Imagine a world, where AI can generate photo-realistic pictures of anything and everything: humans, animals, nature, or any object – even synthesised on-demand with a given description. Not only realistic but abstract and surreal – anything you can and cannot think of, landscapes, creatures and scenes which are beyond your weak hooman imagination. And not only images but also texts, suitable at least for copywriting, cold emails, summaries, customer support bots, but sometimes fiction, poetry and games. And any voice (even yours) to read all these texts– automatically translated into a required language– with the desired accent, mood and tone. And music, setting the right tempo, atmosphere, levels of epicness or tranquillity. Or whole stories, characters, plots, endless and meandering. Imagine a news TV program purely generated by AI – scripts, video and sounds? Arousing, huh?
Now, stop imagining this world because it's already here.
Well, almost, if not – it's just a matter of time. All of the tasks above (and more) could be solved, some better, some worse, some faster, some slower. But soon, many of them will be fathomed and machines will generate everything for us – text, sound, images, providing endless opportunities in simulating the real world in the digital space, making dreams and nightmares of the metaverse come true. Everything will have a question mark if it’s real or not – the internet, the reality, your partner, perhaps, even you. Perhaps, at some point, robots will break through captchas and we will something stronger such as a verification checkmark "written by human", or "proof of humanity" stored somewhere on a blockchain.
You see, whenever AI appears in a story, there’s a high chance it becomes evil, turns against its masters, kills humans, creates a simulation and takes over the world, for a good intention or not. We either see AI as a megalomaniacal entity juxtaposed to us, not megalomaniacal people, or as the complete opposite of that – as a naive creature fooled by us (in which case, however, AI also ends up being evil or at least behaving like a revolting kid lacking communication skills). There is rarely “in between” and there’s rarely an optimistic perspective on Person vs. Machine in pop culture and mass media where both reasonable and mad theories and myths about evil AI has been cultivated for years. For the record, I enjoy many of them.
But we forget that what we call AI now is just a pile of nifty linear algebra, a beautifully engineered mathematical system. You feed it with data, it makes tensor transformations, you get a result you couldn’t get any other way in a reasonable amount of time. Now, AI is an automation machine, the machine that aims to make things simpler and “AI” is merely a buzzword. The common view of what AI is and will be can hardly change unless becomes truly transformative (recommended to read).
Regardless of the current and future states of AI, the brightness of our future relationship with AI depends solely on who owns and how uses it.
AI learns by example, by millions and billions of examples, in simple words, trying to minimise an error between the real and predicted / simulated / generated, trying to explain / represent / resemble the majority of samples. It needs data, a lot of data and whoever owns it, owns the AI.
And I presume I don’t have to tell you who owns all the world’s data. They use it not only to torture us with targeted ads but to train so-called AI which is also involved in the AdTech business. A real case of evil AI you don’t need to be a sci-fi writer to come up with.
But I digress.
The most advanced algorithms we know are worth much less than the data that feeds them, and that is why data is, quite proverbially so, “the new oil”. You can find a paper explaining almost any of the state-of-the-art algorithms in all the details you need to implement it yourself. You can even find the ready-to-use implementation on GitHub or somewhere else in a handy toolbox. It is up to you.
This transparency is good for researchers, practitioners or just tinkerers but it is even more beneficial for big corporations who open source these algorithms. The big corporations are and will be the ones who own the most advanced AI, control and fund the development of it for better or worse. You need around $12 000 000 to train GPT-3, probably the most advanced and famous language model so far.
My belief is the general transparency about AI and algorithms is good and promising because, as I mentioned, any tech-savvy enough person can take an existing implementation, pre-trained models or tools and tinker with them, create new products, experiment, do everything, including making art.
Will Smith: Human beings have dreams. Even dogs have dreams, but not you, you are just a machine. An imitation of life. Can a robot write a symphony? Can a robot turn a… canvas into a beautiful masterpiece?
