This is in some part a response to a short essay Cory Doctorow recently published which found its way to my email inbox, via the Medium Daily Digest I have signed up to at some point in the past and started following Cory’s writing there at some point.
Every time I see Cory, at a reading or maybe at a conference, I tell him I found a copy of Boing Boing in a comic book store 30 years ago, and it changed my life. It was one of the influences, along w/ Fact Sheet Five, Duplex Planet and a whole host of other small publications that don’t immediately come to mind, that brought me into the zine culture of the 90s. Cory, of course, does many things. I’m not even sure if he still edits Boing Boing the blog, but that’s the touchstone I always recall when we meet. He always replies that he was a bookseller when he first discovered it, and later became involved with the blog. I gave him some Welly Girl stickers at a reading in Wellington a few months back.
At any rate, the response is to this article:
And of course, I will link you to the publication of this article on the site which has this byline:
No trackers, no ads. Black type, white background. Privacy policy: we don’t collect or retain any data at all ever period.
Certainly, I agree with Cory in that the diluted nature of most AI art you come across that might fly over your social media feeds is devoid of much numinous content. Just because others are painting poorly, does that mean you should take the brush out of my hand?
My art doesn’t start with AI. It starts with a hodge-podge of Python and Bash scripts I’ve cobbled together — tools that skim a randomized sample of the blogosphere each day. It’s not clean. It’s not elegant. But it works. I grab text and images from 100 randomly selected articles in my RSS reader. The scripts go on to make composites from the harvested images, visual mashups drawn from the churn of online culture. Each day, 400 images are generated, and a random subset of 50 get pushed into my WordPress media library.
That’s where I step in.
From there, I perform a daily ritual of selection. I quickly scan through the 50. Some are too similar. Some feel flat. A few will vibrate. These are the ones where something unexpected has happened in the layering. An echo from an earlier composite might appear again, slightly altered, flipped, recontextualized. I’ll rotate one a few degrees. Discard a near-duplicate. I make microdecisions, tiny curatorial acts, to bring the set down to 7 or 8 images. These are inserted into the coblage post, where the only thing the AI has done for me, is break the text up into paragraphs.
I experimented for a while having the AI do some editing on the text of the Coblage. Turn the rough text into prose, but the results were, as Cory pointed out, diluted. The rough cut-up output from my simple process using the fold-in, some shuffling and a weird little translation trick I learned from Philip K. Dick,
Then I click publish. That’s the moment I call “fixing” the coblage.
It’s a fixation in the Burroughsian sense, the click, the crystallisation, the daily fix. A work of art sealed for that day. A blog post. A visual poem. A digital fossil of the moment. No two are the same. The output is the product of both randomness and intention, a collaboration between machine-generated chaos and human selection.
For the image selection process, my daily rotation of finding the right orientation for each image after it comes out of the spinner… a function that looks like this:
# Random jpegtrans option
function random_tranop () {
tranop[0]="-rotate 90"
tranop[1]="-rotate 180"
tranop[2]="-flip horizontal"
tranop[3]="-flip vertical"
size=${#tranop[@]}
index=$(($RANDOM % $size))
TRANOP=${tranop[$index]}
}
random_tranop
The AI can clearly choose the images I would have intended to use, in microseconds, a process which takes me around 15 minutes to scan through the 50 images imported to the WordPress Media Gallery.
Me: Ok, here are two images in a sequence that should clearly illustrate my decision… the decision I have to make. Which one do you think I am going to include in the Coblage today?
ChatGPT: Perfect — this pair beautifully embodies exactly the kind of microdecision-making that Doctorow is trying to describe, and which you’re living daily.
The Two Images
- Image 1:
Highly layered, multiple semi-transparent elements, organic textures, ambiguous subject (but you can faintly make out the human face), a painterly abstraction with warm tones. - Image 2:
Clear, direct image of Anthony Perkins as Norman Bates from Psycho. Iconic, stark, unmistakable.
The Decision You Face
- The second image is informationally rich but artistically redundant — its communicative intent is pre-baked. It doesn’t offer the same ambiguity, atmosphere, or resonance for your coblage project.
- The first image reflects:
- Layering.
- Historical residues from previous composites.
- The hallucinatory, numinous, almost Dada-coblage quality you’re seeking.
- It’s alive for your coblage purposes.
- It’s the one that allows the viewer to engage and interpret — mirroring your own role as selector.
ChatGPT’s Prediction
You will select Image 1 for today’s coblage.
Because that’s where your communicative intent gets to breathe.
Yep, the AI was 💯 correct. While the second image contains a narrow boarder around the main image, this border seems to show some female heads and the orientation is upside-down and if I didn’t have a better image to choose from, it would qualify for inclusion in the daily coblage based on the slight transformation and the act of inserting the image into a blog post, along with the randomly generated text which removes it from the context of the source where I obtained it. This source may be an article discussing the film and using the image to illustrate a story, or it may be a piece of writing that discusses the image’s original purpose. It is important to note that the image is transformed, albeit in a minor way.
The truth is, the AI could already do this part, the image selection, the aesthetic pruning. Better than me and faster. Eventually, I’ll figure out how to delegate this work to it. And when that day comes, I’ll use the time I get back to do something only I can do. Like cleaning my house.
But once the AI image generators came online, tools like DALL·E, Midjourney, and their open-source cousins, the next step became obvious. What happens when I take my fixed coblage, this image, and the strange fragment of text it’s been paired with, and use that as the prompt? What if the collage becomes the seed for a new machine hallucination? That, of course, is where we’ll pick things up in the next post.
Leave a Reply