Written by Isabel Cheng on Digilah (Student Tech Research)
My Bio:
Hello, I am Isabel. I am a rising junior in high school at Stamford American International School (SAIS). As AI-generated art rapidly gains popularity and advances, questions such as its ownership, copyright, and how it applies to existing intellectual property frameworks arise. Hence, I want to explore the different perspectives toward AI-generated art and how that should be applied to the law with my research article below.
Section 1: Understanding AI-Generated Art
If you are familiar with the concept of Artificial Intelligence (AI), you know that they are algorithm models that programmers train on extensive databases. They analyze vast amounts of data fed to them and learn patterns so they know how to respond to act. However, the technology itself cannot think on its own or create original ideas.
Likewise, AI-generated artworks are created through machine learning models. The databases of images and artworks they are fed with are often without the permission of their owners. With that said, how is AI-generated art affecting the artistic landscape, and where does this technology stand in terms of authorship and the legal status of AI as a creator?
Section 2: Popular Generative AI Websites
Let’s talk about some of the most popular AI generative websites. First off, is Midjourney. Launched in early 2022, it is one of the most recognized AI image/art generators. Though it has a $10-20 USD monthly subscription fee, it is known for giving the best quality pictures fast, which has attracted a substantial user base.
Midjourney Showcase Page (Source: Midjourney Feed)
Leonardo AI, also launched in 2022, has a relatively beginner-friendly interface along with refined tools for paying professionals too. Its free version operates on a token system that resets on a daily basis, allowing users to generate images based on prompts. It produces images relatively well based on the prompts, and compared to Midjourney, it is more customizable because it has a larger variety of style options.
Leonardo AI Home Page (Source: Leonardo AI)
Ideogram, another AI generator that is gaining traction, also caters to a wide audience. It has a range of pricing plans catering to users of different needs. It has a free version where you are allowed up to 10 prompts a day. Its consistent performance in text-to-image generation makes it a reliable choice for many users.
Ideogram Explore Page (Source: Ideogram AI)
Dalle-E 3 is another popular image-generative AI in the market, particularly in AI artworks like drawings and paintings. Though it is still not fully developed, you can access it via ChatGPT Plus.
Notably, Dall-E 3 is one of the few AI image generators that states explicitly on its website that it is designed to refuse requests for images mimicking living artists’ styles. They also allow creators to opt their art out of future training of their image generation models.
However, challenges remain in ensuring artists receive recognition. Is a fake death certificate enough to make Dall-E 3 accept requests for mimicking a living artist’s style? What about the artists who are unaware that their work is being used to train AI models?
With that said, let us dive into the growing concerns about the unauthorized usage of artists’ art styles to train image-generating models.
Dall-E 3 Index Page (Source: OpenAI)
Section 3: Artist Recognition and Copyright Challenges in the Age of AI
Since 2023, the advent of AI generation tools has prompted many to be curious and experiment with them. This includes artists, and an increasing number of them are utilizing these tools to enhance their own work, with some feeding their own artworks into AI systems, so that after inputting a prompt, they are able to produce a new piece of art in their own unique style, all in a matter of minutes.
One wouldn’t exist without the other, and this raises questions about the relationship between traditional and AI-generated art.
In Singapore, the Copyright Act protects original creative work. The owner who made an original art piece is usually automatically given a copyright, and they are granted the right to reproduce, distribute, and display the work.
However, when it comes to work generated by AI, a critical question arises: Does the creator of the AI, the user of the AI, the work of artists who trained the AI, or the AI itself hold the copyright? The existing framework primarily addresses human authorship, leaving a grey area for AI-generated work.
As of now, the Singapore law does not explicitly recognize AI as the creator, which complicates ownership and copyright claims to their generated artwork.
Furthermore, Singapore’s legal system emphasizes that copyright protection is only given if the work is original. Since AI-generated art is often trained on databases with existing art, it raises questions about whether the art it generates meets the originality requirement. This could limit the ability of artists to claim copyright over AI-generated work that they incorporate their style or elements into.
