TikTok, Leonardo da Vinci, and the AI-driven Future of Creativity
The device you’re using to read this on almost certainly has more computing power than the entire server farm used to render Toy Story, which looked like this:
It’s no secret that computing hardware has made truly unbelievable strides since the invention of the transistor some 60 years ago. Your smart lightbulb has more computing power than the Apollo lander. Your watch is approximately as fast as the chess-playing supercomputer that beat Garry Kasparov in 1997. Yeah yeah, we get it, whatever grampa.
As much as these factoids are amazing, they also fill me a weird sense of ennui. More precisely, second-hand ennui on behalf of our PCs: a feeling of listlessness and dissatisfaction arising from a lack of occupation or excitement. These things could be rendering movies, flying spaceships, proving theorems, creating art. Instead they send tweets and play games and dutifully wait for you to read through this Medium article. These are race horses living in studio apartments — power with nowhere to run.
I’m anthropomorphizing… or, anthropo-horse-izing. Of course computers don’t actually care if they reach their full potential or not. The truth is I’m disappointed in myself. I have the power to do incredible things, yet I largely just scroll Reddit and watch videos.
It’s not just me, of course. Every day we humans collectively spend 1 billion hours watching YouTube videos, but only only around 3 million hours making them (this is some heavy estimation on my part: 720,000 hours of content is uploaded every day, much of it fairly low-effort). There’s even a name for this phenomenon, the 1% Rule, named for the 1% of internet users who actually create new content.
Maybe that’s ok! There’s nothing wrong with consuming content, especially if you learn something from it — and that includes viral dances. Still, the technology that unlocked the modern era after centuries of relative stagnation (at least in the western world) was not the musket or the telescope but the printing press. Diversifying the market of art and ideas is perhaps the healthiest and most transformative thing a culture can do. Everybody can contribute something to that.
So what’s holding us back? It’s not like we don’t have the necessary tools: fully 97% of Americans own a smartphone, with the global rate estimated to reach 89% (!) this year. It’s also not that we have no desire to create: a 2019 study found that 29% of kids in the US and UK would rather be a YouTuber than an astronaut, musician, teacher, or professional athlete. We can create, we want to create, but we mostly don’t. Why is that?
The Elephant in the Room
One company has successfully defied the 1% Rule: TikTok. A 2019 report by GWI found that a whopping 55% of active TikTok users have uploaded at least one video in the past month. This is certainly not driven by monetization: TikTok only pays a tiny fraction of its users, and even those rare payouts can be as little as 2 cents per video.
Instead, TikTok’s secret is just that it makes video creation incredibly easy. This is partly thanks to its simple UI and partly thanks to gimmicks that can make low-effort videos engaging, like duets and filters. Users can create something they’re proud of with just a couple minutes of work, providing the quick dopamine hits that drive online engagement.
I don’t mean that to sound so cynical. Putting aside any concerns I have about its addictiveness or its potential for propaganda, for the purposes of this article I have to accept TikTok as an incredible success story! The fact that TikTokers don’t just consume content but also so frequently create it, that they can talk as well as they can listen, is an under-rated virtue.
So, amid all the recent calls to ban TikTok, I have a different but related question: How can we make everything else a little bit more like TikTok? If TikTok’s popularity stems from the way it’s made content creation approachable and fun, it stands to reason that replicating that success in other products and for other art-forms might weaken TikTok’s dominance. Plus, it’s good for us as a society!
Blank Page Syndrome
You know that feeling you get when you pull out a crisp new sheet of paper, or open a new Photoshop project? That blank document could become almost anything — the “possibility space” is near infinite. Your task, put simply, is to turn Nothing into Something. It’s thrilling, and terrifying!
It’s telling that all the words and phrases I’ve found to describe that blank-page feeling are inherently negative:
- Writer’s Block
- Horror Vacui (Latin for ‘fear of empty space’)
- Analysis Paralysis
- Blank Page Syndrome
Leonardo da Vinci once said “Art lives from constraints and dies from freedom”. With that in mind, a blank page is the ultimate art-murderer! It imposes no constraints except its physical size, which in the case of a digital project is effectively infinite.
Conversely then, a medium that only allows one particular kind of short-form video may be the ultimate art-cultivator.
Time for some clickbait!
No! But his observation does help explain its success.
What makes TikTok tick
How has TikTok created so many creators? I don’t think there’s any secret sauce, just a few simple UX principles that work well together.
1. Built-in Constraints
TikTok, like Twitter, is a social network defined by its constraints. TikTok videos have to be under 10 minutes long (until recently the limit was only 3 minutes). They’re also almost exclusively watched on smartphones, which means they’re usually in a portrait 9:16 aspect ratio that is conducive to talking-heads but not to things like world-building (there’s a reason Quibi failed). Overall, the medium and its algorithms reward content that is immediately-engaging, off-the-cuff as opposed to highly-produced, and, ideally, loop-able.
