Magic Avatar, an Artificial Intelligence (AI) portrait app by Lensa, went viral worldwide after it was released last December. After taking a few selfies and uploading them to the software, Magic Avatar provides users with their own selfiebased avatars. However, the app has received considerable criticism for producing distorted images of the original portrait. Accordingly, the Sungkyun Times (SKT) will identify the pros and cons of the AI art industry and discuss what measures are needed for it to coexist with humans.
AI and the Art Industry
-What Is AI?
Oxford University Press defines AI as “the theory and development of computer systems able to perform tasks that normally require human intelligence.” For AI to assist humans or create works of art, AI models must be created and trained by humans. There are three methods of machine learning: supervised learning, unsupervised learning, and reinforcement learning. Supervised learning gives the model a training dataset and allows it to make accurate predictions. By consistently training the model, the AI can later derive the same answers to problems as humans. Unlike supervised learning, unsupervised learning is a method in which models are trained with unlabeled datasets to find patterns that humans have yet to discover. Finally, in the case of reinforcement learning, the model selects an action that maximizes reward in a specific environment, which is a core technology of AlphaGo. These three methods are the foundations for creating AI chatbots and the AI illustrators that we all know.
-The AI Art Industry
AI art, which has recently gained widespread interest due to the AI boom, has actually been developing since the 20th century. In 1973, artist Harold Cohen developed algorithms that enabled computers to create freehand-like drawings, which became some of the earliest examples of autonomous pictures. As personal computers became much more commercialized by the year 2000, programs like Google DeepDream were born to understand how computers analyze and interpret images. Then, the concept of “AI Generative Art” arised with the Generative Adversarial Networks (GANs) method, in which two neural networks compete to give more accurate predictions. Nowadays, anyone can produce high-quality drawings by writing a few sentences using these technologies. These developments in AI art have been showered both with praise and controversies, proving that the AI art industry is at the peak of attention. In the 2022 Colorado State Fair’s Annual Art Competition, Jason M. Allen won the grand prize and seized the public’s attention with his Midjouney AI art generator program drawn art, Théâtre D’opéra Spatial. As an AIgenerated drawing won an award for the first time, it proves that the AI art industry has now become big business. Research firm Coherent Market Insights predicted that by 2028, the current $33 billion digital image market will reach $48 billion.
The Pros and Cons of AI Art
-What Is Good?
1. Growth of the Art Industry
As the AI art generator produces art based on tens of millions of works and continues to store and learn new pieces, it contributes massively to the diversification of the art industry. R. Keith Sawyer, a professor of Educational Innovations at the University of North Carolina, who defines combination as a key element in creativity, stated that AI could invent creative works of art by combining and utilizing existing pieces. Also, images or drawings produced by AI art generators can enter the art industry and boost its diversity. According to Acumen Research and Consulting, the annual average growth rate of the Generative AI market size is expected to be 34.3% and will reach 110.8 billion USD by 2030. In a nutshell, with AI art generators, anyone can easily create high-quality paintings and share them with others, which eventually lowers the huddle to its noble culture. Lee Kyoung-pyo (Bachelor, Department of Industrial Engineering) said, “Being able to illustrate my imagination through the AI art generator Midjourney was fantastic. To draw better, I now check out new paintings of famous artists.” To conclude, the overall demand and supply in the art industry can grow as AI increases accessibility and raises public interest in art.
2. Enhancement of Efficiency
With the development of AI, numerous tools, like AI illustrator, have been developed to help existing industries align with artistic aspects. The AI-based image production function that was added to the NovelAI Diffusion by Anlatan last October is a typical example. NovelAI Diffusion is an image generator optimized to create animation illustrations using images from Danbooru, an imagesharing platform. “AI skills are already professional, and its works make me laugh in disbelief,” said Kim Rak-hee, a cartoonist who serialized Marvel Comics, on his YouTube channel. As such, results can be produced at lower cost and higher efficiency, enabling artists to design even with limited resources. Furthermore, AI can shorten the time to completion and enhance productivity in all businesses that need illustrations or pictures. Russ Maschmeyer, Shopify’s AR/VR Product Lead, revealed a prototype on her Twitter last September that uses textto-image AI that prints images on T-shirts in a matter of seconds. In other words, money previously spent on drawing illustrations for advertisements and magazines can now be utilized in production processes, resulting in enhanced productivity in the overall art-related industry.
