AI and Creativity
Three Things #115: April 14, 2024
I’ve long been fascinated by abstract but important ideas like innovation and creativity. We’ve studied these things for years but we still don’t really understand them. Where do they come from? Why are some people so much more creative than others? Why do some people go through creative periods (Einstein’s annus mirabilis being perhaps the most famous example) and then stagnate? Why do some societies produce more creative, innovative people and more innovations than others?
These are important questions because innovation and creativity are economic drivers for individuals and families, communities, countries, and indeed human society at large. They’re also questions that are highly relevant to our work on Spacemesh, which I view at least partially (and maybe ultimately) as an engine for driving human creativity.
And these questions are especially relevant and interesting in an age when AI agents and tools are becoming more and more creative. After spending plenty of time with generative art and LLM tools over the past couple of years I thought nothing would surprise me, but I was both surprised and delighted recently to play with a new generative music tool called Suno. The songs it creates are are at least as good as most human-created music. It got me thinking again about AI, creativity, and long tail distributions.
Thing #1: History 📜
Before considering where things are today or where they’re going it’s worth reflecting for a moment on the way things used to be. When I say “used to be,” I mean the way things were for millennia until very recently, until the advent of mass media and the Internet.
Historically human society was local. What does that mean? It means that most people were born, lived, and died within a 50 mile radius. It means that people were only really aware of things going on in their immediate vicinity and that news about things that happened far away, even contemporaneously, would only arrive after a long time and after a lot of filtering (think: an epic game of telephone); thus fact was transfigured into fantasy. It meant that people lived within a Dunbar’s number “tribe” of on the order of 100 people and rarely interacted with people outside this group; when they did, they did so with mistrust because they didn’t have the tools of social scalability to extend trust to strangers.
It also meant that notions like identity and time that we think of today as universal were also inherently local. Before the advent of the industrial revolution, trains, and other forms of mass transit, there was no universal notion of time. Every city had its own time, dictated by a clock tower, and that was fine. People didn’t move between cities very often and when they did precise time didn’t matter. Cities didn’t need to coordinate much with each other. A person’s identity was also inherently local: John the Fishmonger was just John. He didn’t need a national ID number or a surname or even a DOB, he didn’t ever need to show ID, and he didn’t need to fit a schema and be a unique row in anyone’s database table.
Reflect on this for a moment, and on the incredible paradigm shifts that humanity has undergone during the past few generations due to regionalization, nationalization, and now globalization. Life today really would be unrecognizable and unthinkable to someone born just 200 years ago.
What about content and creativity? They were mostly local, too: content was produced and consumed locally. The most prevalent form of content throughout human history has always been storytelling: cultural ideas, myths, and legends transmitted by word of mouth. This is an inherently local process. A storyteller didn’t have much competition! They just had to be better than their neighbors, who were anyway probably busy with farming or fishing or smithing or something more productive. There were certainly traveling storytellers and famous playwrights but, again, their competition was pretty limited since most people throughout most of human history weren’t even literate. The bar was pretty low.
It all began to change with the advent of the printing press, which eventually led to the first form of mass media: books and pamphlets. The effects were slow and it took a few generations until ideas began to spread more widely and more efficiently. 150-200 years after the printing press was invented it was still possible for one person to read every book ever written in a single lifetime! Books led to the rise of the first widely popular content creators: authors who in the extreme could reach an audience of millions. But the bar was still pretty high for writing and publishing books, making it impossible to target a niche audience for a long time.
Obviously new mass media technology like telegraph, film, radio, and television sped up this process even further and began to give rise to the first stars known nationally or even internationally. Competition for producing popular content was rising quickly; content creators, producers, and publishers needed to be much more creative than previously to stand out from the crowd. Little did they know how much everything was about to change with the advent of the Internet.
Thing #2: Long Tail 🐒
In 2006 Chris Anderson, then editor in chief of Wired, published an important book called The Long Tail. This book does the best job of explaining the paradigm shift described above, driven by the rise of mass media and accelerated by the Internet and the Web. The gist of Anderson’s message is that mass media in general and the Internet and Web in particular have a tendency to push content consumption in both directions along a curve. Whereas historically most content producers didn’t have much competition and would have a local audience on the order of a hundred to maybe a thousand, or in extreme cases maybe ten thousand, globalization and the Internet has led to the rise of both megastars and lots of niche content.
