Artistry In The Age of GPT-3
Disclaimer: This blog post was NOT written by an AI.
Imagine a story containing Claude Shannon, Alan Turing…and Harry Potter. Now, imagine it being written by an AI.
OpenAI’s GPT-3 or (Generative Pretrained Transformer, 3) has recently surprised many at its ability to be creative. Manuel Araoz made it write a whole blog post about itself.
People have made it write code.
If parsed correctly, it can even create music. This is done with its predecessor, GPT-2.
All of this from a simple task: predict the next words.
Art?!
As with generative art, which has a longer history, a question often arises about the nature of AI (using the term broadly here), and its ability to create. Will it replace human creators? Will the next great story be written by an AI? Will we care as much? Considering that AI, like GPT-3, is informed by existing data, is its creations considered, ‘original’?
It’s all very interesting questions. With GPT-3, it really feels, for the first time, that it could generically create coherent outputs. It can create by mixing together what exists, and in doing so write dialogue about Shannon, Turing, and Harry Potter. In that sense, we can say, with certainty that AI can be original. But that’s a low bar. It’s always been the case. If you task a program to spit out random dots on a screen, it’s also ‘original’. I think the more interesting question is what it means for artistry, and meaning, in general.
AI Intent
Imagine the following: an AI and a human creates the same creative output. Would you judge or experience the output in a similar manner?
In the age of abundance, when a program, could create any, and all creative work, then its differentiating factor becomes its creation. It becomes what’s left unsaid, unheard, or unseen. What matters is not just what’s there.
It’s because creative works, serves a purpose for both its creator and audience. An AI can create a tapestry of interesting experiences, but it on its own has no intent (yet, at least). In the same way that we can look at a beautiful shell, or beautiful rock formation, unless you believed in God, has no direct creator. It just is. A creative work from the mind of a person, however, has output that matters to its creator, and in that, it retains some additional meaning.
An AI writing a song about loss can be poignant, and meaningful. But, it is elevated when you understand why it was created. When Eric Clapton sings about “Tears In Heaven”, it’s grief about losing his son. We also, instinctively feel a connection, even when, say a pub singer sings a cover song, but one which means something to them. You don’t even have to be original for it to matter.
It’s why David Bowie espoused: don’t play for the gallery. Great creative work comes from earnest interest in creation as a conversation in and of itself.
AI intent, Nihilism, & Zeitgeist Art
So, perhaps, we can say that generated creative works exist in its own vacuum. It just is, and thus, any additional connotations aren’t attached to it. But, I don’t think it’s true. AI creative works have connotations to them. You can either see it being devoid of intent, or being full of ALL intent.
Devoid of Intent
In its purity, an AI-generated work, doesn’t ask you to consider any connotation to its creation (and thus, creator). This kind of cold detachment does create a new sense of artistic meaning, something that humans aren’t capable of. We can’t create without having intent, even in absurdism. Yes, someone would guide a computer to create, but there won’t necessarily be a causal link of intent. eg, the person that presses the button, isn’t necessarily pressing it to write a song, or create a story.
Thus: AI-generated works allow us to contemplate creativity as itself. In that space, the output becomes a way to see chaos without necessarily having a directed hand of intent. It could be garbage, words mashed together that have no meaning, or interesting dialogue between Turing and Shannon about magic. It just is. Somewhere in there is an interesting experience. Feels like meditation as creative output.
Full of ALL Intent
The alternative is to consider that with any AI, especially something like GPT-3, is is trained with inputs. It’s estimated that GPT-3, “would require 355 years and $4,600,000 to train - even with the lowest priced GPU cloud on the market”.
It’s trained on a corpus of about 500 billion tokens, with a token, essentially constituting, a word. It is, thus, us. Humanity (and some of our machines). Thus, when, it creates, its intent is a zeitgeist. It’s all of us. And that is also new. When you see GPT-3 spit out a new story, it’s because it was primed from perhaps, reading my blog. I am in all those outputs. You too.
The Second Brain
I feel that, as humans, our creative works are inevitably tied to who we are. If the creative output is the same, the person behind the work matters. I adore the idea of using AI as a second-brain, a way to dance with it, to tease out the adjacent possible. We create our most interesting works when we test our own boundaries. Hearing harmonies when throwing random notes on a piano roll, or throwing new colours together on a canvas. Holly Herndon, who along with Mat Dryhurst, created an AI for Holly to make music alongside with, put it eloquently.
“I don’t want to recreate music; I want to find a new sound and a new aesthetic. The major difference is that we see Spawn as an ensemble member, rather than a composer. Even if she’s improvising, as performers do, she’s not writing the piece. I want to write the music!”
With GPT-3, we can more readily push these boundaries, a companion, a muse, to create with. I’m always reminded of Fan Hui’s commentary on seeing Google’s AlphaGo execute a move it had never seen before.
At first, Fan Hui thought the move was rather odd. But then he saw its beauty.
"It's not a human move. I've never seen a human play this move," he says. "So beautiful." It's a word he keeps repeating. Beautiful. Beautiful. Beautiful.
Regardless of how we will engage with this new frontier of creativity and artistry, it’s ultimately, very exciting. Almost, in some sense, humbling. There’s beauty, wherever we look, and that too can come from a program tasked to predict the next word.
