Beyond Brushstrokes: Exploring the Pros and Cons of AI Art

Note: This article was originally published in The Peak

On June 21, Marvel Studios’ released their new show “Secret Invasion.” The title sequence of “Secret Invasion” was created using artificial intelligence. This comes after they employed incredibly talented artists for the creation of the hit movie “Spider-Man: Across the Spider-Verse.” 

Artists and visual effects creators have called for unionizing and regulations to protect their jobs in the wake of AI being used on as big of a project as a Marvel television show.

Since the twentieth century, when artists began employing AI to make art, the usage of AI-generated art has prompted several debates. Some of these claims have questioned whether AI art can be considered as art and what impact it will have on artists.

What is AI Art?

Art generated by artificial intelligence involves applications that invite users to submit a prompt of anything they can imagine. The artificial intelligent application will create an image based on that prompt. 

Artwork created by artificial intelligence isn’t new, since it emerged in the 70’s. What’s new is that everybody is talking about AI as it relates to all aspects of life. Modern AI has some excellent applications; like transcribing audio and isolating the voice of a Beatle on a lost track. 

In recent years, AI-generated artwork has exploded in popularity. Image and art generator apps like Midjourney and DALL-E have been under scrutiny for claims that they unethically steal human artists’ work and compromise creativity.

Criticisms of AI Art

While advancements in artificial intelligence have undeniably revolutionized various industries, a growing number of critics argue against the use of AI-generated artwork, citing concerns about artistic authenticity and the preservation of human creativity.

From graphic designers to digital artists to tattoo artists, Instagram is a vital way for creators to showcase their work for the public and potential employers. Popular AI art pages on Instagram often have more followers than human art accounts. 

While a person may be behind the scenes writing the prompts for an AI art account, they aren’t putting in hours of work to complete a single piece of artwork for one Instagram post. Human artists are using hashtags such as #artbyhumans, #noAI and #notoAIgeneratedimages to show their opposition. 

There is an argument for AI art that people can’t afford to pay to commission human artists. Tools like Stable Diffusion and DALL-E are free to use. A more advanced AI art generator, Midjourney, is $96 annually.

There are also inexpensive options for commissioning/using human artists’ skills. The widely used application for creating digital art, Procreate, is a one-time purchase of $13. Picrew is a free online tool used for creating images and avatars that only uses images with permission from the copyright holders.

Critics of AI-generated art say that it shouldn’t be called AI art at all, that it is just AI-generated images based on data from billions of human-made artworks.

How do AI image generator apps get their data?

Midjourney (along with Stability AI and DeviantArt) have been less than forthcoming in answering that question. So much so that a group of artists sued them for their use of Stable Diffusion, which is, according to plaintiff Matthew Butterick, a “21st-cen­tury col­lage tool that remixes the copy­righted works of mil­lions of artists whose work was used as train­ing data.” 

Stable Diffusion is informed by LAION, a German organization that scrapes the internet and uses that data to create large datasets for AI companies; all without consent, credit, or compensation to the original artists whose work is being used and profited from.

In another lawsuit, Getty Images sued Stability AI (The developer of Stable Diffusion) for infringing on its intellectual property rights. In a press release, Getty Images said “Stability AI unlawfully copied and processed millions of images protected by copyright” and did so without licensing. Getty points out that they are in support of innovative AI endeavors; in fact, have provided licenses to tech companies for the purpose of AI training. 

Artists aren’t for the abolition of AI-generated art. They state their terms clearly. In the art world, it is known colloquially as the three C’s; consent, credit, and compensation. 

Midjourney-generated image

Positives of AI Art

Art generated by AI is almost limitless. Any prompt you can think of will have a result. Through a process known as “inference tuning”, an AI-art generator will scrape its dataset for any specific prompt that you provide. The quality of results vary, of course. People who have rarely seen themselves represented in art can generate that representation. For these photos, I wrote prompts that included fat women walking the runway in a fashion show and an interabled couple getting engaged. You could still commission an artist to do this work, but it would be a lot more expensive than the $10 per month for unlimited images in Midjourney. And something like a photographic representation in the photo of the fat woman is almost unheard of in society.

AI art doesn’t have to exist in contention with human artists. Comic book artist Dash Shaw said, “Sometimes I think everything I draw is just a combination of all of the millions and millions of drawings I’ve seen.” Austin Kleon wrote a whole book called “Steal Like An Artist.” AI-generated art could be considered a combination of all art that exists on the internet. Regardless, artists whose work is being used as a learning model should be aware of their work being used and fairly credited and compensated. Intellectual property lawyer Kate Downing commented on the Stable Diffusion Lawsuit, saying that companies using the LAION datasets “produce results with markedly different qualities.” In other words, typing the same prompt into an AI art generator twice could yield two completely different images. She says any claims that are made that AI art copies human art directly is deeply misleading.

Midjourney-generated image


Copyright Law/Fair Use

What is Fair Use?

Certain copyrighted content can be used without the permission of the copyright owner if it meets the fair use criteria.

The US Copyright Office defines fair use and identifies the criteria:

In determining whether the use made of a work in any particular case is a fair use the factors to be considered shall include—

(1) the purpose and character of the use, including whether such use is of a commercial nature or is for nonprofit educational purposes;
(2) the nature of the copyrighted work;
(3) the amount and substantiality of the portion used in relation to the copyrighted work as a whole; and
(4) the effect of the use upon the potential market for or value of the copyrighted work.

Fair Use is the use of copyrighted content in a transformative way. Examples of this include using a soundbite or lyrics of a song for commentary and/or criticism in a music review or parodying songs like Another One Rides the Bus by Weird Al Yankovic. Other examples of fair use is content used in news reporting, teaching, or research.

How does this apply to AI-generated art?

Courts have consistently ruled that AI-generated artworks cannot be copyrighted. The training that goes into creating the massive datasets that applications like Stable Diffusion and Midjourney use to operate their platforms is informed by LAION, specifically LAION-5B. To put it simply, the LAION-5B data is not actually images, but rather image-text pairs. The program adds “noise” to the images, corrupts them, then trains the AI to eliminate the corruption. After that, they train the AI to compare/contrast the images and associated text. Then the AI clusters together similar data.

This process is transformative and falls under Fair Use. The images in the dataset are used as just that; data and not for their artistic value.

What is art? Is it the human component? Or is it the conversation, the interpretation around the artwork? Is there something intrinsic to humanity that cannot be recreated, even with billions of data?