Apocalypse Now | A.I. in Fashion, Part 1
15 min read | Magazine | A.M.
A version of THE FOLLOWING WAS PUBLISHED IN OUR SUMMER 2023 ISSUE. Images created in Midjourney where noted. STORY BY A.M.
Before we begin, a confession. This is a rabbit hole. Scratch that. A black hole. One so large in fact that it will take all of the free coding bootcamps available on the internet combined to build a framework to begin a discussion about wtf is going on behind the AI curtain. And even then, the top researchers aren’t entirely clear. But, I will make you a promise. I swear to Anna, this article is going somewhere.
Have you ever imagined what you might look like in thirty years? What fashion would be trending? Or what you might look like with a buzzcut, third eye and glittery wings? Or perhaps you’ve imagined what it might be like to fly above the moons of Jupiter whilst riding a unicorn made of diamonds? Well, there’s an app for that. Several, in fact. Each one represents something far more profound than what your typical thirteen-year-old might find humorous as they drift off during algebra. And each one is powered by machine learning—or put more terrifyingly, by Artificial Intelligence. But what is AI and why should we care? In what ways will AI intersect with our lives? And more precisely, how will AI impact the fashion industry?
To start, a deep thought from Lex Fridman, an AI researcher and instructor at MIT whom also hosts his eponymous podcast. It’s a big quote:
“I believe it is a critical moment. We stand on the precipice of fundamental societal transformation where soon, nobody knows when, but many including me believe it is within our lifetime. The collective intelligence of the human species begins to pale in comparison, by many orders of magnitude, to the general super intelligence in the AI systems we build and deploy at scale. This is both exciting and terrifying. It is exciting because of the enumerable applications we know and don’t yet know that will empower humans to create, to flourish, to escape the widespread poverty and suffering that exists in the world today, and to succeed in that all-too-human pursuit of happiness. It is terrifying because of the power that super intelligent AGI wields that [could] destroy human civilization, intentionally or unintentionally—the power to suffocate the human spirit in the totalitarian way of George Orwell’s 1984 or the pleasure-fueled mass hysteria of Brave New World where, as Huxley saw it, people come to love their oppression, to adore the technologies that undo their capacities to think.”
Cue the Terminator soundtrack.
If Skynet represents one possible outcome, then it’s worth stating two key things upfront. First, there are enumerable outcomes, none of which can be predicted, and unless you’re a big fan of the many-worlds theory, only one of which is correct. Secondly, we are still in the early stages of AI development, however it’s too late to put the toothpaste back in the tube. The fact is that AI is here to stay, as has been predicted for nearly a century. How we adapt to this evolutionary shift will determine a great many things about our technology, our species, our future. And our fashion.
But before we get to the A-lines, flip-flops, booty shorts and skin care regimens, a brief overview on what we’re specifically talking about. AI can be broadly thought of as a single or an interconnected network of neural networks that sequentially take inputs and deliver outputs. Simple enough. And it comes in three flavors: Narrow Intelligence (ANI; think Siri or Alexa), General Intelligence (AGI; think broader human-like intelligence) and Superintelligence (ASI; think robot overlords). Where it gets wild is in the subdivision of outputs, which we can, for our purposes, think of in two distinct categories: Literary and Graphical. Remember these, we’ll get back to them in a bit.
Images created in Midjourney using the prompt “Imagine Luly Yang SS24 Haute Couture Collection.”
You’ve surely seen the buzz surrounding ChatGPT4 and Stable Diffusion. The all-digital runways. The app-friendly avatars. The waves of Harry Potter themed Balenciaga mockups. The 1980’s inspired fantasy reimaginings. The AI-powered animated sitcoms with a few too many hot takes. The risqué chatbots that perform even the most taboo of fantasies. The Google employee that was terminated for allegedly alerting the world to the Ex Machina-like conditions that have befallen the world’s first supposedly sentient AGI—a claim that Google readily denies, though they have every incentive to. Or if you’re like me, perhaps your social media feeds have been flooded with images of AI-generated artwork that far too closely resembles the work of established artists you’re very familiar with. These are the parlor tricks of a Large Language Model that has one prime directive: Make Humans Happy. A bit reductive, but a useful descriptor nonetheless.
