, this isn’t what’s currently shaking the art world.

The recent disruption in the art world can be traced back to the release of Stable Diffusion in 2022, or perhaps even earlier with the release of DALL-E by OpenAI in 2021. The point is, before these models appeared, much of what is being discussed today wasn’t even considered a possibility. Of course, when these models were first introduced, the images they generated were more like random visuals related to specific subjects or topics, not what we would traditionally call "art."

But now, people are using these models and claiming they are generating art—art that should be treated as real art, created by actual people who may have spent hours or even days working on it. And that’s the real controversy surrounding AI and art.

Which brings us to the question: What is art?

What's art?

I'm not an artist, nor do I consume much art. It confuses me to even define what art is, and I think most people, if asked, would struggle to define it (or would define it in their own way). But give me a moment to elaborate a little on what I think art is.

Particularly I like the definition given by Wikipedia:

"Art is a diverse range of human activity and its resulting product that involves creative or imaginative talent, generally expressive of technical proficiency, beauty, emotional power, or conceptual ideas."

That is what’s written in the first paragraph of the article.

I really like this definition, and it brings up some important points to consider about art:

Art is a diverse range of human activity

This means that a human must be involved in the process, taking action in some way to create art. As I mentioned earlier, most artists spend hours, days, or even months creating a piece of art, and that’s what makes it special—the human touch in the creation process.

Different people could create the same piece of art with the same results, but some works would be considered art, while others might not—because of the human touch, the process, and who created it.

It's resulting product involve creative or imaginative talent

This is the most fundamental aspect of art, or at least the part most of us think about when we think of art. Especially the imaginative aspect, the ability to create something without prior knowledge or even a reference.

Famous artists are known for their imagination; they created something no one had ever thought of before, used a completely innovative technique, or introduced a new way of seeing the world.

Beauty, emotional power, or conceptual ideas

I'm not sure about this one. Why? Because beauty is subjective, what's beautiful to me might not be to you. The same applies to emotions; they aren’t the same for everyone, and some might consider a piece of art more significant than others.

But the key idea here is that a piece of art might evoke deep thoughts or emotions related to some part of your life. Alternatively, if you're familiar with the artist and their life, you might better understand their feelings at the time the art was created

Ok, this is all good, but does AI generate art?

Real Picasso

Real Picasso

Midjourney generated Picasso

Midjourney generated Picasso

Does AI generate art?

Given everything we’ve discussed, I would say NO, AI doesn’t create art as we traditionally know it, but it can produce something akin to "art." Let me explain further.

If you ask someone, who has never painted before, to create a picture of a tree, that could be considered "art." If it fits the criteria we’ve discussed, someone might recognize it as art, but for most of us, we likely wouldn’t consider it true art.

This isn’t a perfect comparison, but a similar situation occurs with AI-generated art. In fact, it might be worse, because the human involvement is minimal, making the first criterion we discussed nearly non-existent.

AI can certainly generate stunning images, ones that could be classified as "art" because they’re beautiful, evoke emotions, and have conceptual ideas behind them. This satisfies what we discussed in the third criterion.

However, the other key point that’s often missing is the second criterion. Don’t get me wrong, we could argue that by repeatedly using a tool like Midjourney, a human can refine images until they reach a desired result—thus maintaining some level of human input. However, this still falls short of fulfilling the first criterion completely, and the second one is almost entirely absent.

Why is that? Because AI models lack creativity and imagination; they generate outputs based on the data they’ve been trained on. If a model has never seen a Picasso, it cannot generate a Picasso-like image on its own. We could argue that it could still approximate such an image if the model has learned from many artists whose styles converge toward Picasso’s, but this isn’t creativity—it’s merely an average of all those influences.

But approximation is not art; it’s just a COPY of something. AI models are essentially copy machines—they can approximate or create mashups, but they don’t create art.

At the end of the day, AI-generated "art" is not true art. It’s simply a copy or a combination of things, and that’s not what art is. Let this sink in: AI DOESN'T CREATE ART, at least not in the sense we’ve discussed here!

What's next?

Well, let’s see what happens in the near future, what steps artists will take to prevent their work from being used as training data for AI models, and how industries will respond to the use of their copyrighted material for AI training. We’re already seeing some lawsuits over this, and I believe this is just the beginning.

Gary Marcus has a great article on IEEE Spectrum about copyright issues that those image generation models have . Generative AI Has a Visual Plagiarism Problem

These images, all produced by Midjourney, closely resemble film frames. They were produced with the prompt “screencap.

These images, all produced by Midjourney, closely resemble film frames. They were produced with the prompt “screencap.". Generative AI Has a Visual Plagiarism Problem. Gary Marcus

I mainly discussed images, but music is another area that could soon be a target and impacted by AI-generated content. We’ve already seen some AI-generated music, and it’s only a matter of time before we see more of it.

And here is an example! Recently, I was on BlueSky, and someone shared a post about Lollapalooza with band names, and many of them were imaginary. One name caught my attention: "Only Cinnamon Grass," and I thought, let me create songs for this band!

I use Suno and this is the result:

🌱🤘🎸 Only Cinnamon Grass


This is just the tip of the Iceberg.