Things get psychedelic. These are actual illustrations from a paper in Frontiers in Cell and Developmental Biology, Cellular functions of spermatogonial stem cells in relation to JAK/STAT signaling pathway. They had Midjourney do the illustrations, and they are spectacular! And confusing and uninformative.
Look at the labels! AIs are terrible at reproducing text in images, and these make no sense…and most of the diagrams are random piles of throbbing circles. What do they mean? I don’t know.
But here’s my very favorite image. What have they done to that poor rat?
The labels…they do nothing! The text of the paper makes sense and is a reasonable discussion of the role of the JAK/STAT pathway in spermatogonial differentiation, but then your eyes wander over to those bizarre illustrations and you get totally discombobulated.
I should print that last picture out in color, and place copies at floor level around my house as a rodent repellent. Except my cat already has a problem with frequent puking.
drksky says
I find it hilarious that it felt it needed to put in the label “Rat”.
PZ Myers says
And shouldn’t the “dissliced” and “testtomcells” be obvious?
tedw says
An excellent example of why arts education is important. Also a consequence of the declining number of medical illustration programs in the country. Currently there are only 3 in the US and 1 in Canada. (The one at my school trained the medical illustrator that made the gray and red depiction of the covid virus that went, um, viral in the early days of the pandemic.)
Brony, Social Justice Cenobite says
Just wait. These and the other AIs will be trading data with the ones using it for porn soon. This is how skynet happens.
(There are very serious issues related to AI and porn though).
robro says
Brony @ #4 — Not totally clear what you’re saying, but since porn has been on the cutting edge of technology for the last 40 years (at least), I’m guessing they were using AI long before it became semi-mainstream. It may be Don Norman who says that porn invented the internet.
Alan G. Humphrey says
Must be some amazing rat porn…
… probably AIed that, too.
Reginald Selkirk says
Midjourney knows nothing about human anatomy, it only knows about pictures of bodies, which is why all the extra digits. Likewise, it knows nothing about language. It only know about pictures of text.
raven says
The rat picture doesn’t seem quite right.
It shows the rat with 4 front toes and 4 rear toes.
It gets the back feet wrong.
Jaws says
@8, it obviously traded in one toe on each foot for, umm, enhancement of other attributes of the kind frequently found in one’s spam folder (gender and orientation of recipient irrelevant).
Or maybe the rat just wanted to appear in a David Lee Roth video and had two toes crushed by roadies in the setup?
bcw bcw says
You could use the rat illustration to redo ‘the Matrix” but using rat sperm instead of human brain energy or whatever the premise of that movie claims to be.
Being a movie starring rats, there could lots of squeekquels. <runs>
John Morales says
Just to remind everyone: Midjourney is the equivalent to the predecessor of the Model T.
xohjoh2n says
@11:
Awesome! That means it’s successor is going to be the gothest image generator ever!
(I got a pretty good result once from the prompt “robert smith in the style of titian”, but unfortunately the resolution limit makes it impractical to get a blown up print to stick on the wall.)
chrislawson says
The journal has already put up a notice of concern about this paper. But how did it get past editorial in the first place?
numerobis says
chrislawson: far as I can tell, the cheque cleared so that was the editorial review required.
chrislawson says
In terms of forensic analysis, those illustrations are amazing. Midjourney does not understand language. It was fed on scientific illustrations and then generated new labels based on the shape of the original labels rather than the meaning and as a result it invents new letters never before seen in any alphabet. See figure 2 from the paper for several examples including a ligated u-n conglomeration and some eldritch horror of a letter that looks like some sort of cross between the Greek chi and epsilon.
Figure 2 is a superb example of recreating visual patterns without understanding even the basics of what they mean. It is meant to be an intracellular biochemical flowchart as seen in countless illustrations in the scientific literature. But the same element (JAK) appears multiple times, some of the arrows are just lines or cul-de-sacs. In one case, an arrowhead points to nothing but a side bar grows out of the arrowhead linking JAK…to JAK.
chrislawson says
@14 — very likely
chrislawson says
Perhaps the authors should reconsider their ‘in the style of Cronenberg’ prompt.
