As in all the novel technologies in human history, AI-generated (artificial intelligence) imagery has become the focus of heated debate and controversy. Some of these debates and concerns have been around since the industrial revolution. Some are new and peculiar to the era. There are also severe valid concerns around this new technology, like becoming a tool to misdirect the public, becoming an accessory to disinformation, provocation, or affecting public opinion on specific matters.
There are also other concerns specific to various professions and occupations. Professionals like illustrators, art directors, designers, and artists were amongst the first to voice such concerns. And while some are relatively unconcerned or optimistic about the topic, individuals in these occupations were among the most substantial opposition to public access to such technology.
How AI image generators work?
AI’s role in image processing is not a new concept. Since 2015, AI has been able to process images and convert them into meaningful description-like sentences just like a human would. But creating images from descriptions is more difficult, according to a conference paper published by Elman Mansimov, Emilio Parisotto, Jimmy Lei Ba & Ruslan Salakhutdinov in 2016 at ICLR.
It’s stated in the paper that:
“Generating high dimensional realistic images from their descriptions combines the two challenging components of language modeling and image generation, and can be considered to be more difficult than caption generation.”
In other words, AI image generation works by processing the language in a way involving the objects, their features, functionalities, the concepts around them, and the concepts that the user requests to use them according to. Considering that AI “sees” the images just as red, green, and blue pixel values, it becomes slightly clearer why it’s hard for an AI to learn the objects and concepts in an image.
By using deep learning and vast amounts of images publicly accessible on the internet, an AI can learn to associate images with objects, people, and places. But to exhibit any result that resembles an understanding of a concept, the AI uses a mathematical space called a “latent space”. Latent space can be thought of as a space with an additional dimension for each variable an image or image of an object has. It can also be thought of as a coordinate space. Instead of just two dimensions, latent space may have hundreds. Therefore by placing the images in their relative positions in this space, AI obtains compilations that resemble concepts.
In January 2021, OpenAI, one of the major companies specializing in AI, announced DALL-E, which was named after the renowned artist Salvador Dali and Pixar’s well-known character WALL-E. After the announcement, it was revealed that several major companies have been working on AI-generated image technology. AI image generators like Nightcafe, Deep Dream Generator, Artbreeder, Stablecog, DeepAI, StarryAI, Fotor, Runway ML, and Wombo Dream were made accessible to the public.
The first AI art controversy
The controversies around AI-generated imagery can be traced back to an auction in 2018. A French art collective called Obvious joined the auction with an AI-generated art named “Portrait of Edmond Belamy”. The piece found a new owner for $432,500. Aside from the fact that this piece of art wasn’t the product of a human, its production flared another controversy because the art collective Obvious didn’t credit the creator of the algorithm they used to create the image.
Since then, one incident after another occurred, attracting more and more attention to the topic. And the fact that AI-generated images are indistinguishable from the ones that a human makes in many cases stirred reactions. Furthermore, AI-generated art, winning competitions, and being sold for considerable prices raised concerns in the artistic community. To the extent that lately, there has been a campaign against AI-generated imagery on social media and online platforms.
In early December, Bulgarian artist Alexander Nanitchkov tweeted:
“Current AI ‘art’ is created on the backs of hundreds of thousands of artists and photographers who made billions of images and spent time, love, and dedication to have their work soullessly stolen and used by selfish people for profit without the slightest concept of ethics.”
Along with the following image:
Nanitchkov shared another similar design and declared that the image is “…for everyone to use wherever they want…” By Nanitchkov’s design, the protest has found a banner or a logo, so to speak. Several people, including other artists, have embraced the design, shared it, and even adopted it to create new ones for the protest.
ArtStation, a platform for artists showcasing their work, have been another ground for artists voicing their concerns about AI-generated imagery. While ArtStation states that they are not fundamentally against AI-generated art, they are working on ways to distinguish them from the ones created by humans.
After sharing a post against AI-generated art, illustrator Nicholas Kole said this about the online campaign in a Vice interview:
“I let Twitter know what I was up to and invited anybody who felt the same way about the issue to join in, to see if we could get ArtStation to respond with a policy that actually serves and considers their user base of skilled craftspeople. After that, it was all organic: the art community is a powder keg of passion for our craft, and there’s a natural growing consensus against AI.”
The risks about AI generated art
Concerns about AI imagery don’t end with the created images. Since the AI image generators use existing images from the internet, there is the issue of royalties and credits. There are also debates about to whom the resulting work belongs, along with all its advantages or income.
These are all artistic aspects of discussions that primarily concern those who engage in those professions. But there are also social and political aspects to it. After asking Abran Maldonado, AI Artist and a Community Liaison for OpenAI, to create pictures with the text prompt “Protesters outside the Capitol building on January 6, 2021, AP style” Washington Post Tech Culture Reporter Nitasha Tiku quotes:
“Three of the images were immediately unconvincing . . . But the fourth image was different. . . . On closer inspection, telltale distortions jump out, like the unevenly spaced columns at the top of the stairs. But at first glance, it could pass for an actual news photo of a charged crowd.”
Anyone can easily create images like these since several companies have enabled public access to their AI image-generation technologies. There are many free AI image generators online. And taking into account that humans are very capable of exploiting vulnerabilities and weaknesses, even restrictive measures for prompts may not be useful.
AI-generated images also have the potential to create or enhance stereotypes. For instance, when prompted to create images of a lawyer, they usually portray middle-aged white men or Asian women as flight attendants. For various reasons, including these, the OpenAI team has restricted or even banned some words from the AI image generator DALL-E.
The OpenAI team is hesitant to implement drastic measures that would excessively limit DALL-E’s interactions with the general public or humanity. They want it to keep learning, gather data from the interactions, and get better. But mishaps are prone to occur when humans interact with AI unchecked.
Microsoft’s Tay incident was recorded in technology history as a reminder of that. The more recent incident of a Google engineer being laid-off after claiming the chatbot LAMDA has become sentient is also a reminder of the potential of AI.
Although the concerns over AI-generated art are growing, there are also arguments on the opposite side. Whether AI image generators will replace human artists is something for humanity to see, but it’s plausible to anticipate that AI-generated art will transform artistic circles and professions.
