Notes on a genre

• You might have noticed that most presen­ta­tions of art produced with the dazzling new image models include the text prompt. The pleasure, it seems, is not in the image; rather, it’s in the spectacle of the computer’s interpretation. I keep expecting this bond to loosen, and the images to find their footing as “mere” art, but the prompt sticks fast.

• Nearly every artist is a genre artist. Form precedes content, even when the artist insists otherwise. There are exceptions, but they are rare, and the path is exceed­ingly dangerous: 10,000 illegible projects for every one with magnetism suffi­cient to re-align the world around its axes.

• In that rare case, success simply founds a new genre.

• AI artists are genre artists, too. Our genre is: “I see what you did there.”

• This recalls the early days of synthesizers; what was Switched-On Bach if not “I see what you did there”? I hope that analogy is right, because the synth provides a healthy, sustain­able template for these tools. Ubiq­ui­tous and unremarkable, control­lable and hackable, with flavors ranging from fully corporate to glori­ously DIY … I’m realizing, as I type this, that synthe­sizers might be one of the truly utopian technologies.

• So it becomes a normative question: what choices, right now, could bend AI’s path towards the docility and diversity of the synth?

• Many AI artists avidly metab­o­lize each new technique or model as it appears, rushing headlong to plumb its new capabilities, last season’s cast aside. The objects of our fasci­na­tion are often research papers and Colab notebooks, not “products”, but even so, this pattern fits (a bit too) comfort­ably into the tech economy of subscrip­tion everything, “accept upgrade”, a new phone every year.

• “All fixed, fast-frozen relations, with their train of ancient and venerable prejudices and opinions, are swept away, all new-formed ones become antiquated before they can ossify.” Oy!

• I hope to see more AI artists identify a technique or model they love and stick with it, even as the state of the art advances, leaving them behind. I don’t mean that in a crusty, regres­sive way; the vibe is not “640K ought to be enough for anybody.” I just want to suggest that it’s fun and satisying to form attach­ments to specific tools; to get good!

• I’m proud of the fact that I’ve spent two solid years using OpenAI’s Jukebox, the same model, never amended or upgraded. I’ve embroi­dered the original code with my own customizations, and I feel like I still haven’t encoun­tered a sliver of lurks inside its labyrinth. I hope they never make another one.

• Jukebox has remained inter­esting and moti­vating to me in a way that none of the language or image models have, and I think that’s partially because it’s so difficult to use. Compared to “traditional” digital music production, nothing about Jukebox is faster; nothing is “better”. It’s just different, and, almost by chance, Jesse and I love that difference.

• Using Jukebox feels like exploring the flooded ruins of a once-great city: exactly that slow, exactly that treacherous, exactly that enticing.

June 2022, Oakland