Three women, draped in billowing silk frocks of white and red, stand before a pale, luminous halo of light revealing some intangible city or civilization in the distance. They are surrounded by dark walls and corridors flecked with gold, inside what might be ruins or an old castle. The firelight, from no certain light source, bestows a kind of warmth, perhaps from scattered torches or a hearth within.
From afar, the painting appears to be of digital medium, a combination of oil or acrylic, maybe. But on closer inspection, none of the lines nor the composition makes sense. There are so many chaotic inconsistencies and subjects that are simply inexplicable. Wrinkled folds of clothing seem to melt into the walls. One can hardly tell what the cacophony of ‘brush strokes’ are beyond the foreground– are they supposed to represent the masses? A kingdom? Treasure? Why are there random fragments protruding from the walls? Are the people tucked aside guards? Attendants? Are they sitting or kneeling?
Two years ago, entrant Jason M. Allen submitted an AI generated art piece titled, Théâtre d’Opéra Spatial, for the Colorado State Fair’s annual art competition— and took home first prize in the digital artist category. He created it using Midjourney, an artificial intelligence program that generates artworks from written prompts. The controversial win incited some heavy backlash, especially from art communities who believed the accolade was not reflective of the artist and the art itself. Mr. Allen had only admitted to using AI after winning the competition, to which he was subsequently disqualified. But his story spurred a contested discussion as to what exactly ‘qualifies’ as art.
Generative artificial intelligence refers to algorithms, such as ChatGPT and Dream AI, that can be used to create new content including audio, code, images, text, simulations, and videos. It is unique from general AI assistants like Siri and Alexa, both of which are founded on AI technology, in that developers can train artificial intelligence through models that can “learn” from data patterns without human direction. This method is called machine learning, a type of AI in itself.
For generative AI, two critical advancements have immensely augmented its mainstream value: transformers and breakthrough language models. Transformers are a type of machine learning that enables researchers to train larger models without having to label all of the data in advance, allowing new models to be trained on billions of pages of text, resulting in more complex and in-depth answers to inputted human prompts. Such human prompts fed to AI must then be categorized and converted into numerical data to be quantified. Additionally, transformers unlocked a new kind of notion: attention. Models were now able to track connections between words across pages, chapters within books, rather than merely individual sentences, as if the AI itself mimicked human memory. Rapid advancements in large language models (LLMs, or models possessing billions or even trillions of parameters) have founded the new era in which AI. can produce engaging text, photorealistic images, graphics, video, etc., and serves as a basis for tools to automatically create images, text, or generated video captions from written or spoken text. Specifically for text-to-image generative AIs, they can generate via scanning existing physical artworks across millions of data points to produce digital that match the user’s text descriptions, a technique sometimes referred to as “scraping.”
The story that is being sold to us is similar all across the board— art is being democratized; pulled from ‘entitled’ artists and converted into a mechanized tool that will make everyone an ‘artist’. Midjourney, for example, claims to be “expanding the imaginative powers of the human species,” while StabilityAI say they are “building the foundation to activate humanity’s potential.” Generative AI is described as a new medium of art that allows humans to forgo the effort of making their own art and learning new skills; rather, it is a frontier of convenience that spits out art within seconds of some typing and a push of a button.
To quote artist Molly Crabapple, “I cannot understand why someone would burn a tonne of carbon, just to take away a job people love to do, and give it to a machine.” As of January 2024, OpenAI chief executive Sam Altman finally admitted what researchers have been saying for years— the artificial intelligence industry is hurtling towards an energy crisis. AI servers are productors of electronic waste; they are large consumers of water, rely on critical minerals and rare elements that are harvested on unethical and unsustainable labor, and utilize massive amounts of electricity, spurring the emission of planet-warming greenhouse gasses. To reiterate, the artworks produced by OpenAI (creator of Dall-E), StabilityAI (creator of Stable Diffusion), and Midjourney– the three biggest image generators— all trained their successful models by scraping millions of peoples’ images from the Internet, entirely without the owners’ knowledge nor permission. The billions-worth generative AIs we see today in mainstream use are trained on the copyrighted works of poorer artists, illustrators, and photographers, and are taken directly from their online portfolios and community sites like DeviantArt. Users can even directly prompt the image generators to create an artwork in the style of another artist simply by typing in their name, indicating these artists’ works were absorbed by the model. Over the last few years, commercial adoption of AI image generators has profited off the stolen works of real illustrators and creators who are losing paid work and control over their own creations. Companies who’ve previously hired artists can simply turn to the cheap, quick, mechanical equivalent, and so long the option for AI art continues to be available, most won’t be able to resist its convenience. Videos generated from AI like Deepfake are being used to falsify information and can be used to demean, intimidate, or harass individuals through videos that are nearly indistinguishable from reality, even creating pornography of unsuspecting victims through their images, many of whom are disproportionately female.
Currently, artists have sought and found methods to protect their art from AI. One such way is utilizing Glaze, a software tool designed by a team at the University of Chicago, which allows artists to invisibly encrypt their digital artworks to prevent AI models from using them. While the image remains unchanged to the human eye, Glaze serves as an active digital watermark that completely inhibits AI’s ability to analyze the work itself, affording creators some peace of mind and genuine protection before legal and commercial AI industries try to adapt AI to suppress the effects of Glaze. Furthermore, while there are unfortunately no laws prohibiting the sale of AI art, the utilization of AI artwork is not currently considered viable for copyright in Canada. The Copyright Act, R.S.C., 1985, c. C-42, (“Act”) does not consider copyright infringement in the context of AI-generated works, and works created using AI may additionally violate copyrights held by others. However, there are still concerns regarding authorship attribution as to who may possess copyright due to the fact there are multiple parties involved in AI-generated work— who owns the rights? The creator of the AI, the user who curated the data, or the investors who financed its development? Can AI-generated work even meet the standards of originality?
We are living in an age of witnessing the degradation of artist culture. Rather than taking joy in the process and time of creating, AI-generated art emphasizes the finished product. Intrinsically, AI generated art is contradictory. It is so instantaneous and effortless, its disregard for intent, labor, and effort, has given birth to stolen, plagiarized works lacking in meaning, substance, and utterly disposable. The very definition of art itself is, “the expression or application of human creative skill and imagination, typically in a visual form such as painting or sculpture, producing works to be appreciated primarily for their beauty or emotional power.” By its nature, art has always been a concept society struggles to firmly define and precisely establish a common constitution of what exactly art encompasses. Artists are compelled to discover, to push boundaries and experiment with the unknown, making art an ever-evolving and expanding field. But AI art is very different. Boundary-pushing has never threatened the livelihoods of other artists before, or threatened to monopolize the means of creation. It may be shocking, uncomfortable, but has never possessed so much destructive power to seek success without any of the work.
Art has always sought to explore humanity; we must not forget to think, learn, and discover for ourselves.
Claire Chu
November 2nd, 2024