-What Is Bad?
1. Copyright Controversy
Discussions regarding plagiarism and copyright infringement have existed since the initial creation of AI art generators. AI art generators function on the basic principle of receiving input data from existing artworks and learning from them to create new images. In this process, an AI generator may plagiarize the artists’ painting styles, resulting in copyright infringement. Stable Diffusion, an AI algorithm for Magic Avatar, is a new service launched by the Lensa application last December. The algorithm scrapes billions of images and texts on the Internet and trains them on AI to recreate new images. However, illustrator Lauryn Ipsum stated through the Washington Post that Lensa copied her color combination, brush touch, texture, and painting style without consent. Also, artists’ signatures were found in some images generated by Lensa, which has led to heated arguments regarding copyright infringement. In addition, it is unclear whether the work made using an AI art generator owns copyright, who owns it, and how the profit should be distributed. Last October, the Japanese DMM management team caught many selling AI-generated images without marking them as AI-generated work, misleading people to think that they had been created by a person. To be more precise, even if an artwork created by AI gains copyright, ambiguity still exists whether the work belongs to the creator or the corporation that possesses the AI art generator.
2. Ethical Issues
As AI constantly develops, ethical problems inevitably emerge. First, algorithm bias can occur when an AI art generator creates distorted images. The Lensa app, which creates a magic avatar from users’ selfies, became controversial when it made avatars brighter and slimmer than the photos users had submitted. In fact, numerous female users have received such distorted images that “Why do female users tend to get results featuring an over-sexualized look?” was often asked on the Frequently Asked Question (FAQ) section of Lensa’s homepage. Besides problems residing in the algorithms themselves, users can easily use AI in unethical ways as there are not many limitations regarding the usage. Anyone can maliciously use AI to create inappropriate photos or pictures of someone without their consent. For example, websites like Mage enables users to create high-quality sexual images by typing a few sentences on the prompt. In fact, there are countless websites that allow users to customize sexual imagery using AI without restraint. To put it in a nutshell, there are no regulations or countermeasures for users creating unethical images using AI art generators.
Preparing the Future
-Solving Ownership and Copyright Issues
A bill that prevents AI art generators from being trained upon artists’ works without their permission is needed to solve copyright infringement problems. A viable solution is to make it mandatory for AI art generator companies to priorly obtain permission from artists and present a certain amount of payment for usage. In addition, compulsory disclosure of the list of works put into an AI’s training database is another way to prevent copyright infringement. As the number of pieces using AI continuously increases, ownership and copyright laws for AI-generated works must be established. For instance, last September, New York-based artist Kris Kashtanova’s graphic novel Zarya of the Dawn became the first ever AI-assisted work to be copyrighted by the United States Copyright Office. As more works relying on AI will enter the market, it would be more realistic to give ownership to the artist of the AI-generated piece and instead make them pay a certain amount of royalty to the AI art generator company. Furthermore, the proportion of AI in the art industry will significantly rise in the future, necessitating the amendment of the current copyright law. Lawyer Park Sang-oh of the law firm Barun also emphasized the importance of revising and refining current copyright laws in order to determine who will be the subject of future lawsuits involving AI-created works.
-For a More Ethical AI
In order to solve ethical problems, AI algorithms must first be ethically designed. AI art generators must be trained to avoid featuring discrimination against skin color, gender, and body shape. Initially, excluding violent or sexual images from the training data set of AI models is also a possible solution. For example, in the past, Lensa’s model was trained with unfiltered data from the internet. However, since last November, nude and pornographic images have been excluded from the training input, making it difficult for users to create sensitive content. Furthermore, it is necessary to enact legislation that imposes substantial legal penalties and immediately suspend services of algorithms that excessively distort avatars into sexual or certain appearances. In addition, stricter regulations on users for creating sensitive content with AI art generators are needed. In particular, legislation should be revised to impose a harsher penalty for utilizing photos to create inappropriate pictures without one’s permission. Finally, ethical awareness education for developers and users of AI art generators must be mandatory. Educating employees about AI ethics at corporations can be one way.
Art and creation will soon become a field that coexists with artificial intelligence. Thanks to AI art generators, creative works surpassing those of humans are being produced, and the general public now has better access to art. Simultaneously, copyright infringement and ethical issues are on the rise. The new era of art that AI is opening for us is just over the horizon.