What are the ethical implications of AI generated art?
Artificial Intelligence (AI) has become an integral part of various industries, from finance, healthcare, and transportation to entertainment and education. The creative industry has also been vastly influenced in recent years by artificial intelligence, revolutionizing the art industry and bringing new possibilities for creative expression. As artificial intelligence enters the art world, it brings new modes of artistic creation, new challenges to traditional forms of art, new artists, and new audiences, which raise questions about the extent to which AI will impact the art industry. The concern is about whether AI tools will enhance and encourage creativity and innovation or whether they will become increasingly autonomous and overtake human art.
Artificial Intelligence Art
Artificial Intelligence Art AI is a technology that enables computers and machines to simulate human intelligence and problem-solving capabilities. The technology, which is based on machine learning algorithms, processes large amounts of data and develops sophisticated AI algorithms that allow computer systems to automate different tasks such as speech recognition, identification of patterns, decision-making, and problem-solving. Hereby AI is defined by its ability to learn from data and adapt its behavior accordingly. More sophisticated than AI is generative AI (GAI), which not only processes and identifies data but also creates entirely new content, such as text, images, videos, and other media, in response to inputted prompts. Over the past years, several AI generators have been developed, such as ChatGPT or DALL-E 2, which have gained worldwide popularity due to their facility in assisting humans in writing, researching, and designing.
With respect to the art industry, AI has the ability to create artworks in the form of text-to-image generator tools that generate original images based on textual input provided in natural language. These generators draw their outputs from a massive database with an extensive amount of images and other human-made contributions, which make up the training set. Through a training process, the generators learn different characteristics of the training data to eventually generate completely new and non-repetitive images. Through continuous iteration, the system refines its creations and produces highly complex artworks. It combines and mutates the inputs in such unexpected ways that it is capable of creating new aesthetics that challenge traditional art forms and push the boundaries of human taste. Nowadays, a broad range of text-to-image generators is widely available, such as DALL-E 2, Stable Diffusion, and Midjourney. DALL-E 2, developed by OpenAI in 2021, counts as the first text-to-image software to capture widespread public attention.
Art creation and appreciation have often been constrained by access, skill, and knowledge barriers. Being accessible to anyone, image generator tools have the potential to break down these barriers, expanding the possibilities for art forms and offering any individual the chance to experiment with art in new ways and create something new. People who have never picked up a paintbrush can use a simple text prompt to generate a portrait in the style of Vincent Van Gogh. This provides the possibility for art to be produced on a much larger scale and at a lower cost than traditional art forms, thereby further increasing the accessibility to the production of art to a wider audience. This vast access not only increases the possibilities for art creators but also for others to witness and appreciate different forms of art. In addition, AI tools have the ability to assist artists by improving the effectiveness of the creative process. Through the automation of certain tasks such as color selection or composition, more focus can be put on the conceptualization and creativity processes.
Ethical Considerations
Even though AI-generated art has the potential to revolutionize the art world and pose new possibilities for artists and art appreciators, the use of AI image generators also raises significant ethical implications and debates.
One of the major concerns revolves around the ownership of AI-generated artworks and whether it should be assigned to the creator of the AI work, to the AI developer, or to the artist whose images have been used in the creation of the AI artwork. The latter is related to the responsible use of data in AI art creation. The training data used by AI tools in the image generation process is often sourced from existing works, which raises important concerns about copyright, intellectual property, and the value of human creativity, as the use of this data without consent can lead to infringement of rights and violation of privacy. AI tools may replicate protected elements and styles without proper accreditation to their rightful owners, which infringes on copyrighted material and affects compensation for original artists. In fact, most AI tools are trained to recognize and utilize large datasets of images collected from the internet without any indication of whether these images are originals or have been used unlawfully. Adobe has built its own image generator, Firefly, around training sets consisting only of public domain and licensed works, while others like DALL-E 2 or Stable Diffusion have made no efforts to license material in the training data before making their GAI models available to the public, thereby profiting from the use of artists’ work and depriving them of their income. This has resulted in multiple lawsuits that have been filed against these companies. The most relevant lawsuits regarding image generation and copyright are Andersen v. Stability AI and Getty Images v. Stability AI, concerned with infringement at the level of the training data and output.
When it comes to the licensing of training data sets, diverse perspectives come into play. On one side, data licensing is crucial for the protection of intellectual property rights of original authors and for respect toward their rights in receiving commissions for their works used by AI tools. Nonetheless, there are multiple arguments against the licensing of training sets. Some consider the technology to be a net social good, with the purpose of encouraging creativity, research, and innovation. Others believe the scope of the material used in training sets to be far too broad, making licensing overly burdensome and complicated. Moreover, machine learning is seen to be so transformative to the extent that it completely changes the purpose for which the original works are used, decreasing the importance of licensing the works used.
Although this argument lies within the attribution of copyright, there are also countries in which AI-generated works are not even copyrightable. For example, the US Copyright Office states that copyright law only protects the “fruits of intellectual labor” that are “founded in the creative powers of the mind”, therefore excluding non-human works from copyright protection. Several works have already been revoked from copyright registration once the US Copyright Office learned that they were produced with the use of GAI models, such as Kris Kashtanova’s graphic novel ‘Zarya of the Dawn’ as you can see below, whose illustrations were created using Midjourney.