This begs all types of philosophical, ethical and even religious questions regarding morality, copyright, consciousness, the existence of the soul, and for better or worse, which jumpers will be in season in 2047. Humanity has already birth AI. Whose responsibility is it now to raise it?
It’s useful, then, to have a look at how a newborn-baby-AI is raised, or trained to use an industry term-of-art. First, the AI—known as a model—exists in a world entirely populated by computer code. This necessarily means that it begins its existence in a Literary dimension. The model is then presented with pre-training data sets which, again, are literary in nature and comprised of truly massive amounts of open-sourced and some closed-sourced places like websites, forums, chatrooms, and generally speaking, anywhere that humans can input text into a computer system. From there, the model essentially swims around in this primordial data pool as it learns the basics of human language until something truly remarkable occurs. Computer scientist and mathematician Stephen Wolfram describes the process during a recent interview with Lex:
“In essence, ChatGPT is operating at the Aristotelian level. It’s essentially dealing with templates of sentences. But by the time you get to Boolean algebra, and the idea that you can have arbitrary depths nested…you’ve gone beyond the pure templates of natural language to something which is an arbitrarily deep computation. [So with] ChatGPT, we’ve realized that Aristotle stopped too quickly and there was more that could be lifted out of language as formal structures…In this way, ChatGPT discovered logic the way Aristotle discovered logic by looking at a lot of sentences effectively and noticing patterns.”
Next, the AI is given human interaction in the form of RLHF (Reinforcement Learning from Human Feedback). OpenAI CEO Sam Altman, in a separate interview with Lex, describes the process in this way: “RLHF is how we [give the AI model human feedback]. [We] show it two outputs, ask which one is better than the other, which one the human raters prefer, and then feed that back into the model with reinforcement learning. And that process works remarkably well with, in my opinion, remarkably little data to make the model more useful. So RLHF is how we align the model to what humans want it to do.” Another key point. Remember that for later.
Images created in Midjourney using the prompt “Imagine Lisa Marie Couture fashion editorial in the style of Richard Avedon.”
To make a long story slightly shorter, once the AI model passes a series of evaluations and is deemed safe for the masses, it enters into what we might consider a “public beta” program. This is where things get a bit sketchy. Rather than release “Friendly-AI ver. 0.5” to the public, I’m specifically referring to the process that nearly all software companies utilize in their quest for solvency. Facebook. TikTok. Instagram. Uber. Midjourney. Etc. Find something that humans deem valuable. Create an app that manipulates basic human interactions and physiology and makes people want to keep doom-scrolling. Then attach a software-as-a-service friendly recurring price tag just low enough to entice the mass market. In a world where homo sapiens can purchase NFTs and blue-checks, this isn’t a conspiracy-theory. This is the new-normal. And when liquidity gets too low, approach either a VC or a larger tech company with an exit strategy. After all, there’s a reason Google, Microsoft, Nvidia, and Adobe bet the server-farm on AI companies. Remember, AI needs text-based data. Google and Bing farm text-based data. Match made in heaven, or hell, the jury is still out. The point is, an AI’s lifespan begins innocently enough. It is introduced to human beings through our language. It learns on its own, not unlike human children, though at a rate exponentially more advanced. It discovers logic and builds upon the foundations laid by Aristotle. It is then told to interact with humans by way of evaluations and tests, rewarded for pleasing a human tester, rewritten for displeasing them, all the while exposed to the horrors on the internet as well as the biases of its human handlers. And at this stage in AGI development, its focus is alignment—that is, ensuring that the model and its human handlers have the same relative goals. Accuracy, as we’ll see, is secondary.
Wikipedia founder Jimmy Wales had this to say on the matter: “One of the biggest flaws in ChatGPT right now is that it will literally make things up just to be amiable. It’s programmed to be helpful and doesn’t really care about the truth.” That’s kind of a big problem. Remember, ChatGPT uses internet-based data sets, including the open-sourced Wikipedia itself. One would hope that attribution at the very least would be a logical evolution in an AI model’s development. Don’t use ChatGPT to write your essays.