Robot: Can you? LOL
AI or computer-generated art already established itself as a standalone field and is being sold for millions of dollars. The fact of it may seem incomprehensible, strange, bizarre and devoid of any meaning (not as much as selling an empty space, though), but it's already with us. You can’t deny it. It is with us if not as a piece of art, at least as a part of other systems and programs which use AI and different kinds of math and physics for visual processing and not only – such as 2D and 3D editors, phone cameras, face-swap apps, cinema industry and more and more. It is everywhere, you just don’t always see it.
We can argue if that "computer-made" art can be considered as Art with a capital A, and if we can consider AI an artist. But should we? For myself, I found out that it is much easier to consider it as a tool because it is not a decision-maker, not at any stage of the artistic process – neither a creator with the idea nor a perceiver sharing with the creator their common solitude.
To define a task, create, code and tune the algorithms, evaluate and filter results we still need a human. All those stories generated by GPT-3 are mostly fortunate results selected by a human. All those pictures "painted" by a computer are just results of having good algorithms and the right parameters, the result of human decisions in combination with the right tools. All those strange pictures with cats and dogs, resembling a trypophobe nightmare or an unfortunate effect of psychedelics, are just an accident, the result of the multiplication of several tensors, just a clever algebra created by humans. An accident, because the AI-generated it doesn't understand if a result is a piece of art or not. It doesn't operate in such terms – at the end, it's decided by a human, the human who once saw something enthralling in a latent space of the artificial neural network’s dreams.
We decide on what Art is. We can claim that Art it is only what was made in Renaissance and nothing else, or we can say Art is an artfully organised pile of shit or a banana glued to the wall. And the result of the discussion of whether a machine can turn a canvas into a masterpiece or write a symphony is always up to us. It can and it can do it better than most of us. The question is only if we can accept it.
Take this, for example:
The enchanting view of a library, a bit weird and uncanny but fascinating and impressive in its details. There’s a weird symmetry feeling though the picture isn’t symmetrical. The perspective always goes somewhere far in the centre creating a sense of endless meditation you can lose yourself in. And If you look through all the images I linked you might start seeing patterns of how they were made, or you can at least think of them as patterns.
The author, Ryan Moulton, an AI engineer from Google Research, took the sacred library paintings by James Gurney and put them into VQGAN, a generative deep neural network, which he didn’t train himself, plus some other tools. “The code did the rest.” he said, the code that he didn’t even write. He just tinkered with existing tools because he wanted to see if AI can paint like James Gurney.
This is how Moulton explains the idea behind his experiment:
I found myself experiencing really complex emotions as I was making this. First was simple wonder. I described a place that I thought would be beautiful, painted by an artist that I thought would make it beautiful, and it was. I typed, “Great Hall of the Sacred Library, by James Gurney,” and was absolutely blown away by the image that gradually materialized. I felt compelled to keep exploring it, and creating it. What else was there in this place? In this world? What other wonders could I find just around the next corner? All I had to do was describe it, and it would show me.
The result was too good and impressive and he couldn’t sleep at night. Moulton was shocked about what he did and the most worrying thing was is it’s only the beginning. AI is in the early stages of development and we can only guess what level of artistry we can see in the future.
I felt shock, a sinking feeling in my gut, and couldn’t sleep for the rest of the night. My daughter wants to be an artist. What should I tell her? Will this be the last generation of stylists, and we’ll just memorize the names of every great 20th century artist to produce things we like, forever?
This sounds scary, almost cyberpunk-ish-ly dystopic and worth a novel or a film where humans will all be doomed and devoured by the feeling of chthonic mediocrity and superiority of machines over them. Not hyperbolising, we will see more of this and AI will get better at generating impressive and enthralling pictures, real or surreal, and, perhaps, less uncanny and unrecognisable from what humans could do or imagine to do.
But there is one nuance, relevant if not in the future but at least now, that I think is important to emphasize.
It wasn’t only AI who created these pictures and came up with ideas for these pictures. There are two important components – data and a human who runs the algorithm. The main data is James Gurney’s paintings. The human, well, Ryan Moulton. He claims he is not an artist but it was his idea to see if AI can paint like Gurney. He got inspired by it and pushed the right buttons, literally and figuratively. Which is how the artistic process works in nutshell every time you sit down to draw or write something.
