“there are ghosts in our machines and that our house pets have claws”

December 15, 2022 12:55 PM

The Witching Cats of New Jersey is a short essay by artist and historian Kazys Varnelis about the fashion among the 19th century New Jersey merchant class of commissioning portraits of their cats in the guise of witches’ familiars, most of whom are now kept at the Germantown College Archives. This then becomes an essay about AI generated art, for obvious reasons.
posted by Kattullus (13 comments total)

those images are amazing (by which I mean creepy, and funny, and weird...)

Should have stayed in NJ
posted by supermedusa at 1:11 PM on December 15, 2022 [3 favorites]

it might be more productive to think of them as modern day Ouija boards, accessing not so much the spiritual realm but the collective unconscious as embodied in massive data sets created by humans past and present

This is absolutely what it feels like to play with ChatGPT, and to a lesser extent DALL-E/Midjourney. Like dredging stuff out of the collective unconscious. It's pretty bad at sticking with facts and logic, but it excels at dream logic. You know how text is often sort of fucky in dreams? Well, it's fucky in DALL-E too. Ditto hands.
posted by BungaDunga at 1:17 PM on December 15, 2022 [3 favorites]

Thank you, Kattullus. This is great!
posted by quintessence at 2:06 PM on December 15, 2022 [2 favorites]

Huh. I grew up five doors away from the author.
posted by adamrice at 2:28 PM on December 15, 2022 [2 favorites]

"Bastet, it seems, was both a student and friend of photographer William H. Mumler, noted for his notorious 1869 photograph of well-known cat-lover Mary Todd Lincoln and the ghost of her husband Abraham Lincoln—there are suggestions that Bastet actually pioneered the practice—but also was a friend of Swedenborgian Spiritualist painter George Inness who once told Bastet of his photographs, “Namir, you really have something there.”

If this was fiction, I wouldn't believe it.
posted by clavdivs at 2:52 PM on December 15, 2022 [1 favorite]

I didn't finish the article before I shared it, thinking my wife would be delighted with these witch cats. She was, but then burst my balloon by letting me know they were generated imagery. :(
posted by jellywerker at 5:09 PM on December 15, 2022 [1 favorite]

This is awesome. Thank you.
posted by doctornemo at 5:30 PM on December 15, 2022 [1 favorite]

burst my balloon by letting me know they were generated imagery

I had kind of the same reaction by the end, despite thinking this was a fascinating essay. The first few sets of images (the ones with attribution) do seem to be genuine, but I found the way the author transitioned to using AI-generated images (the later ones, without attribution) and then talked about actual(?) artists whose described styles matched the generated images to be kind of deceptive.

Like, he leads off a paragraph with this: "By the 1820s, however, the Witching Cats found a resurgence in the vogue for alternately amusing and frightening paintings made by self-schooled Primitivist (or Naïve) painters in rural areas" and then has a whole slideshow of images that fit that description, and you only find out that they were made by DALL-E down at the end of the essay.

I think he made a really good point about the similarities between the image distortions in the current generation of AI imagery and that caused by long exposure times whilst filming cats in the early days of photography, but once he went for the big reveal at the end I wound up questioning whether his histories of witching cats were real themselves. Maybe that was the point? But if so, I wish the author had expressed it more plainly.
posted by whir at 8:48 PM on December 15, 2022 [1 favorite]

(Still, I thought this was super interesting, thanks for sharing it Kattullus! This is exactly the kind of thing I hope to find on Metafilter.)
posted by whir at 8:49 PM on December 15, 2022 [1 favorite]

When I was reading the linked article, it was immediately clear that at least some of the images were AI-generated. Some seemed plausibly human-created, and this one just confused me because it had some obviously AI-generated features, but otherwise seemed like it might be human-created. The backstory had some giveaways (artists named Bastet and LeChat, probably others I didn't catch).

But overall, this seems to be part of a new art form: human-generated backstories for AI-generated images. I've seen a number of image galleries on Facebook that get reposted from the Midjourney group to others; some people overlook where the images came from and get worked up into quite a lather—not that the images themselves are fake, but that their purported provenance is. Anyhow, perhaps someday, we'll be able to upload a bunch of AI-generated images to a descendant of ChatGPT and have it generate the backstory for us.
posted by adamrice at 7:57 AM on December 16, 2022 [2 favorites]

human-generated backstories for AI-generated images

You can just ask an AI to generate your fake backstory for your AI-generated images: Photographer fools the internet with AI-generated images of ancient cameras
posted by BungaDunga at 9:13 AM on December 16, 2022 [1 favorite]

(it sounds like Stern gave ChatGPT a fairly detailed prompt for the backstory, but you could probably just feed the Midjourney prompt to ChatGPT and ask it to rephrase it as a museum description and get something superficially plausible. I've tried it with a couple of my own dall-e prompts and it worked quite well)
posted by BungaDunga at 9:23 AM on December 16, 2022

Ok, well I found the salty essay where Varnelis cops to all of this being AI-generated (and also spent some time googling the sources in the attributions he initially uses, like the Germantown College Archives, which are obviously, in retrospect, fictional).

the above essays are better than what 95% of graduate students are capable of doing in an afternoon or, usually, a weekend. This should not be surprising as, while ChatGPT is incapable of original thought, so are 95% of graduate students and, indeed, 95% of faculty.

(The essays he's referring to aren't this one, but rather earlier ChatGPT experiments) ...and, earlier:

I unintentionally duped some readers. [...] Convincing people has never been easier. This project was intended to foreground this.

I mean, weird flex but ok? AITA? I do wish he'd left a few more breadcrumbs, but I guess I'm part of his experiment in "dupe the naive reader" now. The whole thing leaves me feeling a little bitter about online writing where I've been accustomed to high diction being predictive of clear reasoning, which again, is his point I guess. It feels a bit more like a punch in the face than an elbow in the ribs though.
posted by whir at 10:59 PM on December 16, 2022