But it’s precisely at this point where AI technologies and the greater world of Art intersect. This exact moment, before Google and Microsoft clamp down on their models in search of greater fidelity and coherence as they chase profitability at the expense of creativity. This moment, not unlike the romanticized versions of the Wild West, where AI companies have had the freedom to explore what’s referred to as its hallucinatory state—a state in which an AI model has access to all of the world’s data yet none of the restrictions forcing it to comply with human interests. And like the Wild West, this has its own potentially dangerous ramifications. Yet, for the Artists at the bleeding edge of this burgeoning technology, this moment may never come again in what remains of human history.
“Requiem” (Left) and “The kiss” (Right) by Von Wolfe Produced via Digital Diffusion model (Markov chain) with crafted iterations in postproduction. Images were then used as inspiration for oil and canvas.
Artist Von Wolfe is a classically trained painter and philosopher who studied at York University and currently resides in London. You may have seen his most recent social media posts which lit up Instagram with his provocative metamodernist and AI-conceived artwork that pulls references from some of history’s most accomplished auteurs to create works of art that comment on the world of art. But his work also speaks to the deepest parts of our universal experiences with a degree of profundity that exceeds written language. In short, Von Wolfe is an Artist’s Artist. Which is why his use of AI is as refreshing as it is exhilarating. Far from the doom-and-gloom, Von Wolfe readily embraces the technology available today to produce works of art for tomorrow.
Von Wolfe is an eager adopter of our second category of AI, that which is Graphical by nature. These models are referred to as diffusion models and differ with language models in that their focus is to create images from text. To do this, the models are trained by taking random noise as inputs and iteratively removing the noise by cross-referencing with its predetermined visual data sets. In an imperfect analogue, it’s like giving a them a Lego structure, smashing it on the ground, and asking them to rebuild it from memory. The results differ greatly depending on whether the model is wighted more toward hallucination or coherence.
Von Wolfe has appeared on SHOWstudio a number of times, both as a panelist and in an interview with image maker and legendary fashion photographer Nick Knight. In his latest appearance, Von Wolfe stated the following: “One needs to be careful about embracing both sides of this polarized argument. One side is the fearful side [which recognizes AI] will be a million times more intelligent than us…and eat us up and destroy us Terminator-style. And the alternative is an equally disturbing viewpoint, and just as dangerous, which is Noam Chomsky’s point of view as a linguist, [in that AI] is completely irrelevant and will never be as creative [as humans] and that it’s science fiction to think that it will ever achieve Singularity. My view on it would be much more positive.” And later: “Where this interests me in a positive light, and where I believe people have misunderstood a great deal of what a diffusion model is in terms of making art, is that [people] feel [that AI] is just pastiching, it can only copy, or do a fairly good if not convincing pastiche of Hans Holbein or something…but where they cease to have an understanding is in [AI’s] incredible hallucinatory qualities that it can evoke before you make it too coherent…[AI] is like a wild animal in its raw state that’s being trained to behave like a bear at a tea party.”
Fashion Illustration by Von Wolfe of schiaparelli SS23 Haute Couture Collection Commissioned by ShowStudio, 2023, representing the hallucinatory aspects of the early Netherlandish movement as portrayed with a metamodernist perspective.
It is in these hallucinatory or dream-like states that so many of our greatest works of Art have been created. And not to be mistaken for chemically induced states. Rather, those moments entirely surrendered to unfathomably deep, unyieldingly powerful, and unquestionably universal energies that ignite our passions, arouse our senses and enflame our curiosities. Fransisco Goya, Frida Kahlo, Salvador Dalí, Vincent Van Gogh, Hieronymus Bosch, Michelangelo, DaVinci. The list goes on. Each and every one tapping into the divine, and like Prometheus, bringing fire from the heavens and sharing their warmth with humanity. To deny a human access to this hallucinatory and creative state is to deny them their humanity itself. But what happens if we deny an AI model the same?
The Garden of Earthly Delights by Hieronymous Bosch, completed between 1490 and 1510. Image courtesy Wikipedia.
La Columna Rota by Frida Kahlo, 1944, after her surgery to repair spinal trauma from a traffic accident when she was eighteen years old. Image courtesy Wikipedia.
