.jpg)
If you are connected to any art related forum on social media, like comics or paintings, chances are, you may have read about this ongoing debate about banning AI created artwork. The artist community is very vocal about this, as they claim that AI engines are basically stealing their work and letting anyone create similar images based on their own art. Some of them tagged AI work as plagiarism as this new art created by copying existing artist’s work with no permission or compensation.
Recently, Australian Supanova Comic Con banned AI generated arts from being sold at their convention. Kickstarter, a portal where new projects are listed for crowdfunding, suspended the project “Unstable Diffusion: Unrestricted AI Art Powered by Crowd” as they would wait for more clarity on AI generated artwork’s stand on copyright and plagiarism. DeviantArt changed the policy for its AI engine DreamUp to not use any of the member’s artwork for learning. Earlier, every member had to individually opt out from AI using their artwork on the site for learning. And a much recent interesting case is related to a copyright claim from an AI artist. The United States Copyright Office (USCO) has initiated a proceeding to reverse an earlier decision to grant a copyright to a comic book that was created using "A.I. art", mentioning that copyrighted works must be created by humans to gain official copyright protection. Although the final decision is still pending and the copyright will still be in effect until the proceeding is completed.
To understand this issue, let us first understand how an AI Image Generator works. It is similar to any other AI engine, that takes thousands and thousands (maybe millions) of available artwork, uses its complex algorithm to generate data points, and once it reaches to a level where a decent structural data is ready to be used, it starts creating new art based on keywords given by the user. So, basically, you can go to any AI Image Generator, and give a prompt like - “lady in red dress, standing near closed window, looking outside into the raining forest”, and it will create an original artwork with raining forest outside the window, and a lady standing on the window in red dress. You can also specify any artist name like Van Gogh, along with your input, and it will create your artwork in Van Gough’s style. In simple words, an AI Image Generator Engine is trying to mimic human artists by learning from existing artwork, and then using that learning to create a new artwork.
Now it may sound as if AI is trying to copy an existing artist’s artwork, but this is exactly the same path any new artist follows to become a professional artist. Any new artist studies existing artwork from various artists, trying to practise and enhance their skills by creating similar artwork multiple times. Basically, they train themselves using available artwork from different artists. So, if the AI engine is doing the same thing, learning and enhancing its skill using available artwork, would that be plagiarism or considered as copyright infringement? The answer is YES, and here is how it is different from an individual taking inspiration or learning from an existing artwork vs AI using Machine Learning from the same set of artwork.
AI image generators, such as those based on Generative Adversarial Networks (GANs) or Variational Autoencoders (VAEs), are trained using a dataset of real images and a machine learning algorithm. The goal of the training process is to learn the patterns and features present in the training data, so that the AI model can generate new images that are similar to the ones in the training dataset. First, the model is fed a large dataset of real images. This dataset might include images of faces, landscapes, objects, etc., depending on the task the model is being trained for. The model then processes the images and learns to recognize patterns and features present in the training data. For example, if the model is being trained to generate faces, it might learn to recognize features such as eyes, noses, mouths, and so on. Once the model has learned to recognize patterns and features in the training data, it can then generate new images that are similar to the ones in the training dataset. For example, if the model is asked to generate a new face, it might use the patterns and features it learned from the training data to create a synthetic face that looks similar to real faces, but that does not directly copy any specific face from the training dataset.
For example, based on your text prompt, AI will choose to create one part of the face from one learning, whereas it will combine it with another learning as you wanted to show something like a scar or maybe the nose is different in your task. Similarly, AI will combine its learning from various images, and generate something completely unique and original, still using the images related data points it has consumed. Just because AI does not have the capability of creating something new, it is simply applying snippets of what it has consumed earlier. So, it is effectively copying images from different artists, and creating a new image. This is not the same as training new human artists and creating their own art even if that is inspired by some other artist’s artwork. Although, when some new artist creates something that is even inspired, the expectation is that they should credit the original artist for their inspiration. In the case of AI, it is not inspiration, but stealing the images from existing artists. And as per copyright law, it is illegal to do so. Although as we know, any legal battle is going to take time, and not sure how easy it will be to prove that an AI created image is copied from someone’s work. The problem is, it is not illegal to study the available art for research purposes, and the biggest AI image dataset is open source, created for research purposes. Basically artists today are fighting a battle in the 21st century, and they have laws to protect their interest written in the 19th century.
