“Prompt hoarding” and the future of art

Maybe it’s just that I lean towards the verbal over the visual, but the tweet above from Adam Holwerda struck a chord with me. As much as I agree that having serious conversations around algorithmic appropriation of artistic styles and tech firms profiting off of the labour of emerging artists is essential if we want to understand the impact of computer-generated imagery—as someone without much artistic ability, I can’t deny the appeal of being able to come up with ideas for imagery and within seconds seeing how a machine-learning model can approximate my vague visual suggestions. It’s something different from the feeling I get from actual artistic creation, but it was, to put it crudely, neat.

Switch the medium from visual art to the written word, though, and the criticisms of AI hit closer to home for me. The idea that “future writers are hoarders of prompts” strikes me as deeply dystopian, an abdication of the creative impulse to something superficially related but profoundly different. Because it reduces art to strictly something to be consumed, ignoring the other, literally creative side of the creative process—which is a profound and deeply rooted human drive, with its own inherent value for the creator.

The art that machine-learning creates is, in a way, utilitarian. This isn’t to say it can’t also be beautiful, just that regardless of the intent of the prompter, the goal of the software isn’t expression, it is matching a set of specifications. Its decisions stem from some mix of subject and style that is specified in an initial prompt and likely refined through a series of iterations, the text returning an image that inspires tweaks to the prompt, which returns slightly different images, and so on until an image arrives that is either sufficiently close to the prompter’s original vision, or interesting enough in its own right to be chosen as an end point.

In this sense, creating AI art, at least with our current tools and models, is somewhere between a commission and a negotiation (which all commissions are, to some extent). This isn’t to be dismissive of commissioned pieces, which account for a significant portion of Western art’s canonical works—although notably the credit for those works goes to the artist who made it, not the patron who commissioned it. Some artists are creating fascinating and compelling works with AI tools already, and I’m sure those works will only get more sophisticated. But it strikes me as a very different process from what I’ll call “direct creation” for lack of a better term (acknowledging that all artistic creation is mediated and indirect to at least some extent).

Direct creation involves a different sort of negotiation—a constant self-negotiation, both conscious and unconscious, to refine the ideas you’re exploring. This is something I’m much more familiar with in terms of the written word, so I’ll focus on that here: the value of writing for me isn’t in having an idea and formalizing it in words, but in having an impulse and working it through in the process of writing. It can be slow and painful and full of revision; it can be wonderfully quick in rare instances; but however it happens, it’s a process that relies on reflection, self-knowledge, and some degree of personal growth, however small or indirect.

Getting a finished novel that matched your initial idea to a tee strikes me as almost a monkey’s paw situation. It eliminates the possibility for growth in the artistic process, replacing it with something closer to wish fulfilment. Even ignoring the fact that the algorithmic version of art is one that almost by definiton limits itself to styles and techniques that have already been created, archived, and tagged as art, this sort of creation taken to its extreme (and the tech isn’t there yet) is essentially stagnant. At the risk of romanticizing struggle, eliminating any friction between the creative impulse and its execution robs the artistic process of opportunities for personal growth. You’re no longer working through ideas if the idea leads immediately to the finished work.

All that said, I could see the publishing industry going the route of AI-generated novels, as Holwerda’s tweet imagines. If it went that way, I doubt authors and publishers would be involved at all. It’d more likely be algorithms all the way down, a mix of trending topics and deep personalization that wouldn’t need human interference to maximize engagement and profit. If writing novels can be handed over to the machines, generating ideas for those novels certainly could be, too, with other algorithms surfacing the content most likely to be consumed by each individual. The notion that prompt generation is somehow more immune to automation than any other artistic field strikes me as almost wilfully naive.

The impulse for people to tell stories isn’t going to go away, though. The personal value in hammering out specific wordings, developing metaphors, and working through your own contradictory thoughts is probably significant enough to outweigh the easier but relatively shallower process of creating written works from prompt-generation, at least for a significant subset of people who have the impulse to write.

Maybe what ultimately comes of this is the separation of content generation from artistic creation. When a version of every imaginable image or premise (or melody or whatever else) is available with a few second’s effort, the consumptive side of art, of getting exactly the niche content we want to see, will be as easy as clicking OK. But the need for the creative side of art will still remain. The need to make things—to work through thoughts, to fine-tune compositions, to put your fingers on an instrument and see what sounds you can coax out of it—that experience of creation will remain essential to a well-rounded life.

Our relationship to art has changed dramatically in the last century. Before recorded music was widespread, group singing was common. Recorded music led to professionalization and a percieved separation between performers and listeners that has contributed to the idea that making music is a rarified skill instead of a fundamental part of being human. We’ve become increasingly isolated from forms of expression that should be as natural as breathing.

The onset of AI art might make that worse. If algorithmic art can achieve the sublime, or even approximate it, the entertainment industry is in for a profound reckoning. It might be enough to scare off human creators from even trying to match machine-generated works. Or, it might be freeing. It’s a long shot, but the glut of content might be the reminder we need that the value of art is as at least as much in its creation as in its consumption. At the very least, it’s unlikely that the impulse to create the old fashioned way will ever totally disappear. Whether those more hand-made creations hold any economic value in an algorithmically generated mass entertainment complex, well, only time will tell.