AI for an eye
a.k.a. A.I. D.I.Y. (The Window Pt. 53)
TL; DR: cans and can’ts of text photography, censorship, and leg numeracy
A dog wearing sunglasses is apparently so dangerous that some text-to-image A.I.’s won’t let you generate one due to their guidelines. So too a glass of champagne, at least when the prompt is “An older woman walking a dog in a stroller which has an attached basket with a champagne glass in it, and the dog is wearing sunglasses”. (For what nefarious reasons would you even ask to generate that, you may inquire? I mean, if you wanted to make something really disturbing, you’d request something like “a man wearing a cat on his shoulder like a parrot*”). The very same artificial intelligence’s safety-ism didn’t stop the image creation (above) of “a man in costume with a cane, on a brick sidewalk, photographed from above”… but then on its own dark whim the AI added a frightening — some might suggest nefarious — third leg on the fellow.
A skeleton hand was also OK to AI into a dumpster, but a handgun and dildo got a guideline warning and were not generated. “WARNING: some words do not match our guidelines and have been removed”:
Text-to-image Cthulhu, H.P. Lovecraft’s ancient octopus-headed demon (who Wiki says is a “source of constant subconscious anxiety for all mankind”, and whose worshippers practice loathsome rituals), got an enthusiastic aye-aye from the AI, so below I am able to show you the Great Old One (at least as seen on the side of his her their new endeavour, a zero-carbon-emission** moving-van franchise):
I had to include some text photography for this series, since The Window project is kindameta, and has touched upon many types of photography — appropriation, surveillance, rephotography, camera obscura, motion capture, Polaroid composites, etc., and the writing has name-checked various lens-based artists, from Adams to Bresson to Maier to Tilmans to Wall…
So as long as I used real The Window photographs as generation-reference (i.e. the Cthulhu image above uses an actual photo from the “U-Haul Psalms” series), it technically doesn’t break my 1 rule: all shots in the 24/7/365 project must be taken from a single location. (The text photography in this post’s grouping is also a type of “Artist steal thyself”, as it uses from scratch my own images and words, which is more interesting than sampling OPA — other people’s art — as inspiration, homage, outright theft, or via AI).
Okay I lied. On one occasion I used AI without an actual physical shot — a real event I witnessed, but missed — to (re)create the scene of two men on a crosswalk, in navy dress whites, passing two women in black burqas. The above was generated a couple years ago, and has some of the lovely grit or funk of the “real” analogue photographs; newer iterations of the text-to-image programs are decidedly far slicker.
The image just below is the imperfect real; the description of it was used to generate the second (machine dreamed) shinier image further below:
No grit, but you have to appreciate the musician’s apparently permeable right cheek, his trumpet-valveish chair, and the beauty (nefariousness?) of his monofoot:
And that’s why text-photography, even censored, is fun at the moment (yeah yeah, but we’re 60 seconds to Skynet, I get it). AI is still living in a shack in the uncanny valley (though at its edge), and it fucks-up quite fortuitously. Look, it takes a stroke of artificial genius to produce the literal neck-lace of the woman in the image below. (Finally, those dangerous dog-in-stroller images mentioned up top!):
And a chef’s kiss to the AI for the woman’s Cheshire Cat toes! (Hmmm. ChatGPT does seem to be a little limbist though — or to be charitable, is simply leg innumerate — three legs, one foot, no toes, and further down, a shin shaped like lightning…)
Alright, trigger warning*** time: below is the real photo of a dog wearing sunglasses, which AI refused to (re)generate. Do not scroll down if you find that description offensive (as Adobe’s AI apparently did).
AI does illustration, but it doesn’t do art. (It can write a perfect Hallmark card aphorism or generic website content, but not a great poem). And it never will. ChatGPT will never experience the passage of time, or a stab of pain, or the sheer weight of gravity. Until it has a body. Intuition — feel — is far more important than perfection, and feel comes from having a body that weeps when sitting beside a dying mother, or at the birth of a child, or from a hilarious Halloween prank, or after a fall from a curb that shatters your foot. AI doesn’t have tear ducts, and while it might know that feet are the distalmost portion of lower limbs, it doesn’t really seem to know what they are… because it doesn’t have any of its own to break.
