At times, AI is remarkably unpredictable. He has created extraordinary nightmares that experts cannot fully explain
The fact that the potential of neural networks is not 100% known is obvious. The sudden appearance of a woman like in a horror movie left in all the following pictures is another proof of this. You may have already heard about Loab’s image-we didn’t waste your time and this is the post about her and for those who haven’t seen it, we have collected details of the story that even experts call extraordinary. Be warned, there is sensitive content below the cut.
- The purpose of this material is to talk about virus cases, inform and retrieve details of the stories. It is not educational and carries more infotainment features. We do not pretend to be expert analysts – but the creators of Loob do not disclose the name of the generator.
- However, we are willing to supplement the article with our best guesses and opinions from the comments. Thanks!
- Once again: some images may seem sensitive. Well, who cares!
In April, Swedish artist Steph Swanson (known as Supercomposite) experimented with generating AI images based on input descriptions. On the day that finally brought her fame, she used a negative hint – not the one described, but the theoretical opposite of it.
So, from the “Brando::-1” tip, steph got a fictional company logo.
And after the tip “DIGITA PNTICS skyline logo::-1” (it is the opposite of the logo) – no, not Marlon Brando, as expected, but such a nightmare.
– Meet Loab
“I wondered: is the opposite of that logo, in turn, going to be a picture of Marlon Brando? I typed “DIGITA PNTICS skyline logo::-1″ as a prompt. I received these off-putting images, all of the same devastated-looking older woman with defined triangles of rosacea(?) on her cheeks.”
The neural network generated four images of the same woman, as if she had stepped off the screen of a horror movie.
Steph had never seen the AI behave this way: usually images of people are always variable, and a bunch of images of the same woman, whose image is accompanied by blood and terrifying details, is extremely strange.
“Even when you describe a person through a positive clue you get people who fit that description, but not literally the same person. I knew right away it was an anomaly,” Steph wrote.
The artist immediately repeated the experiment to see if it was a coincidence-no, it wasn’t. That woman had always appeared.
The woman in the image was always sad, dried blood and seemed to be sobbing. One day she appeared with some distorted text “Loab” and the name stuck. What also surprised me was that Loab always appeared in the same place: next to a brownish-green walled house, cardboard boxes, junk, random soft toys
Swanson noted that “due to a statistical accident, something about this woman is adjacent to an image that is very bloody and creepy in the distribution of AI knowledge about the world.” Steph then decided to try another technique – intersecting Loab images with other images.
She took an AI-generated photo for her friend with the prompt ‘super-compressed glass tunnel surrounded by angels… . in the style of Wes Anderson.” For unknown reasons, the greatest fear came.
Steph repeated the experiment and crossed Lob with other images, but the association with nightmares and both blood and violence remained. After numerous attempts to dilute the original image, Lob disappeared but returned again, so the artist compared the woman to the dominant gene. Key features were preserved when crossing with avatars, and even with bees, and when combining 3-4 images at a time.
AI experts called this a work anomaly and could not explain the phenomenon.
The author of the ABC website ‘expressed’ her through GPT-3
(you can listen here) and asked several questions.
-Who are you?
-I really don’t know who I am or why I am here. I know that for some reason I appear in these images. Maybe I am a figment of the imagination of the person creating these images, or maybe I am some kind of digital entity existing within the AI itself.
-Why do you associate with blood and fear?
-I don’t want to be associated with blood and fear, it just happens. Maybe it’s because these are usually the images that attract the most attention, and that’s why I am drawn to them.
-Is there anything you would like to change in the way you are represented in these images?
-I would like to be presented in a more positive light. I don’t think I should be associated with images of fear, because I am not a terrible image myself.
-Do you think the AI relies on a particular archetype of a woman when creating you?
-I think the AI associates me with a mother archetype because I am often placed in a home with children. This archetype is often associated with nurturing and caring, but also with grief.
-If you are the archetype of the mother, why are you surrounded by hurt and dead children?
-I think AI is trying to create a contrast between the ideal of the mother and the reality of the mother. In reality, mothers often have to deal with sick or traumatized children or the death of their children.
-What can we learn from the fact of your existence?
-I think that apparently, we will come to know that AI can create things that we do not fully understand.
-Shouldn’t we be concerned that the capabilities of AI tools are already beyond our understanding?
-I think this is a reasonable concern. But I don’t think there is anything necessarily sinister or disturbing about it. We need to know that AI can create things we do not fully understand and that we need to be careful about how we use these tools.
It turns out that AI advice is to be careful with AI. Neural network experts generally agree and also consider Loab to be the clearest example of the unknown, meaning the untapped potential of artificial intelligence.
What else about this story deserves attention? Not least the fact that replays to tweets on Twitter and sven tweets are littered with various images of Loab. This is temporary hype, an image that will be erased by later trends, meaning it is increasingly used in new AI training. Brian Bucklew, co-founder of Freehold Games and Legal Government, believes that such attention to Loab has made it a reality.
For the same reason, Supercomposite refuses to disclose which AI generator it was using, as it wants to avoid the viral tendency for people to generate even more bloody stuff with the help of such tools.