Nonetheless, this American perspective might undermine the creativity and dedication involved in writing prompts. The high complexity inherent in the use of AI tools requires skill and creativity in the generation of GAI prompts. For the user to achieve the exact desired output, they must combine different text elements and undergo a long iteration process. In addition, the skills required to effectively craft prompts for one AI tool type significantly differ from the skills needed for another type. There are already individuals and businesses that advertise their skills in writing effective AI image prompts. All this questions whether these prompts should be considered creative works by themselves. Just as creating art requires skill, so does crafting the language that guides the AI tool to create what the user attempts to achieve and see. In this sense, the output of a GAI model could be characterized as a derivative work based on the independently copyrightable prompt. If then the AI-generated artwork is not copyrightable, then the prompt might be. However, giving the prompt writer the sole credit for the entire output could undervalue the mechanical contribution, which is why GAI models should also be considered co-authors of the works. In this regard, it is important that policies encompass both the input and the output stages of GAI.
A further important issue around AI-generated art is the cultural and social impact of AI tools on broader society. This is related to the possibility of AI-generated art leading to a homogenization of art, a decline in human artists, or a perpetuation of biases and stereotypes. With regard to the latter, biases and harmful data are evident to be inherent in the training data sets, especially in the form of race, gender, religion, and abusive, hateful, and violent images. Because of AI’s lack of a moral compass, these biases and stereotypes are then reproduced by AI image generators in their outputs, raising issues about the violation of moral laws. Can machines produce art that is offensive and dangerous, or can it lead to a devaluation of human creativity and loss of artistic and cultural diversity?
An additional concern that might bring significant social and economic impact is the threat that AI art tools pose to artists, content creators, and experienced graphic designers. As companies may consider AI to be the cheaper and easier alternative, demand for human creatives may drastically decrease, causing thousands of creatives to lose their jobs. This poses the question of whether AI will replace human artists, compete with them, or facilitate collaboration among both. Are machines tools used by humans to enhance creativity, or are they becoming increasingly autonomous and independent, capable of creating art without any human intervention?
The way AI art is valued poses a different cultural impact on the art industry. In 2018, the first-ever AI-generated artwork to come to auction was sold at Christie’s for $432,500, highly exceeding its estimated price of $7,000. The artwork ‘Portrait of Edmond de Belamy’ was created by the arts-collective Obvious using a Generative Adversarial Network (GAN), questioning the value, authorship, and originality of the work. While traditionally the value of an artwork is determined by the artist’s reputation, culture, experience, skills, and emotions embedded in the work, the anonymity of AI makes the generated artwork much more complex to be valued. Is AI art simply lacking in depth and meaning, or are machines capable of creating art that is just as valuable and meaningful as human-generated art? The focus shifts from the artist’s reputation to the artwork itself and the emotions it evokes, emphasizing the viewer’s subjective experience, leading to a more personal engagement with the artwork. In this regard, AI art forces the art market and the art community to reconsider the valuation paradigms of AI-generated art, such as emotion, technique, and originality. Especially originality poses an important concern for AI-generated art. Being ultimately generated by machines, this new type of art is discussed not to be truly considered creative or original. Even though algorithms are able to produce continuous new variations and combinations, the outputs originate from existing works, which is why the output will ultimately never be entirely new or unique. But what contributes to the complexity of valuing AI art is the lack of transparency when it comes to the process of AI art generation. Even though the training and image generation processes are known, the parameters and decisions taken by the AI tool remain unknown. For an effective valuation, both consumers and artists need to understand how the AI artwork is created, exactly what data is used for its creation, why and how certain works are combined.
The generative adversarial network (GAN) portrait painting “Edmond de Belamy “
A Future for Human-AI Coexistence
When it comes to the role of technology and machines in creative processes, important concerns arise around the originality and ownership of artworks, and the potential economic, cultural, and social impacts on human artists and society as a whole. The question of AI art arises as to whether human artists and AI will eventually be able to coexist harmoniously, enriching the art world with new possibilities without harming society as a whole.
From the diverse ethical considerations discussed above, the licensing of works in AI training sets seems to be one of the most important concerns when it comes to AI-generated art. Even though the absence of data licensing provides free access to AI tools, the infringement of IP rights is unlawful and can lead to diverse negative consequences for culture and society. Unlicensed works in AI training sets not only disrespect original authors’ rights but also facilitate an overtaking of AI art over human art, posing a risk to the elimination of human artists and to the devaluation of human art.
In this regard, licensing the data contained in training sets indicates the only plausible approach that will allow the protection of IP rights and the safeguarding of human art without hindering technological development. Promoting the intellectual property rights of original authors and licensing of their works maximizes the happiness of creatives and minimizes the suffering from the negative consequences of disobeying the law.
As current licensing strategies are ill-suited to deal with this emerging issue, new models of licensing tailored to the protection of works used in training sets should be considered. In addition, a further consideration could be the licensing of prompts. Given the complexity and skills involved in the creation of GAI prompts, only the licensing of these would respect and protect the rights of prompt creators, especially if the final AI artwork is not copyrightable, as in the United States.
Instead of feeling threatened by AI and its possible impacts on art and society, artists should embrace the opportunities it brings and take advantage of its capabilities to enhance and evolve creativity. A careful collaboration could combine human emotion and AI precision with the result of new and interesting art forms. The ultimate objective for AI in the art industry is to expand human creativity rather than replacing it, encourage creativity and innovation, while safeguarding the rights of creators and respecting the essence of human creativity.