When da Vinci talked about constraints he probably meant things like the limited pigments available in his time. Were he alive today though, I bet he’d be a kick-ass Tiktoker — even at the ripe old age of 570. If you’re not convinced, he once said “the smallest feline is a masterpiece”, which is a very TikTok thing to say.
2. No Blank Pages
TikTok’s cure for Blank Page Syndrome is to just, never show you a blank page. Opening TikTok’s video editor doesn’t drop you into “Untitled Project (1)” with a blank timeline, like Adobe Premier does. Instead it shows you something so much worse, your own face.
TikTok has already put some opening strokes down for you. You can see the foundation of your video already: it’s you, in your current room, with your current outfit. If you don’t like any aspect of that you can change it until you’re happy, but you’re never starting from nothing.
3. Prompts
Twitter’s input box asks users “What’s happening?” Facebook prompts “What’s on your mind?” TikTok doesn’t do anything quite so obvious or boring, instead it cleverly adds features that serve as de-facto prompts.
Many of TikTok’s filters conceptually resemble the kinds of creative writing prompts you may remember from High School English class, just zanier: You encounter a dancing hotdog, how do you react? You’re being FaceTimed by a hamster, what do you say to it?
Or the most popular TikTok filter prompt: You suddenly look super hot, how do you make your classmates jealous?
The Duets feature also falls into this category by turning existing content into prompts for new content. Plus, TikTok is well known for its “challenges”, which are more explicit prompts generated ad-hoc by the community.
Prompts are a solution to a less literal version of Blank Page Syndrome — Blank Mind Syndrome. In the same way that TikTok’s video editor pre-fills the screen with your face, everything about TikTok’s design tries to pre-fill your mind with content ideas.
4. Encouraging Exploration
A great TikToker like Leonardo da Vinci might start a new post knowing exactly how they want it to look and sound. Unfortunately, we can’t all be world-class artists with that kind of singular vision.
TikTok does something that traditional creative apps fail at: it makes it easy to experiment with new looks and sounds non-destructively, meaning if you go down a route that doesn’t end up working out you don’t have to hit the Undo button until you’re hopefully back to where you started. At least, in some cases — there are still some destructive edits in the TikTok app. But the easy things (filters, effects, songs, speed-changes) are all try-before-you-apply.
Experimentation is relatively risk-free, which encourages the non-da-Vinci’s among us to poke around recklessly until we’ve finally created our own (relative) masterpieces.
Art and Centaurs
Here’s a creative writing prompt: It’s 2023 and you’re a tech blogger, write a post with a novel take on Artificial Intelligence. Understandably, it’s all people want to read about these days. The problem is I’m not an AI researcher, so I asked AI itself for help:
ChatGPT: Artificial Intelligence (AI) has come a long way since its inception, and we have seen many incredible applications of the technology. From self-driving cars to facial recognition, AI has revolutionized the way we interact with machines and has opened up new possibilities for innovation.
However, as we enter a new era of AI development, it’s becoming increasingly clear that the true power of this technology lies in collaboration. In the past, AI has been seen as a tool for automation and optimization, but now we are beginning to realize that it can also be a powerful tool for collaboration between humans and machines.
It turns out that ChatGPT and I are both fascinated by the concept of centaurs: a cute term for problem-solving methods that involve input from both humans and AIs. The idea originates from the chess world: Deep Blue’s defeat of chess grandmaster Garry Kasparov was not game-over for humans — for a couple decades after that moment, a human player aided by AI software could still beat a pure-AI player.
Writer Gwern Branwen breaks the advancement of automation technology into four phases:
- Subhuman: The technology is able to complete a task passably, but not as well as a reasonably talented human. “the AI is unimportant and used largely in cases where performance doesn’t matter much or where a human is unusable for some reason (such as environment)”
- Human: The technology is able to perform about as well as a human. “as it will be better than many humans and have other advantages; this may prompt a global revolution in that field”
- Superhuman: The technology can out-perform a human, but centaurs may still be the most productive. “The humans using it become more productive and more valuable and employment increases”
- Ultrahuman: “the technology becomes autonomous in the sense that a human no longer contributes to it at all, and that occupation disappears”
In Chess, AI systems have finally reached the Ultrahuman level. To quote our hero Leonardo da Vinci: “Poor is the pupil who does not surpass his master.” Humans no longer add any value to the Chess-playing process.
Gwern also uses the example of calculators, which perform their task so well that we often forget “calculator” used to be a job title for human math-doers.
This is mostly a linear path — most technologies will eventually reach the Ultrahuman level. Given the extremely fast-paced development in Generative AI tools like ChatGPT and Midjourney it’s reasonable to worry that human artists will soon find themselves totally out of work, surpassed in every way by cheap and tireless software.
But art is different. Human calculators are a relic of the past but human painters still exist despite the invention of cameras, which do a far better job of capturing scenes. Speaking loftily: “Where the spirit does not work with the hand, there is no art” (that was, of course, Leonardo da Vinci). Or speaking more practically: Humans like to consume art that has a human story behind it. The Mona Lisa is a great painting, but it’s only worth a billion dollars because of the near-mythic reputation of its creator. No AI-generated art will ever be worth even a fraction of that, no matter how proficient.