John Morales says
chrislawson @17, :)
Oh, were it ‘in the style of H.R. Giger’!
StevoR says
(looks at second image here) What the.. ?!
That’s one messed up image!
@1.drksky : “I find it hilarious that it felt it needed to put in the label “Rat”.
Well, given what they done it it maybe not..
OTOH, it could be quite a feww similar-ish species in the rodentae family. Or maybe even a marsupial like a some species of dunnart ( https://en.wikipedia.org/wiki/Dunnart) bandicoot (https://en.wikipedia.org/wiki/Bandicoot ) or antechinus.. (https://en.wikipedia.org/wiki/Antechinus) Convergent evolution at work I presume. Adding the Linnean latin species name in brackets after the rat part would’ve been nice – & the least they could’ve done after the other awful things that AI did to it!
Marcus Ranum says
xohjoh2n@#12:
(I got a pretty good result once from the prompt “robert smith in the style of titian”, but unfortunately the resolution limit makes it impractical to get a blown up print to stick on the wall.)
The latest version of midjourney has a some pretty cool pan/zoom features for images. You can also use image2image AI upscaling in other tools, to enlarge existing images. You might enjoy going back to the well on that prompt.
Since I happen to have A1111 running in another window I just asked stable diffusion for a version of my own. Very good. Its working on a version of Smith by HR Giger right now. (it also gave me a doubletopped version of Smith with an extra layer of eyes like a wolf spider’s)
John Harshman says
It’s just been retracted, because “The article does not meet the standards of editorial and scientific rigor for Frontiers in Cell and Development Biology”. You think? Somebody ought to be fired over that, or at least never asked to review again, assuming this is not a predatory journal.
Reginald Selkirk says
People who bought this item may also enjoy:
Penn State Professor Previously Charged With Bestiality Faces New Charges Including Masturbating On State Vehicles
Akira MacKenzie says
“Rat”
Oh THANK YOU! I WOULD NEVER HAVE GUESSED!!!
muttpupdad says
But is it Rattus rattus or Rattus cartoonus?
felixmagister says
I mean, specifying “rat” as opposed to “abomination from the pages of HP Lovecraft” isn’t entirely a waste of ink.
Ridana says
8) @raven:
You say that like that’s the only thing wrong with that illustration. :D
microraptor says
John Morales @18: Not phallic enough for Giger.
John Morales says
microraptor, :)
I always thought a collaboration between Giger and Philip Jose Farmer would’ve been amazing.
Hemidactylus says
I’m sorry. This perhaps conveys way more about my depraved mind than I should allow but that rat. It kinda looks like a boner that got out of hand and exploded. I thought that every time I scrolled past. I’ll show myself out of this thread now as I am already regretting sharing that image as seared forever into my memory. Sorry you won’t ever unsee it either. Please asteroid. Please!
nomuse says
And now we know where the Voynich Manuscript came from.
outis says
This is getting ree-dunculous, only there’s no one laughing.
It’s always VERY difficult to convey exact information by illustrations/diagrams, and creating scientific illos is an art form by itself. And these clowns actually got this stuff published? With those pizza cells and Mighty Boner Mouse? Everyone concerned should get the sack and go live in a cave forevermore.
It’s not enough that magazines like “Nature” are ringing all alarm bells about paper mills spewing out fake papers all over the place (paywall, sorry):
https://www.nature.com/articles/d41586-021-00733-5
now we gotta see the rise of the demented pictures… if this goes on scientific publishing is going to go under irretrievably.
outis says
@30: on a lighter tone, if you like that sort of stuff:
https://en.wikipedia.org/wiki/Codex_Seraphinianus
with that, the author has been having a laugh at the whole world’s expense since 40+ years, and it looks gorgeous (and expensive).
pbdg says
It’s now been retracted. Being Fronteirs it names the editor (a bull semen expert from India who’s published some nice research) and his two reviewers (one of whom is a colleague). I read that the process from acceptance to publication was 2 weeks (more proof that Fronteirs is MDPI for rich people) Hopefully they are now banned from reviewing/editing any more papers for Frontiers.