I started typing out a tweet last week in the midst of the Ghibli frenzy. It got long so I slapped a title on to post-ify it, and I wanted to share here as well, with some light editing. Enjoy.
Last week, OpenAI released 4o image mode, their latest image generation offering. It’s capable of some things we haven’t yet seen with AI image models, which you probably witnessed through the Ghiblification craze inundating social media. It really is incredible. But while truly incredible—in fact because of it—the model obscures the technical process it relies upon to create images. It works so well that there's little need to familiarize ourselves intimately with it to better our control. There's no craft to it beyond the verbal. Language is the only interface.
That’s a good thing and a bad thing. It’s a good thing for usability. It makes the technology “indistinguishable from magic.” However it also wrests control away from the user by a less obvious route. It exposes the product, but not the process. While exposing a powerful and smooth interface via natural language, it doesn’t provide any tools beyond this to take real creative liberty with the model, try to break it, make it do weird things, or otherwise come to know the technology well.
Contrast this with open source models, which expose their code to the user, allowing them to become acquainted not only with the product but with the process, the technology. In serving only polished product and eliding the underlying tech, we lose out on the ability to make art that reveals the technology to us in new and unexpected ways.
I'm not critiquing the new model in itself. I'll be using it myself extensively. But I think the best AI work of the last few years has been stuff that gets into the guts of these image models, spills them out, and reincorporates them into the work. This is proper "AI art"—art made with AI to tell us something about AI.
I think the period for this mode of AI image creation still has some time left and people, artists, whoever, should continue to get their hands dirty working with models that don't do exactly what they say.
To do so enables us to more thoughtfully critique and offer improvements to, as well as more deeply appreciate and champion the new technological paradigm by understanding its ins and outs. Whichever ideological direction you swing, your position will be more potent for having familiarized yourself in this way.
Otherwise we accept the tech as given, consuming what has been authoritatively handed to us as a prepackaged product. One of the great roles that artists can fulfill in our society is this kind of exploration, denying the complacency that such prepackaging fosters (whether intentionally or not).
A lot of great art will probably come from thoughtful use of 4o and whatever better models lie beyond, but that art is more likely to be built on top of those models and take them for granted, rather than digging at the roots of the paradigm of generative media itself. That will be a post-AI art, art made from a cultural context where AI is no longer a newfangled thing to gawk at and question, but rather where it’s the water we're swimming in. The boundaries are fuzzy and of course this has already kinda begun (I consider my upcoming collection, Sky World, to be post-AI work), but this model release feels like a real turning point into that era, where the tech works so well that we forget about it and its fruits become table stakes. We’ll lose the patience to engage with anything rougher around the edges, despite the degrees of creative freedom they may afford.
This smoothing, of course, is part of the natural trajectory of technology. But I want us, especially artists, and especially with AI, to avoid the acceptance that it solicits from us. AI, ML, generative media—whatever term for or segment of this nascent technological regime we want to refer to—is world-shaping. If artists can/do/should play a role in its progression, it's as conscientious observers, explorers, and actors (not necessarily objectors). Holly Herndon and Mat Dryhurst are a great example of the way artists can serve this role.
Post-AI art signifies a world where we've relinquished our opportunity for conscientious examination and already accepted this next regime. But its changes are so seismic and the stakes so high that it's crucial we take our time questioning it, exploring its edges, despite the prevailing tailwinds of sociocultural accelerationism.
The rate of development at these technological frontiers is so fast that we hardly have time to think before the opportunity to question has passed. Despite claims of technology being a totally neutral force (it’s not, and that doesn’t mean it’s bad), technological developments always come bundled with implicit ideology, and through sheer speed companies can manufacture tacit consent. Politics can't keep up, so it's largely on culture to do so in defense of itself and its constituents.
One way to contend with this speed is to keep making "AI art" in the sense used above: art that explores the nature of AI (Sasha Stiles and Pindar Van Arman are two great artists of this ilk that come to mind). This kind of work, done aptly, does much more to vie for a better future with this technology than do the all-too-common, unnuanced "AI art is evil" shouts (I’m definitely not saying AI art is beyond reproach, mind you). By incorporating the tools of AI into our practice as a real craft we learn them and come to better know how they might shape our lives.
If for none of these highfalutin reasons, I still urge artists to keep making with tools that expose their tech because it will result in work that is weirder, more distinctive, and more important than the anodyne images of the prepackaged model. Such tools have been essential to my work for the last however long and my career as an artist is much better off for it. Yours probably will be too. The time for them is still now.