AI is, to many creatives, frankly terrifying. Until now, technology has largely threatened repetitive, uncreative work. But now, AI can create. It can bring text and images into the world that have not existed before. And video and music are on the horizon, too.

Is this the doom of creative endeavour? Or should it be a trigger for a revitalisation of creative output in the face of new competition? At first glance, the doom-mongers have a point. How will any form of art or creativity survive, when you can spend a few seconds generating something nearly as good with a modern AI?

Well, for a start, quite a lot of the output of AI is, well, not good. They don’t truly understand what we ask of them — they just guess as to what the best output will be based on the input we give. That’s why there are such problems with AI hallucinations, in text and images. A machine designed to make things up tends to make things up. And sometimes trying to intervene in the model just makes things worse.

The blessing of creative competition

But that’s also the heart of the issue. Modern generative AIs are, unlike their ancestor algorithms, designed to generate new things, not follow pre-planned pathways. They are creative in as much as they do that. And so, any of us in a creative profession has to face the reality that we’re suddenly dealing with a competitor that is, for any given output, much cheaper than us. (Although, their actual costs, especially their environmental ones, are being obfuscated.)

We can fight back — and there are court cases doing that right now. Or we can work with the new competitor to see what new things we can create.

A lot of creative innovation comes from a mix of cooperation and competition. Look at the crowd of writers known as The Inklings, of whom CS Lewis and J. R. R. Tolkien are the most famous members. Their work, and that of Tolkien in particular, essentially birthed a whole new genre of fiction. No mean feat. And we could look at any number of examples, from Der Blaue Reiter in Munich in the early 20th century, to the cluster of Silicon Valley founders who created the digital revolution as we know it.

AI as creative antagonist

Could we welcome AI creativity into our communities, to give new energy to what we produce?

We’ve talked before about how NEXT23 speaker Harry Yeff uses an AI version of his voice to force himself to get better. For him, AI switches between being antagonistic intelligence, where it forces him to get better, and assistive intelligence, when it inspires other work. And this is another way we can embrace AI without letting it supplant us.

Taking Yeff’s idea and running with it, you can see AI’s output as the new bare minimum of creativity. If you can’t do something that is, in some way, better than the AI produces, then you either need to improve or you need to get out of the game. Harsh — but a common reality in a period of digital-mediated change. Just as an artist might do sketches or use photographs for inspiration, so too might a creative use AI-produced ideas as a starting point. A photographer could use AI output as a source of composition he can build on with more control once she’s on location. A musician can interweave AI-created music with their own melodies.

Short of a massive change in copyright law which changes the economics of AI, these tools are not going away. We need to find a way to live with them.

And that means making sure our work escapes the mundane.

Against the mundane

The corporate art world has shown a marked lack of creativity of late. The style that’s known as “corporate Memphis” — think of the people illustrations Facebook uses — is everywhere. And critics are not enthused, even if the average consumer seems to find it amiable enough:

Corporate Memphis is flamboyant. It’s spirited. It’s joyful. It’s most certainly prepossessing to the general public. But it’s exactly these elements that warrant greater scrutiny. Illustrations in the style, with its aggressively friendly expressions, portray a world that is uncannily utopian. The deliberate oversimplification in Corporate Memphis can give a false sense of security in situations where the opposite behavior should be observed.

This style emerged in 2017, or thereabouts. It’s still in use today. That’s a degree of creative stagnation that’s not appealing — and which can easily be recreated by AIs. I generated this image, which is exactly in the right style, using Midjourney in under a minute:

If you’re looking for a field of creative work to get out of, this is one…

AI needs human creativity

My image above, though, is very close to a parlour trick. I’ve specifically targeted a field where creativity has been supplanted by familiarity, known and conforming patterns of unchallenging art. This is where AI loves to operate, if you’ll excuse me for attributing emotions to a system that has none.

To some degree, the creativity of AI is oversold. LLMs are not truly creative. They are very sophisticated guessing machines, that predict what the output should be based on patterns they’ve seen before. That’s why they’re great at working with code, for example, because it follows predictable patterns. But art is more challenging because it doesn’t. The more codified the art style, though, the more easily it can be replicated. AIs can only create something genuinely innovative accidentally — and are dependent on their user to realise when it’s happened.

They can’t create in the way humans can, they can only work within the boundaries already established. And that’s why, in fact, LLMs need human creativity more than anyone.

Why AI needs human creative work

LLMs need to be trained to improve. So, they require new examples of human creativity to learn from — because training LLMs on AI-created material does not go well. It creates a problem known as Model Autophagy Disorder:

Feed a model enough of this “synthetic” data, and the quality of the AI’s answers can rapidly deteriorate, as the systems lock in on the most probable word choices and discard the “tail” choices that keep their output interesting.

So, even the LLM creators need human creativity to still exist, to keep the model supplied. We’re already seeing that play out in journalism and some forms of social media, where AI companies are paying for access to human-created data.

And, besides, human beings create. It’s one of our most basic impulses from early childhood. Sure, artistic expression is sometimes beaten out of us by life, the education system, and corporate culture. But people will still create art. They will still create music. The theatre will not be replaced by AI.

Creative supply and demand

So, ultimately, we have something terribly dull: a demand/supply rebalancing. The market for quickly produced, generic art, for example, is likely to crater. That’s the sort of “creativity” that AI can eat alive.

But specific imagery for articles or reports? Perhaps not. Prompt crafting to get exactly what you want from an AI is a skill, and one that takes time both to develop and maintain as models change. At what point does it become cheaper to work with a human artist, rather than to spend time wresting with a generative AI, or hiring a prompt crafter to do it for you?

And, as our friend David Mattin is fond of pointing out, human beings are essentially status-seeking primates. Something that is in abundant supply — like AI-generated creativity — does not confer status. Human-created art will. And so the economics of creative markets are bound to shift, as mundane creativity is commoditised, but the best human art with a story behind or around it, becomes prized as more valuable.

Humans will create. Other humans will want the results of that creativity — and some will pay. We just have new competition — and new tools to work with. The game’s afoot.

Picture by Jr Korpa from Unsplash.