Will A.I. make creative professionals obsolete?
Artificially modeling human creativity is incredibly hard
Until recently creative fields were considered to be immune to any form of automation or artificial intervention. Unlike other conventional fields where the processes can more or less be standardized, the act of creation offers a complete antithesis. It is iterative, organic, and in most cases downright messy.
Creating something that stirs humans at a visceral level requires cultural context, intuition, and ingenuity. There is no standard formula. No recipe. Artists/designers often use the phrase “trust the process” to bring some semblance of certainty into a process that’s largely vague. It almost feels like being on the edge of chaos. Introducing an element of certainty inadvertently ends up removing the element of creativity. You always start from a blank canvas and get thrown into the deep end.
Moreover, the success of artistic output is non-binary. It depends to a large degree on the emotions it evokes in people, and how it makes them feel. Not only is this tricky to quantify, but even the emotional responses aren’t consistent.
OpenAI took up this challenge with their ground-breaking text-to-image research project DALL·E in early 2021. It stemmed from a simple idea — instead of machine learning algorithms automatically captioning images (a technology that’s been around since 2015), what if we flipped the logic to generate images from natural language descriptions?
This idea inspired breakthrough AI research, and the systems that followed DALL·E are nothing short of a glimpse into the future.
DALL·E 2, Midjourney, Stable Diffusion, etc. allow users to input a text prompt and generate photorealistic images in mere seconds. Most of these systems are still in beta, but the generated images are already at a level of fidelity that makes them indistinguishable from the works of professional artists.
AI systems available today can combine concepts, attributes and styles. They can make realistic edits, generate variations of an image, add or remove elements all while taking shadows, reflections, and textures into account!
This really puts things into perspective. For artists/designers, it is both empowering and scary. It shakes our sense of security and makes us wonder…
Do humans overestimate their creative ability?
Creativity has long been romanticized as this divine gift endowed upon a select few. In moments of creative brilliance artists, poets, musicians, scientists, and the likes are known to conjure innovative ideas out of thin air.
In essence, however, creativity is not as fanciful. In my opinion, it is a function of a person’s knowledge and experience. Creativity manifests itself in a person’s ability to synthesize known information in novel ways.
From this viewpoint, creativity isn’t necessarily a scarce resource. On the contrary, it is quite abundant. Everyone can be and is creative. It is like a muscle that can be trained. The training involves building a large mental library of concepts and being open to diverse experiences.
Modeling Creativity
Once we demystify creativity and break it down into its constituents, it starts becoming clearer how AI can be modeled to facilitate humans in their creative pursuits.
To begin with, text-to-image models are trained on hundreds of millions of images along with their descriptions (alt text) scraped from the internet. Through deep learning, the system is able to learn the relationship between objects and their natural language terms.
For an AI system, every image is an arrangement of pixels. It perceives images very differently than humans. Every time the model encounters an image with let’s say a “shoe” descriptor, it validates its existing model, checks for new attributes, and improves its overall understanding of what metrics constitute a “shoe” by plotting the object’s location in a mathematical space.
As this process is repeated for a diverse set of images, the AI model becomes increasingly complex and grows into a “multi-dimensional latent space”. To put things into perspective, this space is a classification of objects on criteria so extensive that humans don’t even perceive these differences through their senses.
The image generation process starts with a text prompt that directs the system to a specific location in its latent space. Each point in the latent space can have up to 500 coordinates and is a recipe for a potential image. Through an iterative process called Diffusion, the AI system approximates a new image based on the text prompt that makes sense to humans.
Because the model has been exposed to such a huge repository of data, it has started showing signs of contextual understanding. It understands that “shoes” are worn on the “feet”; that they are used for activities like “running”; If the prompt says “A person tying their shoes”, the model understands that the user is referring to tying “shoelaces” even when that specific term is missing from the prompt.
The fascinating part is that the generated images aren’t simply being recycled from the training data set. The program is able to create completely new image compositions from the text prompt!
“AI models have the capacity to see the totality of the entire recorded human experience. They can emulate every famous designer, every art style and every scene in all of recorded history. It has a larger catalogue of all cultural, historical and artistic movements than any human will ever possess in their mind.” - John Mauriello
Democratizing art
The distinct advantage of text-to-image AI systems stems from their ability to supplement human craftsmanship. Never in the past have non-artists been able to express themselves creatively. Their imagination was always sort of trapped in their minds. In this aspect, prompting is a game-changer. It acts as a huge enabler for people who haven’t spent years honing a craft or learning a technical skill. Once this technology goes mainstream, it might lead to an explosion in high-fidelity creative output, and that’s amazing!
But will AI systems eventually become advanced enough to completely remove the human element in artistic endeavors and make craftsmanship redundant? I don’t think so. Going by historical trends, every time a technological advancement has threatened to kill craftsmanship, it ends up making hand-crafted products more desirable. And this time it would be no different. Craftsmanship is anti-fragile.
Concept artists, illustrators, and graphic designers who have worked in the industry for years, and understand the nuances of their craft already see this as “one of those” events. Once synthetic images have saturated the market and the hype has died down, AI systems would default back to being just another set of tools in the artists' toolbox, that enable them to do what they do best, more effectively.
Staying relevant
In the long term, staying relevant is less about taking specific steps and more about building a mental framework that helps us stay curious and embrace change.
If we consider graphic design as an umbrella term for various kinds of digital art, most of it has more to do with applied logic and visual problem solving rather than mastering a tool and creating standalone pretty images.
AI systems are here to stay, and hungry to disrupt. They would trigger both positive and negative long-range consequences that no person would be able to accurately predict.
As creatives, if we try moving up the value chain, and build a distinguished preference for our unique points of view, we can potentially position ourselves to take advantage of many such technological advancements, and stand a chance to ride the next big wave of creativity, without being left behind.
