Artificial Intelligence-Generated Art

Artificial-intelligence-generated art, or AI art, is any piece of art generated through artificial intelligence. AI art is constructed using machine-learning algorithms, which are self-learning. These algorithms incorporate elements from digitally available images and artwork. A user will provide specific parameters that train the algorithms to look at a certain selection of these images. These parameters, or prompts, are mainly provided to the algorithms by the user in the form of phrases (text) or images.

The specific method through which the image is produced depends on the decoder that the algorithm is trained on. For text prompts, the algorithm can put the text through an autoregressive or diffusion prior. The autoregressive or diffusion prior then produces an image embedding, which is then fed to a diffusion decoder. The decoder produces the final image based on the image embedding. For image prompts, the image can also be fed to a diffusion decoder to produce the final image. Neural networks can also be trained on the style of particular artist and used to produce images or videos that emulate that style. For example, the platform DALL-E is trained on an OpenAI model called the Contrastic Language-Image Pre-training (CLIP). CLIP serves as a bridge between text and images.

Many of these platforms are available for use by the public. AI can be used to produce a unique image that has never been created before or illustrate a concept that would otherwise be expensive to create through traditional art. Some of the more well-known AI art platforms include DeepDream, DALL-E, WOMBO Dream, GauGAN2, ml5.js, Midjourney, Stable Diffusion, Artbreeder, and Art Recognition. Some of these platforms require a subscription to use, while others are free.

Some platforms encourage users to explore the use of unexpected or creative phrases or combinations of phrases. On the website DALL-E, the main page provides example prompts to stimulate the user’s imagination, with phrases like, “An astronaut,” “riding a horse”, “as a pencil drawing”. With simple tutorials like these, users can recognize that they can input a subject for the art as well as style of the art.

Ethical Dilemmas

There are ethical dilemmas arising from the questions of ownership and the morality of the sources from which the algorithms pull their data and over the finished product. It is unknown how effective these platforms are at vetting the finished products and ensuring that all the content follows these guidelines.

Copyright Infringement

Some platforms have guidelines that explicitly say that users should only share photos and videos that the user owns or has the right to share. For example, Deep Dream Generator says that users should “post authentic content” and not post “anything you’ve copied or collected…that you don’t have the right to post.”

The terms and conditions of the platforms WOMBO and WOMBO Dream state that they do not allow content that infringes copyright. However, they also recognize that there are exceptions to copyright infringement that allow the use of copyrighted content without authorization. In certain cases, these exceptions are provided by under the fair use doctrine in the United States. If users believe that their copyrighted content is being infringed on, they may submit a claim to the company.

Data Scrapping

Data scrapping is also a concern. In regards to AI-generated art, companies that host these algorithms may use data scrapping or extraction tools to provide images to train the algorithms. However, the sources from which the companies take the art can include copyrighted content or even sensitive materials, such as private medical or police records. These materials are often taken without the creators’ knowledge or consent. For these creators or the people whose records are being used, it is difficult to get any recourse to remove the materials from the data set.

NVIDIA, the parent company of the generator GauGAN2, expects that users will not use any data scrapping or extraction tool in posting, submission, creation, or transferring any content to and from the platform.

Harassment and Discrimination

Many algorithm platforms have community guidelines or usage policies which state their anti-harrassment and discriminatory policies. This includes, but is not limited to, using data or producing content that relates to sexual intercourse, sexual content involving minors, violence or discrimination aimed towards a certain group, unlawful activities, deception, and threats to personal safety.

Kim Jung-gi Case

to be written

Responses to Ethical Issues

Previous Court Rulings

In previous rulings, courts such as the have determined that AI cannot hold copyright. A review from the US Copyright Office ruled that a piece, titled “A Recent Entrance to Paradise”, generated by an algorithm called “Creativity Machine”, could not be registered as a copyrighted work. The review discussed what the phrase “original work of authorship” means and concluded that the piece lacked “human authorship”, which is needed for protection under copyright. In terms of AI-generated art, this can mean that the art produced by these algorithms belong to the public domain.

This is especially concerning for artists who are worried about algorithms that can copy the style of a human artist. This issue is in line with the growing concern of automated jobs. Some artists have used AI-image generators to produce images of Mickey Mouse and other copyrighted characters in order to bring the issue of legality and copyright into the forefront of the conversation around AI art.