As Singapore and countries around the world navigate these complexities, there is a growing consensus that intellectual property laws need to be updated to better fit the world today. Policymakers and legal experts should work together with artists and AI experts to shape a new legal landscape that protects creativity and authorship rights in this new age of AI.
Section 4: Protecting Artists – Tools and Views
Artists are increasingly turning to AI-generative-art poisoning software such as Nightshade, Glaze, Have I Been Trained, and more to protect their work from being used in unauthorized AI training.
These tools allow creators to alter the pixels of their artworks that are imperceptibly different from human eyes but will get picked up by machine learning algorithms and act as a poisoned data sample, resulting in deformed outputs from AI models.
Glaze More Info Page (Source: The University of Chicago)
However, the development of all these poisoning tools is not perfect, and most end up compromising the quality of the artwork. Additionally, some artists view the anti-poisoning software as another unnecessary step, given that another training model could come out to invalidate this approach.
Furthermore, another perspective on AI-generated art is that there is no difference between AI learning and an artist. Most artists take inspiration from other artists anyway and make their own art as a derivative of that.
AI needs artists, but if you look at human history, modern art is built upon thousands and thousands of years of art from different cultures across the globe. Artists need artists too, suggesting that AI could be seen as an extension of this creative process.
Contrary to that though, another valid argument is the idea that AI art is theft? It’s a common consensus that it takes at least 10,000 hours to learn a new skill. Most artists invest years practicing and honing their craft, while all companies do is download datasets of millions of artworks and use them to train their algorithm models.
With that, the models can then make combinations from what they “learned” and instantly generate pieces, often profiting from it through consumers without having done anything artistic themselves.
This situation is unfair to artists, as no human can feasibly digest billions of images and create amalgamations of all of them in seconds. This could lead to a culture of complacency, ultimately stifling creativity and innovation. It is concerning to consider that all “new” and “original” art could end in the next few years if lawmakers do not adequately protect artists’ rights.
However, a lot of this issue involves how AI interacts with the current intellectual copyright laws, further driving the need for policymakers to amend them.
As previously noted, law and legal experts should work together with artists and AI experts to consider adding something like an AI Commission system or requiring AI developers to get permission/credit from artists whose art is used in training datasets.
Conclusion
In conclusion, the rise of AI-generated art is both an opportunity and a challenge for artists and the legal system. Especially with the evolution of AI technology, education, and awareness are crucial for both artists and lawmakers to navigate this new landscape.
Legal frameworks need to reform and adapt so that artists, AI, AI creators, and users get fair recognition for their contributions. As we move forward, it becomes clearer that the ambiguity surrounding artistic authorship around the world has to be changed, fostering an environment where creativity can thrive alongside technological advancement.
References
https://docs.midjourney.com/docs/plans https://leonardo.ai/
https://www.midjourney.com/showcase
https://openai.com/index/dall-e-3
https://dezgo.com/text2image/sdxl
https://lawgazette.com.sg/feature/generative-ai
Most asked questions
Which are the commonly used AI-generative-art poisoning softwares?
Artists are increasingly turning to AI-generative-art poisoning software such as Nightshade, Glaze, Have I Been Trained, and more to protect their work from being used in unauthorized AI training.
How can I access Dalle-E 3?
Dalle-E 3 is a popular image-generative AI in the market. Though it is still not fully developed, you can access it via ChatGPT Plus.
Most searched queries
Midjourney
Ideogram
Leonardo AI
Hello readers! Hope you liked what you read today. Click the like button at the bottom of this page and share insights with your colleagues and friends!
Industry leaders and students researchers, for more such amazing articles and research on technology follow Digilah.
These motifs were re-colored with colors of an inspiration image using a statistical approach of global color transformation, and the design was post-processed to a grid that could be readily used for dyeing, as each cell is of a single color. 