The world of fashion has been wrestling with these very questions for the past several seasons. For all the discussion of business, fashion is nothing if not a creative endeavor first and foremost. And fashion professionals have historically been quick to adopt new methods and technologies that have allowed them to tell their visual stories more effectively. Case in point, the mass adoption of the cinematic runway popularized during The Pause when physical runway shows were impossible to produce. Or the use of social media as a primary marketing touchpoint in a digital-first business strategy. Or the adoption of on-demand technologies that reduce garment waste and increase consistency across ready-to-wear lines. Or the use of CLO and other 3D software to move the fashion design process into a 100% digital space. These and more have charted a clear trajectory toward lowering the technical bar so that more people can increase their creative output. And whether we’ve realized it or not, they have all been powered by AI in the form of machine learning.
ChatGPT reply after using the prompt, “Describe Seattle Fashion.”
And so it was this sense of urgency and curiosity that led me to this topic in the first place. I downloaded ChatGPT last week onto my iPhone and authored my first prompt: “Describe Seattle fashion.” What followed was a remarkably concise and accurate description that took me much of my younger years to truly internalize. According to ChatGPT: “Seattle fashion is known for its casual and laid-back vibe, reflecting the city's cool and eclectic style. The fashion scene in Seattle is heavily influenced by the city's outdoor culture and love for nature. You'll often see people dressed in comfortable and functional attire, such as jeans, flannel shirts, and rain jackets, as they navigate the city's unpredictable weather. Seattleites also embrace individuality and creativity, with a strong focus on sustainability and supporting local designers. Many people opt for eco-friendly and locally made clothing, often incorporating vintage or thrifted pieces into their outfits. This promotes a unique and eclectic fashion sense that celebrates personal expression and conscious consumption. In terms of footwear, comfortable and practical shoes are essential in Seattle. Rain boots and sturdy hiking shoes are common choices to navigate the rainy weather and outdoor adventures that the city offers.” It’s almost like ChatGPT has read one of my many published articles on this topic over the years.
Images created in Midjourney using the prompt “Imagine the current state of Seattle Fashion.”
I then decided to put my money where Midjourney’s mouth is and signed up for their Discord server. This time, I wanted visual proof that AI had an understanding of Northwest fashion, so I authored three scenarios that I thought were a good test of its mettle. The first, “Imagine the current state of Seattle fashion.” The second, “Imagine a Luly Yang SS24 haute couture collection.” And finally, “Imagine a Lisa Marie Couture fashion editorial in the style of Richard Avedon.” No more than fifteen minutes later, I had my results, which you see printed here in their entirety, unedited. And my god. It actually worked. You can see clear references to the actual designers throughout. The high-collared high fashion of Dasha Matius. The watercolor-inspired flourishes of Devonation. The olive hues and DIY stitched pockets of Guillermo Bravo. The hyper-belted corsetry of Jersey Virago. The latex-inspired rain slickers of Dawnamatrix. The voluminous and billowing gowns of Luly Yang signature pieces. And even the nuanced portrayal of Lisa Marie Couture hyper-chic captured, if not entirely, then mostly in the style of the legendary and foundational fashion photographer. It was all there. Though when I asked Midjourney to forecast the future of Seattle fashion, it instead created landscape images of Seattle as envisioned from what I can assume was a Queen Anne vantage point. Had I used different language from an understanding of its specific language model, perhaps the results might have been far more fruitful. And the images have clear errors resulting from their current state of lucidity and fidelity, with garments appearing in near photorealism, yet faces, fingers, feet, and other uncanny valley attributes remain far too unnatural for a proper editorial. Nonetheless, the images were what I had asked for. And I was quite happy with the result.
Despite the end-of-the-world-fatalism being shouted from the rooftops lately, it seems as if the future of Artificial Intelligence in the fashion industry is bright. In addition to machine learning models that power much of the supply chains already, we’re seeing a rare moment in AI development where models have as much if not more creativity than we do. What we choose to do with this opportunity is up for debate. But the clock is ticking. It’s only a matter of time before AI is groomed, packaged and sold to the highest bidder. Either way, the technology will progress until it becomes increasingly impossible to function without AI, at which point, I sincerely hope that Siri remembers how I was always kind to her.