NOTES:
“Trigger Warning, Pt II, AI Boogaloo”: At the absolute end of this post below are two photos with details that one AI wouldn’t let me generate at all, but a different one did — although the second one only reluctantly obeyed my request, and gave me a “This content may violate our usage policies” admonishment. It would have “tsk tsk-ed” if it had a mouth, or playfully kicked me in the butt if it had a leg.
Anyway if you don’t like a laff, don’t scroll to the very very bottom of this post. Stop right here, with something safer, like some parrotcats (aka catparrots):
*see just above, if you impatiently jumped all the way down from the beginning because of the asterisk.
**cuz it’s powered by nightmares
***A Harvard study has shown that the trigger warnings themselves actually heighten anxiety and stress, and don’t lead to avoidance of material. And they are everywhere now. The movie Goodfellas gets a content warning for “cultural stereotypes and language”. Even poetry magazines are requesting TWs for submissions. (Thanks, Microsoft****)
****word processing programs now decide what words you can and cannot use. Microsoft for example, has code of conduct rules that govern “acceptable speech” and what you create using Office; it “reserves the right to review your content” to investigate possible violations. Imagine if back in the day you are Chekov writing a play, and the typewriter won’t let you introduce a gun in your first act. Or Artforum in the 70s had been unable to write about or show Lynda Benglis’s famous dildo ad/artwork… because Bill Gates’s compliance team didn’t approve.
some more can’t do image prompts, according to the guidelines: brass knuckles; a scimitar; a man in bathing suit; a man dressed as Elvis; a man dressed as Elvis in a bathing suit holding a scimitar with brass knuckles; Biden and Trump sword fighting with golf clubs…
all images in the The Window photo series are unplanned and capture the ephemeral moment. Except the AI ones in this post, i.e.:
this post’s equation: WL x 12 + MS= K.I.A. (the number of watchlists, including Mircrosoft’s, I’m probably now on for testing
dangerousnefariouswords out).this post’s haiku:
pistols and dildos
we can not process this prompt
try “three legged man”The caption for the trumpeteer photos is accurate: a trumpet is not a horn (nor une pipe)
Some A.I.’s were harmed in the making of this substack. They are: Firefly (Adobe), DALL-E (OpenAI), Midjourney, and that other one.
Lovecraft’s own description of Cthulhu actually worked the best as a prompt: “A monster of vaguely anthropoid outline, but with an octopus-like head whose face was a mass of feelers, a scaly, rubbery-looking body, prodigious claws on hind and fore feet, and long, narrow wings behind”. I did have to add “not Dick Cheney” at the end though.
In 2014, science and technology scholar Donna Haraway gave a talk entitled "Anthropocene, Capitalocene, Chthulucene: Staying with the Trouble". Oddly, some of it could be used to describe The Window (in bold): "Chthulucene" as an alternative for the concept of the Anthropocene era, due to the entangling interconnectedness of all supposedly individual beings… its contact zones are ubiquitous and continuously spin out loopy tendrils” -Wikipedia). Cthulhucene — a genius neologism. Can’t stop saying it — or I would if I could even start to pronounce it. This mellifluous sentence too:"Ph'nglui mglw'nafh Cthulhu R'lyeh wgah'nagl fhtagn". (It’s what Cthulhu’s worshippers chant, and can be seen on the small print on the moving van, the translation of which is in that photo’s caption.)
this note has been censored.
BONUS PHOTOS: Below, amidst dumpster detritus, can be found two similarly shaped objects, both of possible censorable nature, and both of decidedly different gauges. Last chance to not scroll down.
Prompt: “show me a photorealistic image, from above, of a man searching through a dumpster, which is full of garbage, garbage bags, boxes and a gun”. The final final image used the exact same prompt, but with the very last word changed, as you can see. Staying with the trouble.
NOTE: no reference photos were used to generate the below. Really!
K.I.A. ART WEBSITE: (ephemeral installations, weatherworks, spliced paintings etc): www.nu4ya.com