Is Ai-art Art? Can it be Art?

Simple answer is: Not every generated image is art, but some can be.

for example themechanicalcomedian.art to many qualifies as art.

This is because the idea is the art part.
Execution is how you make it available to other people. In combination they can then be judged if it’s successful art or not.
In other definitions “the expression or application of human creative skill” – the skill is in the use of the Ai tool. Skill in using that tool to bring your idea into the material world.
The question whether something made with Ai can be considered art or not, comes down to how much of it is your own idea, and how much is not. Ai is just another medium for making those ideas available to others.

Art History

This definition can be verified from art history over 100 years ago when the same questions were asked. From the movements like Dadaism and postmodernism. Works like Duchamps Fountain, After Walker Evans, and even Comedian (by Maurizio Cattelan) all share aspects to the conclusion of idea being the art part. Notably these pieces are widely regarded as art. (In December 2004, Duchamp’s Fountain was voted the most influential artwork of the 20th century by 500 selected British art world professionals). So by these standards its easy to deduce that the idea is the soul of the art piece. I don’t see any point why an Ai generated image could not be art as well, if the idea itself can be considered art.

Dadaism incorporates randomness and the piece not being 100% planned down to the tiniest detail (as well as many other art pieces that take their final shape during creation). So utilizing a tool like Ai generator that you usually don’t know for sure what it spits out, but is based on your core idea, is completely okay according to historical art movements and the processes of art creation in general.

Duchamps Fountain and other ready mades are all about the idea of the piece. People were not interested in who created the items. They were interested in who presented them, and in what light. Comedian on the other hand gets remade by staff over and over again when the banana goes bad and the tape loses it’s stickiness. Yet Maurizio is still the artist and the creator of the current Comedian, not the staff members who re-create it because it’s the idea that matters. After Walker Evans is just pure self-pretentious theft imo. It’s the peak ridiculousness of postmodernism, but the art world has seen it as art solely based on the ideas behind the work, so there is that.

Comissioning art

Now some people like to compare generating image to commissioning an image from another artist. This is tempting but misleading.
Ai image generators are not sentient beings or entities that insert their own visions on top of your idea, they are not alive nor actually intelligent. Ai image generators simply output a fixed image based on your fixed input and your input alone. This is precisely similar functionality as a camera and you are not commissioning photographs from the camera are you. Using words as the method of operating the Ai generator does not change this, as photography doesn’t become commissioning even if for example a disabled person uses a voice controlled camera.
The program just runs its code where it makes different shapes based on different inputs. The prompter is the only sentient being affecting the process and that make’s it a tool.

Ai as a tool

You can think Ai image generator just like a camera, but instead of taking your camera to the place you want to take a photo of, you bring the place to your camera. Either with words or with and image or other data files. Just like a camera takes in light and turns them into an image in film or 1 and 0’s. An Ai generator intakes words or other files and turns them into 1 and 0’s.
Now Ai image generators cannot output things they lack in their training data. This is just like a camera cannot capture wavelengths of light that its sensor or lens are not manufactured to capture. This is just a limitation on the tool and something the person using it has to take into account when trying to execute their idea into the real world. The “skill” in using Ai is to know the limits of the tool. Using models that allow you to adjust the parameters of the process instead of automated models like Dall-E 3, is similar to moving from automatic mode to manual mode on a camera.

For example Stable Diffusion with ComfyUi has Area Composition, InPainting, controlNet for poses along many other variables you can control. You can read more here about the ComfyUi controls for Ai image generation. But lack of certain amount of control is the trade off for automation and limitations of the tool, and just needs to be accepted at this time. It’s similar to choosing a catapult to yeet paint into a wall for abstract art, it has randomness built in it and is part of the idea, if you cannot accept randomness you should choose another tool.

Ai and ethics

From that we come to the ethics of Ai image generators.
The training process and output itself simply are not stealing. The whole point of Ai image generation is to abstract the contents of the dataset by learning the shapes and their form. This way you can output shapes and styles without having to create copies or uses samples of existing works. The training process is like a human learning from examples and that is not stealing. Even if the images are influenced by the images in the training data they still are new pieces. If they aren’t the Ai generator has failed its purpose just like a human has if their art is a rip-off of someone else’s work.

Now the actual problem comes from how the images are brought into the training dataset. In many cases these images are outright stolen without payment to the creators.
Downloading images without permission is theftr whether you like it or not. Using the images in training requires physical access to the images thus them needing to be downloaded. This is usually done by simply scraping the images without payment and by law again this is theft. Link to further sources proving this: https://www.lerattus.blog/how-ai-companies-use-copyright-infringement-to-steal/

Then we have the “ethical” datasets where the images are selected to actually be either free domain or licensed. Now many of these like ShutterStock Ai and Adobe Ai who claim are trained with ethical datasets, still have practiced ethically questionable behavior for the licensing parts at least. Usually these “licensing’s” are done by offering another service, for what people might even pay the companies for, and there having a pre-selected opt-in clause for your images to be used in training. This pre-selected “yes” is not ethically sourced. Sometimes these clauses are added to your account after you have uploaded them meaning unless you check your options everyday they might have already used your images in training without you actually ever consenting to it.

Why artists don’t want their works opted in?

Imagine you have used years of your life in order to study for your job and R&D. Now without compensation to you someone can use all your acquired knowledge and skills to flood the market with adapted works of your product, and drive your market share to the ground. Furthermore, the problem is that now your niche is flooded by mediocre imitations of your work. This means innovation will be hard as it’ll easily drown in the noise. A lot of the generated images will not be art. They’ll either be content or products that’ll hog up people’s attention away from actual art and screw with the algorithm of many of the media platforms artists use.
The problem is that the people who are the main reason the development of these tools were possible, are not compensated accordingly, but are instead being taken advantage of and losing visibility and thus income at the same time.

This is the industrial revolution for the non-realism art world just like the camera was for the portrait painters. A lot of people will go out of business.

 

In conclusion.

Art is in the idea and execution is how you make it available to others. Executing your idea with Ai does not disqualify the idea itself from being art. Ai generated images might not qualify as ethical art, or in many cases as good art due to their limitations, but that does not mean they cannot be art.
If you do Ai art, well firstly be sure it’s actual art and not just content.
Secondly try to use the most ethical generator that you can find. At the moment Shutterstock Ai does compensate the artists whose works were used on the dataset. Even if they might have been misleadingly entered there.
Thirdly respect those people who will lose their income, and whose life works are being used to replace them without them getting anything from it. There is no need to be salty against them. They are the reason these tools exist and imo should deserve a part of the cake.

 

Suggestion for solution:

By law define Ai datasets to be auditable thus allowing cracking down on copyright infringement and steering these models to a healthier business model.
Furthermore require a training license for all images in the dataset.
This training license should force for lets say 80% of the price of the generated image to flow into a pool. Now from this pool it would be distributed equally to everyone’s % of images they have in the dataset (meaning more of your images = more share for you) with the free-use images removed. Now with correct labeling it would be quite possible to divide the images and their creators in the categories and pay a share by the prompt for lets say “Beautiful dragon in a forest at the style of X” you’d focus the revenue sharing for 33% to the whole pool and 33% targeted to those who have labels suiting the prompt and final 33% for the style of X artist.

These changes to business model of these companies would ensure that artists and the Ai companies both would benefit ethically from generative Ai as it seems to be here to stay.