Because content creators today are able to reach a national or even a global audience the top stars have followings in the hundreds of millions, something that was never possible before. The “curve” is therefore dominated by a small number of stars who produce “pop” content that appeals to a mainstream audience. At the opposite end are the niche producers who would never have been able to support themselves as content producers prior to the Internet. This is because the Internet enabled aggregation of demand for niche content and services over vast distances. There may be only a handful of vegans, travel hackers, or garage gym junkies in a small town, not nearly enough to support local shops dedicated to these things. But there are many thousands of these people scattered around the country and they can all find each other and find related content, services, and products online.
Modern mass media and technology has thus had a bifurcating effect, tending to push creators of content, products, and services in one of two directions along this global curve. While you can still run a mediocre hair salon or pizza shop in town—since haircuts and pizza slices don’t travel well—mediocre local content like local newspapers and magazines are all but dead at this point. Local has died and we mourn its passing.
Producers today have two choices: go big or stay niche. They can try to “make it big” and produce content that appeals to a mass audience but they risk “selling out” (watering down their product and message) in the process and potentially alienating early followers. Or they can decide to target a global, niche audience with content, products, or services that are highly relevant to that group but stand no chance of appealing to a mass audience, although this second strategy inherently limits the potential market size and thus potential growth and profitability. There’s no longer an option to stay in the comfortable middle of the curve, producing content that appeals to a medium-size mainstream audience, due to fragmentation and competition. Such an audience no longer exists, and it’ll never be profitable or sustainable.
The result? Today we live in a golden age of content. There’s no shortage of flashy but unsatisfying “pop” content: movies, music, trashy romance novels, etc. As audiences have grown pop content like this has gotten more and more competitive and more expensive which has had the sad effect of reducing creativity on the part of content creators. This is the main reason mainstream film has stagnated for the past decade and today almost entirely consists of sequels and expansions of existing film franchises like Marvel and DC. These are relatively low risk for producers, but also bland and uninspiring.
And in the long tail we also have an incredible long tail of niche content including blogs, random news sites, podcasts, YouTube, and of course social media. AI-generated content is just beginning to appear in some of these places. The floodgates haven’t yet been opened but all of that is about to change.
Thing #3: Singularity 🦾
We’re now entering a new paradigm shift around how content is created, distributed, and consumed, one driven by AI tools and specifically by generative AI technology. On the face of it this shift might seem completely different than the one that came before—cue headlines about the AI singularity, about AI taking over, and about fake AI-generated content swamping the world. But I believe it’ll have very similar effects to the shift that came before, and thus in a sense could even be thought of as a continuation or a second phase of the same paradigm shift. These effects include efficiency, audience growth, aggregation, and the long tail effect described above, as well as changing the rules of the game for human creators.
Let’s start with efficiency. The printing press made publishing orders of magnitude faster, cheaper, and easier than writing books by hand. Other forms of mass media including radio, television, film, and the Internet made it more efficient still. It may seem like publishing can’t get much more efficient than it is today! I can type this article in a matter of hours, hit publish, and an infinite number of copies appear instantly around the world for close to zero cost.
But it is becoming even more efficient thanks to AI tools. Rather than research and write the article myself I can instead “meta write” the article by carefully constructing a prompt describing what I want to say, in what order and in what style, and then let the LLM do the work of writing the article for me. All that remains is to proof it and hit publish. What took a few hours could in theory be done in minutes. I’ve experimented a bit with LLMs and I don’t feel that their writing is up to snuff yet: the grammar is excellent and the points are well structured but the prose is formulaic, predictable (surprise surprise) and dry. But it will continue to get better!
Next up is audience size and growth. One of the most obvious effects of mass media was to dramatically increase the maximum size of a publisher’s potential audience. In the past a storyteller or writer catering to a local audience could expect 100 or, at most, 1000 fans. A good storyteller today can realistically reach millions. How will AI tools increase audience size further? The answer seems self evident. For one thing AI will perform automatic translation, making content instantly available in every language. This is sort of possible today, but no one likes reading long form content passed through an auto-translator. While the meaning is more or less conveyed it sounds unnatural and unpleasant. AI will make the experience much better, providing not only smoother, more natural language but also big picture translation of cultural nuances that are lost in anything but the best, most thoughtful and creative human translation. It’ll also allow content to be translated across mediums and styles, which will also vastly increase the potential audience size. Don’t like social media apps or infinite scroll? Your AI agent will summarize the day’s happenings in long form, or in podcast form, or as a magazine, or as a TV broadcast if you prefer.