Ongoing Court Cases regarding Copyright Infringement

In the week of Jan 15, Getty Images filed a claim against Stability AI, creator of the algorithm Stable Diffusion. In its statement, Getty Images says that Stability AI infringed intellectual property rights, including by using content owned or represented by the former without seeking the appropriate license to use such content. The former also states that the content included, but is not limited to, millions of images and the associated metadata.

Resistance to Data Scrapping

In response to concerns that algorithms may be using creative content without the consent or knowledge of the artists, a group of artists named Spawning created a website named, “Have I Been Trained?” On this website, artists can search the LAION-5B data set by text or by uploading an image to see if their content is being used.

Added by peer on 1/26: 2: There are several different companies that specialize in AI generated art and the most popular of these is an app called Lensa. Lensa has taken social media by storm, creating custom portraits based on ten images that the user inputs. Lensa used an AI technique called Stable Diffusion to generate the portraits, but there has been controvery as several artists claim that their work has been stolen from the algorithm.

Added: Art professors have begun to worry about the prevalence of art-generating algorithms, particularly pertaining to its simplicity of use. Users can create abstract and sophisticated images, which has teachers worried their students won't develop necessary skills for work in art.

Artists have brought a lawsuit against three major AI Art companies, claiming the companies illegally used the artwork and illustrations of artists to train their algorithms.

A Canadian illustrator has discovered that an AI Art algorithm used his art without his consent.

Discussion from Peer

The article is around 1220 words, which definitely hits the word requirement for the draft portion of this assignment. The article is about Artificial intelligence (AI) generated art. The opening paragraph of this article is strong. Comprising of 4 paragraphs, each paragraph is densely filled with information that not only teaches readers about what AI-generated art is but also how it works. For example, the author explains the different methods for how images are produced through an algorithmic process that uses an autoregressive or diffusion prior to text-based prompts. For image prompts, the author explains the process of how the algorithm can use a diffusion decoder. In addition to introducing the topic and explaining how information technology works on the surface level, the author also briefly talks about the different apps on the market that use these technologies and the affordances that go with each application. Overall, the opening paragraph is very strong when it comes to summarizing the issue without going too in-depth about it. The body of the article is also separated into multiple sub-sections that go with each section. For example, the subsections “Copyright Infringement”, “Data Scrapping”, “Harassment and Discrimination”, and the “Kim Jung-gi Case” all go with the section called “Ethical Dilemmas.” The subsections “Previous Court Rulings”, “Ongoing Court Cases regarding Copyright Infringement”, and “Resistance to Data Scrapping” all go with the section called “Responses to Ethical Issues.” Lastly, the author has over 30 references to reliable sources. I think structure-wise, the author did a really good job in staying true to the theme of “ethical issues” surrounding their chosen information technology. They did not ramble or go off track and got straight to the point of the ethical dilemmas and the responses to those ethical dilemmas.

The issue at stake is very clear to me. AI-generated art has been booming in popularity lately and has been very prevalent across many social media platforms. The ethical issues surrounding AI-generated art can be very serious and the article does a good job of bringing the issues to light and highlighting why people should care about it. Copyright infringement is a common issue among human creators, but when AI is involved, it is often difficult to really know when the AI will step into the zones of copyrighted work. As stated in the article, a previous court hearing ruled that an AI-generated piece of artwork “lacked ‘human authorship’, which is needed for protection under copyright.” This issue is especially concerning because we truly never when the AI will accidentally draw from the works of existing art pieces. The issue of automated jobs and how AI-generated art can overtake real artists are also at stake. The article also able to help me understand the issue of privacy surrounding artificial intelligence. AI can sometimes draw personal information online without consent, such as medical and police records. This creates a huge privacy concern for literally anyone who uses the internet.

Lastly, as for objective reporting, I think the author does a good job of not stating any personal opinions and staying true to just facts pulled from reliable sources. The article does mainly talk about the negative implications of AI-generated art. While the issue itself is definitely a serious topic of discussion, I think the author can add a section at the beginning where they talk about how AI-generated art can be good, and then start talking about the negative consequences. I think they can still mainly focus on the negative implications, but I think briefly mentioning the positives won’t hurt. Also, they can mention the recent popularity boom and how TikTok even has built-in filters for people to do self-portraits. But going back to objective reporting, I believe that the author does have a neutral point of view all throughout the article and does not state any personal opinions. All of their arguments are based on facts pulled from a variety of sources. When talking about lawsuits, the author mentions both sides of the argument.