You’ve heard about AI-generated art. Maybe you’ve even played with some of the commercially available AI art tools. But are we really artists if the AI is doing much of the work? Should we be embracing AI-generated art or rejecting it? Is AI-generated art really art at all?
Let’s find out.
How Do AI Art-Generators Work?
AI art-generators grew out of image recognition programs.
The developers trained the image recognition programs by showing them millions of images and descriptions of those images. The programs absorbed this information in order to classify what things look like. Eventually, the AIs could ‘understand’ similar images and what they represented.
So, for example, they knew what an aeroplane looked like. More impressively, when shown a new photograph, they could say whether it was of an aeroplane or not.
Image generators built on this image recognition model, but reversed the process. So, because the AI knew what an aeroplane looked like, it could output an image of an aeroplane. For example:
Although this is identifiable as an aircraft, which was good enough for a recognition program, it doesn’t really work aesthetically. It’s distorted and there’s two engines on one wing and only one on the other, for example. So, there was a way to go.
But the things that an AI can recognise are not just objects like aeroplanes, it’s also concepts, styles, anything really, and that’s where it gets interesting thing from the point of view of an artist.
The eureka moment was when the developers realised they could ask the image generators to combine objects and concepts. For example, they could ask the AI to output an image that looks like an aeroplane, and looks like a Van Gogh painting:
This was when it became possible to create art using AI.
But is it a Good Thing?
Some people say that pictures like the one above either just aren’t art, or that using AI to generate art is somehow wrong. The arguments often used are:
- AI-generated art isn’t real art.
- There’s no artistic skill involved in creating the picture.
- AI-generated art is soulless. It has no emotion or artistic process.
- Any AI generated art is by its method of production, derivative, unoriginal and banal.
- AI-generated art just isn’t very good.
- AI-generated art will destroy the livelihoods of real artists.
Let’s explore some of those issues.
AI-generated Art Isn’t Real Art
That AI-generated images are not art at all is not a sustainable argument.
Artists have presented innumerable unlikely things as art. Marcel Duchamp created artworks using unaltered everyday objects, for example.
What makes anything ‘art’ is entirely subjective, and it’s impossible to identify ‘real’ art.
AI-generated Art Lacks Artistic Skill
If an AI is generating the picture, then clearly, the artist isn’t supplying as much of their own technical ability. There’s potentially no technical skill at all involved.
But then again, did Tracey Emin use much technical ability to put her bed in an art gallery? Does photography lack artistic skill because you don’t need the same hand-eye-coordination to take a photograph as you do to paint a picture?
Art can be powerful, emotional, and exciting without being difficult to produce. Damian Hirst’s assistants produce his spot paintings by the dozen on what amounts to a production line, but they still speak to lots of people.
This goes to show that technical ability doesn’t define artists, and that art is not just about how hard it is to produce. The AI might facilitate the work, but it’s still a product of the artist. Just like a photographer using a camera, the AI augments human creativity rather than replacing it. The role of the artist is like a photographer choosing images that suit their vision and discarding those that don’t.
I’d also argue that removing technical ability from content creation democratises art. Art doesn’t have to be produced for the rest of the world. People can produce images for their own pleasure. If they want to do that by supplying an idea to an AI generator, picking the image that suits their vision and then tinkering with it in Photoshop, how is that wrong?
AI-Generated Art is Soulless
Again, this is entirely subjective. It’s true that until the invention of conscious AI, they will lack intentionality, but philosophical questions about concepts like intentionality are notoriously hard to formulate. Which reminds me of the Turing test, which is more complex and subtle than people realise.
Turing proposed that the question “Can machines think?” requires a definition of what “thinking” is, which is philosophically difficult. He proposed a more pragmatic question: “Can machines do what we (as thinking entities) can do?”
His test then proposed that if the responses an AI gave to questions were indistinguishable from the responses humans gave, then the AI was, for all practical purposes, alive.
In the same way, I propose that “Can machines produce art?” requires a problematic definition of what “art” is. Instead, we should ask a more pragmatic question: “Can machines produce art as we (as artists) can?”
If an AI can produce art that’s indistinguishable from human art, then the AI is, for all practical purposes, an artist.
AI-Generated Art is Derivative
There’s some truth in this. AI-generated art is unlikely to be startlingly original and innovative. At the moment at least, no AI will invent its own school of the Avant Garde. At least until AI gains true sentience, no AI is going to sweep the art world with a ground-breaking trend like cubism or impressionism…
…but then again, neither are the overwhelming majority of artists.
AI-generated Art Just Isn’t Very Good
It’s certainly true that AI-generated images can be odd.
This isn’t necessarily a problem if the artwork is supposed to be surreal, but the more representational the artwork, the less likely it is that an AI can produce it without errors. For example, human figures with anatomically impossible bodies are very common.
In my experience, for every image that’s halfway decent, an AI generates a dozen that are unusable. Even the usable images need a human artist to edit them to make them more aesthetically pleasing, also known as ‘fixing them’. Occasionally one comes out perfectly, but most only have potential, along with several problems, such as misshapen bodies, odd faces and colours not being aesthetically pleasing.
This is where the human artist using the AI comes in. Selecting the images with promise is the role of the human artist, just as it is for a photographer, and the artist then has to manipulate the AI-generated image in Photoshop or a similar tool to make it work.
As long as a human has to select and edit the AI-generated images, i.e. until AI gains true sentience, AI art generators are simply tools that human artists can use, aids that help to create visual images.
In this way, AI art generators are analogous to cameras. A photographer can use a camera to take snaps, or they can use it to make superb art. Similarly, an artist can use AI to produce images that are banal or that are incredible.
An Example
Here’s the process I used recently to make a book cover:
- Get the AI to generate a bunch of images (lefthand image).
- Choose the image that has the most potential.
- Edit the image. If you look carefully, you can see that I’ve made a bunch of changes to the AI-generated image. I’ll expand on this process in a future article.
AI-generated art will destroy the livelihoods of real artists
I doubt this is any more true than that Photoshop destroyed the livelihoods of artists. Remember, the skill of the artist is not just in the production of images, it’s the selection of meaningful images.
Also, computers and robots have already affected lots of jobs. For example, robots have already replaced lots of factory jobs, and AI drivers may well replace human taxi and truck drivers soon. Why should artists be exempt from this trend? To announce that artists can’t have their jobs changed by computers seems like special pleading.
Ultimately, art has always been a difficult way of making a living, and that’s unlikely to change. AI tools might actually help if they speed the artist’s process so they can make and sell more art.
Conclusion
I’ve tried to argue in this article that, at least until AI gains true sentience, AI art generators are simply tools that human artists can use, aids that help to create visual images.
I’ve suggested that although the argument that AI-generated art is derivative has some weight, the other arguments against AI-generated art are questionable. They suggest a narrow approach to what art is and what the role of the artist is.
In the end, it seems hard to argue that AI-generated art will have any greater effect on artists than the invention of photography did. Rather than damaging art, AI, like photography, may open new fields for the artist. We can integrate AI into our artistic process where appropriate, or we can decide not to, just as many artists don’t use cameras.
Discover more from Graeme Shimmin
Subscribe to get the latest posts sent to your email.