There are few arguments that put AI as a tool like Photoshop, but it is actually not. Yes, AI engines are a tool, but not for artists. It is a tool for non-artists to generate new artwork by simply defining some ideas. Photoshop or any other image editing tool enhances the artist’s capability, letting them do their job much faster and with more ease. This engine cannot let artists do anything, but provide some instructions to replicate images from various sources, and generate a new one.
Having said all these things, ban, artist protests, legality issues and all, the AI revolution in the art industry is inevitable. Just like the Industrial revolution, we are witnessing the AI revolution, and it is not just art, soon we will see the impact of this in other areas of society. We can regulate it, refine the laws to protect the work of artists from being “consumed” by these AI engines without their consent. But we cannot stop AI from being the largest producer of artwork in the near future. The artists community will struggle, and the most impacted individuals are the students or new upcoming artists, still trying to learn and hold their foot into this world. We need to accept the fact that AI is not going away, and with regular enhancements, soon it may not need to “consume” or learn from other artist’s artwork. Take the example of any 3D animation engine or game engine, these are sophisticated tools that can simplify your 3D animation, or game play without the need of any reference material. Yes, you need to create your characters, but that too is way too simplified in these tools. You get the base structure ready, and just have to do minimal work to get your characters into action. Similarly, an AI engine can use free available artwork to create its dataset, and use that to create new original artwork against simple prompts. Maybe some AI engine companies will pay some artists to create and provide artwork for their AI to consume and enhance the capabilities. It looks like we can address the ethical and legal part related to these new AI engines, but the fact remains that there is a strong future of AI generated art. And this future is not very bright for lots of existing artists as they will find less assignments due to AI taking over the majority share of their workload. There will be non-artist people or companies that will get these assignments, and will run past multiple prompts to generate multiple new artwork in a short span of time. You may still need some artist to do touch-up or editing on the AI generated artwork, but the work and pay for these artists will be very less. Similar changes have already happened in other fields that have disrupted the entire industry and taken away almost the entire job of those communities. The manual work of craftsmen like carpenters or blacksmith is replaced with machines creating all kinds of stuff much faster and cheaper. A similar creative example would be photography. With today’s mobile phone camera and tons of filters, anyone without any knowledge of photography can become a professional photographer. High end video cameras in your mobile phone have resulted in many individuals creating really good content videos without any knowledge of video shooting or editing.
Any new technology or tool will create some disruption in the existing way of doing things. Even in the field of art, the introduction of tools & processes like Photoshop, digital coloring had impacted the existing artist, as the new generation of artists were much more efficient and started providing faster results using these new techniques. It was obvious that the majority of new assignments went to these new generation artists.
Today, this AI tool is very new, and more or less at experimental level. But with more and more refinement, it will continue to provide better results. There is a need for clear law around creation and use of AI artwork. For example, mandatory disclosure of AI generated art, no copyright should be issued to AI art, and penalty for tagging AI art as human artwork. Similarly it should be illegal to use dataset generated using artwork of artists without their consent for producing any commercial artwork. The AI engine should only use license free artwork or pay artists for using their artwork for machine learning.
If the artists need to survive, they need to learn this new tool, and adapt accordingly. They have the capability to edit and update the AI generated results, to make it way better than any non-artist producing it from an AI engine. I will quote an example of photography. You can experiment and click nice professional looking photos using your mobile camera and filters, but you know that for any professional job, you need to hire a real professional. This is going to be the ruleset for the artist community as well. Any client willing to get the artwork done, will be hiring real professional artists, not some kids who can experiment and will have no control on the outcome. So, in order to survive, the existing artists need to be in control of AI tools and able to produce artwork according to their client’s need, no matter if that is generated using AI, or manually, or a mix of both. This is only possible when they start using AI as a helping tool, and then enhance the results using their existing artistic skills.
In the end, you call for its boycott, you blame it for plagiarism, you sue them, but you simply can’t ignore these AI engines. This technology will get more and more enhanced over time. So, it is better to accept this new force and use it to your advantage.
Comments