Perspectives on AI-Generated Art
March 2024
Midjourney v6 dropped in late 2023, and something shifted in my thinking about AI art. The artifacts were gone. The hands looked like hands. The images felt intentional rather than assembled.
I had been skeptical. The earlier generation of AI art felt like remixing — impressive technically but hollow aesthetically. You could recognize the training data bleeding through. Prompt engineering felt like searching through latent space for something that already existed rather than creating something new.
The new models changed that. The relationship between prompt and output became more like collaboration. You could iterate with the model, refining intent rather than hunting for lucky accidents.
The Authorship Question
The debate about whether AI-generated images constitute "art" seems increasingly unproductive. Better questions: Who makes the meaningful choices? What is being communicated? Does the output reward sustained attention?
A photographer chooses framing, timing, and context. A painter chooses color, composition, and gesture. An AI-assisted creator chooses prompts, iterates, curates, and contextualizes. The chain of intention is longer and more abstracted, but intention persists.
The craft shifts from execution to curation and direction. This isn't unprecedented — photographers have faced similar critiques. Ansel Adams didn't paint his landscapes; he chose where to stand, when to click, and how to develop. The darkroom was a tool. So is the prompt.
The Economic Disruption
Where I think the conversation gets more complicated is economics. Stock photography, concept art, and illustration are being disrupted in real time. Commercial work that once required skilled practitioners now requires skilled prompters — a smaller pool, with different skills.
This isn't the first time technology has displaced artistic labor. Photography disrupted portrait painting. Digital tools disrupted darkroom techniques. Each shift created new forms while rendering others economically marginal.
The difference now is speed. The gap between tool availability and economic impact is months rather than decades. There's little time for adaptation or transition.
Where I Land
AI art is a tool with genuine expressive potential and genuine economic consequences. Dismissing it as "not real art" misses what's interesting. Celebrating it without acknowledging the displacement misses what's harmful.
I'm using these tools more now — for prototyping, for exploration, for generating variations I wouldn't have conceived. But I'm also more conscious of the human labor that trained the models and the human labor being displaced.
The technology will continue improving. The interesting questions aren't about the outputs but about the systems we build around them — compensation for training data, support for displaced workers, and norms for disclosure and attribution.
—
Building Do Little Lab. Previously: SRE, engineering manager, humanitarian relief organizer.