How will AI help with aggregation? Quite simply by making indexing, search, and retrieval much faster and easier. The Internet allowed niche audiences to find each other for the first time on places like bulletin boards, newsgroups, email lists, and eventually Facebook groups, blogs, and podcasts. But the search and discovery process is still quite manual and can be slow and frustrating, especially for truly niche content. I can easily imagine an AI agent that acts like “everyone’s friend” in the sense that it knows exactly what everyone’s up to and what everyone’s interests are. Tell it your interests or what you’re curious about, ask for a recommendation, and it’ll give you a warm intro to exactly the group you were looking for and might not have found otherwise. In this way it’ll bring niche audiences even closer together than is possible today (and might even give rise to some new ones).
Finally, what about the long tail effect? I think this effect will be extraordinarily strong in the AI future with one important caveat: AI-generated content will fill the entire middle of the curve, pushing remaining human content creators even further to either extreme. I don’t think it’ll be too long before we see a true global AI pop star with an audience in the millions—it may sound strange but it’s already happening in Asia.
However, the few pop stars at the very top of the curve, the Taylor Swifts of the world, will still be human. They may use AI tools to help them generate content or appeal to different audiences but their brands will be very much personal and human. They’ll release more and higher quality content than ever before and they’ll reach a bigger audience than ever before, thanks to AI. I don’t actually see a fundamental technical or social obstacle to virtual mega pop stars, but I do think that human artists will continue to be more creative than their digital counterparts for the foreseeable future.
At the opposite end of the curve, in the long tail, human content creators and publishers will continue to find niche audiences but those audiences will shrink. Any profitable, large enough niche will be quickly colonized by an AI. It’s not hard to imagine profitable AI publishing houses with advanced agents that consume, organize, produce, and curate content for audiences as small as a thousand people. This content could include articles, memes, social media posts, images, music, magazines, even podcasts and books. But there will always be tiny niches that AI fails to discover, fails to understand, or isn’t allowed to participate in for whatever reason.
Meanwhile AI will colonize the entire rest of the curve because AI tools will be able to generate art that’s better than what 99% of human artists can create, music that’s better than 99% of the music out there today, and writing that’s better than that of any but the very greatest human authors. This is literally already happening, as the example I mentioned of Suno shows.
What does this mean for human creators and for human creativity? Human creativity isn’t going away but its nature is evolving. For one thing, “meta creativity” will start to matter more due to prompt engineering. The most successful artists, writers, film producers, and musicians will no longer necessarily be the people who have a specific vision for what they want to create or who do all the legwork themselves. Instead they’ll be the people who have a rough outline and can describe it succinctly, and who can orchestrate an army of AI agents who will do most of the work of bringing it to life. To be clear this is still a form of creativity, discrimination, and taste, not to mention hard work, but it’s a different sort of creativity, more about logistics, management, and complexity than sheer creativity as we think of it today. It’s akin to saying that the greatest builders today aren’t those who are best with the hammer and saw, but rather those who can inspire and organize others to bring their vision to life.
For another thing it means that creators need to embrace these tools and start incorporating them into their daily workflow. If I were creating a video game, a book, or a film today I’d seriously consider having AI tools generate large portions of the project—they really are good enough. As the effects described above intensify, I’m sure we’ll see creators who are unwilling or unable to adopt these tools to be forced further and further down the curve, and many will simply fail to compete. Many will simply choose to retire rather than compete with AIs or be forced to learn new tools, which will further accelerate the transition.
It’s not all doom and gloom. To the AI doomers I like to cite the example of photography. At the time photography emerged painters were primarily employed in portraiture. There were fears that photography would destroy this profession and put painters out of work. This may have happened in the short term. But over the medium to long term it instead had a powerful liberating effect: it freed painters to pursue more abstract subjects and led to a renaissance of the medium. In retrospect, painting evolved in some important ways in response to photography.
Human creativity is durable and it’s a mistake to bet against it. I think something similar will play out over the next few decades. In the same way that painting in 1900 didn’t resemble painting in 1800—but there were more painters and more great works of art being produced after this transition—I’m certain that we’ll see a renaissance and an explosion of all forms of human creativity thanks to the AI transition. However, the nature of creativity will evolve and the transition will take some time.
In the meantime I’ll keep trying new tools and learning how to incorporate them into my workflow, and how to use them to increase both my productivity and creativity.