There’s also the simple fact that humans like to create art. It’s not a degrading job like weaving textiles or tabulating sums. It’s a calling. People will continue to create art even in a hypothetical future where ultrahuman AI makes their mortal efforts wholly obsolete.
To me the most relevant question isn’t whether generative AI will replace human artists, but how it will aid human artists. This is basically the same point ChatGPT made in response to my prompt: AI is not just a tool for automation, but also (uniquely) a tool for collaboration.
In other fields, human-AI centaurs are doomed to eventually be superseded by ultrahuman AIs. But in art and other creative outlets, the centaurs may reign for the foreseeable future.
The UI of AI
If AI is going to assist humans in our creative endeavors, we’re going to need a better way to interact with it. The groundbreaking AI tools of the last few years (ChatGPT, Dall•E, Midjourney, Gen-1, etc) are effectively standalone toys. They are destinations where an artist can explore ideas, but they do not slot nicely into professional workflows. They export poor-quality “finished” products that are not easily tweak-able, or fixable — practically useless for professional creatives.
But on the opposite extreme, Clippy is a reminder that tightly integrating AI into professional workflows can be more annoying than helpful.
I used Dall•E for many of the images in this post. It’s magical but also frustrating, like a fickle genie. You get 15 free wishes per month, and if you don’t phrase your question exactly right the Dall•E genie will curse you with some unholy abomination of a drawing. There’s no feedback loop, no way to guide Dall•E to improve a mediocre image into a great one. No, you either get your wish, or you start over.
The Human-AI Feedback Loop
Generative AIs are currently somewhere between Subhuman and Human in their artistic abilities. The work they produce is rarely polished, or even coherent: Dall•E is famously bad at drawing the correct number of fingers. Plus, they’re starting to run up against the uncanny valley, which has long stymied videogame artists.
AI is not ready to be in the driver’s seat — it is better suited to the butt of the centaur, with the human in front.
In its simplest form this may just involve adding a button to Photoshop or Premier or Blender that sends the current state of your document to an AI and displays some resulting “hallucinated” alternatives. This may be enough to provide inspiration, but probably not immediately-useful assets. It’s still all-or-nothing, there’s no way to say “I like how it did Da Vinci’s TikTok pose, let me grab that — but his couch looks all wrong”.
I believe the ideal near-term solution is for AIs to learn to use the same creative tools as their human counterparts, creating art not by directly generating pixels or waveforms but rather by simulating clicks in Illustrator or Ableton or some other app. This would let artist edit and remix the AIs creations with ease, or extract the good pieces from it while leaving the hallucinogenic nonsense behind.
In chess centaurs, humans make moves and AIs show likely outcomes that follow from those moves. Creative AIs could work similarly, showing some paths forward as the human works, contributing in real time but never taking control.
Mapping this back to the lessons we learned from TikTok:
- Built-in constraints: Generative AIs currently can do anything, from line art to photorealistic renderings, from whale noises to classical orchestrations. It’s a jack of all trades but a master of none. Having the AI work inside more specialized tool like Illustrator or Reaktor constrains it to a particular kind of work.
- No Blank Pages: If the AI can output a useful starting point instead of (what it considers to be) a finished product, there’s no need to ever start from scratch.
- Prompts: Where do those AI-generated starting points come from? Maybe the artist writes a text prompt for the AI, or maybe the AI just comes up with something totally random and the artist takes inspiration from there.
- Encouraging Exploration: AI-suggested “paths-forward” could show potential outcomes of different ideas, helpfully ruling out some paths quickly and preventing the need for a lot of undoing.
So it begins
So far I’ve talked about creativity in terms of visual art and music, but I’ve glossed over something obvious: AI is already an excellent collaborator for writers. I wrote every word in this article myself (it’d be self-defeating to outsource my own hobby), but I did consult ChatGPT along the way to check my work and get its “thoughts”.
The UI for AI-assisted writing is still a bit awkward, involving a lot of copy-pasting. Plus, with all due respect to ChatGPT, it’s a bit dry as a writer. And yet it’s already helpful to me! That’s the beautiful thing about using AI not for its end results but for fragments and inspiration: the AI can be helpful without even being all that good. All it has to be is interesting.
In a sense, the AI becomes not the artist but the muse.
Coincidentally, GPT3 was codenamed “da Vinci” — but maybe it would’ve been more fitting to call it “Mona Lisa” after the woman who inspired the famous painting.
Is a good muse all it takes to turn the average person into Leonardo da Vinci? Probably not, but it may be enough to turn a good artist into a great one, or even a terrible artist into a passable one. It may lower the barrier to entry enough to overcome the 1% rule. It may finally allow us to use our computers, and our minds, to their greatest potential. Or maybe we’ll all just write horny fan-fic, but who am I to judge — if we’re being honest, Medium posts about AI are basically the nerd equivalent.