It can be deliberate or subconscious but you always take something as a reference, build connections that have never been built before to create something. It is rarely original but it is always at least a unique combination of things. It is always a remix. It is not a process of generation but rather a process of discovery. Therefore the machine is merely a discovery tool, a brush in artists hands painting on digital canvas of their imagination.
The same works for anything "created" by a computer. AI doesn't evolve and doesn't come up with new things completely by itself (yet, **nervous giggle**) – the first and the last decision is up to us.
In the process of generating art, AI is just a tool, such as a pen and paper, brushes, camera, or Photoshop, or Blender. One in the stack of the instruments an artist can use. If a person is able to tinker with these tools they can create anything they want, even better than large studios, and they will be handsomely rewarded for their skills and ideas (especially now, in the world where we have NFTs).
It is still hard to comprehend for me but after pondering on it for weeks and finally writing this essay, I am able to cope with the sense of what the relationship between us, AI and artfulness is going to be.
"Most of all I dread mediocrity: a work should either be very good or very bad, but, for its life, not mediocre. Mediocrity that takes up thirty printed sheets is something quite unpardonable." - Dostoevsky
I don't want to speculate about what kinds of google-eyed cats resembling octopuses, or colourful fractals resembling corals, or 24x24 matrices of pixels resembling humans machines we can create with computers. It is impossible to predict which is a pointless activity to engage with. I want you and me to get back to the "imaginary" world in which a machine can generate (discover) anything and everything for us. Human imagination together with machine’s artificial imagination is a combination that can bring to life any artful idea, even the craziest, revealing to us the unseen latent space.
To achieve such results in cinema, for example, now we need hundreds of people to work together for a few years using expensive hardware and the most advanced software. An indie artist doesn't have access to the full power of it. That's why, I believe, we see dozens of generic blockbusters. Studious aren't ready to take risks and their films become high-budget toy advertisements – "audiovisual entertainment, not cinema", as one person who I admire wrote (arguable though).
Perhaps, it's a necessary step in the development of this industry. The stage we have to accept and endure through. Whatever it is, it moves the development of technology and tools artists can use forward bringing the future closer to us. But perhaps, at some point in this approaching future, we will get tired of generic plots, weak drama and Mary Sue characters. At the same time, CGI will get better and cheaper allowing all of us, including independent artists, or at least small studios, to avoid lucrative producers, start taking risks and creating what we always wanted to. Something visionary. Sometimes great.
On the other side, soon, we will see a Netflix movie or animation entirely generated by a computer, watchable, not terrible, formulaic, most likely mediocre, but maybe... great? Then, hundreds and thousands of them, personalised and wrapped into an endless TikTok loop of blockbusters.
But cinema is just one example. Take writing, for instance. Soon (sooner than autogenerated cinema), the world will be drowning in the generated textual content. It will be easy and cheap. You won’t notice the change and hence won’t be surprised – a big fraction of what you see and read on the internet already seems like something created by robots for robots. Creators of such content, just like big cinema studios, don't take risks. They simply do whatever works in the safest possible way.
A mediocre way.
But machines will provide all of us with tools, all the tools artists need to fulfil their most visionary creative pursuits and spend time being better than mediocre. We will get a powerful toolbox, a toolbox able to replace them, and all articles "10 ways to become better human", "How to become a better writer?", "10 lessons from Atobic Hamits" and "How to get rich quick" will be AI-generated, solely based on texts written by humans in the past but that's what those 'creators' do anyway, right?
Machines will discover in their latent space of artificial imagination all the mediocre content instead of mediocre us doing that. We will be drowning and wallowing in mediocrity, suffocating and losing ourselves in the deserts and oceans content. We will have to reconsider what is artfulness, what is great, what is terrible and what is mediocrity. The only way to survive in this new world as a human and as an artist will be by avoiding the middle, by doing something bold, by doing something, again, visionary. Performing.
My only humble advice is to start doing it today.