The Disturbing Art of A.I.

(Nexpo theme music) (picture hums) (delicious jazz music) (delicious distorting jazz music) (images flashing) (mouse clicks) - It's Tuesday night. (spooky music) Wildfires ravage California's west coast. CD Projekt Red announces DLC. And in the far reaches of cyberspace, a musical artist named Supercomposite makes a discovery. (thunder) (spooky music) Experiment with AI-powered art, made possible by feeding text prompts into machine learning systems like Stable Diffusion, DALL-E and Midjourney, and like everything, at first everything seems fine . Animals, landscapes, trees, the human form, the potential of AI-powered art is infinite, so it decides to push the limits of what this system can do and brings the following to the table. (keyboard click) - We have known each other for many years, but this is the first time you have come to ask me for advice. - Marlon Brando's last name.

A warning that will surely test the absolute limits of the capacity of artificial intelligence. Interestingly, you enter a modifier, -1, also known as negative prompt weight, which tells the AI ​​that they are looking for an image that is conceptually as far away from the prompt as possible. To be honest, it's an interesting thought experiment because it makes you wonder how exactly the AI ​​sees the world and what it considers the opposite of everyday things. I mean, what do we consider the opposite of everyday things? If I asked you what the opposite of birds is, what would you say? (tense music) (machine humming) Supercomposite sends his message and waits. (machine whirring continues) Minutes pass without much concern, until finally - (computer whirring) (haunting music) Innocuous, but interesting, that the opposite of Marlon Brando is perceived as an abstract horizon.

Written on it appears to be two words: Digita and Pntics and in the background there appears to be text on a white page. Curious, he wonders if the inverse of this photo would result in an image of Marlon Brando, so he enters another message: and sends it off into the vast unknown. (ominous music) (atmosphere humming) (ominous music continues) (atmosphere humming) Before them were four images of a very strange older woman with severe rosacea on her cheeks. She has high cheekbones, dark hair, and can be seen in a living room, in what appears to be a hospital, in a bathtub, and against a green wall, with only one word discernible above her: L-O-A-B or Loab.

Over the next few days, Super experiments with creating her, becoming curious as to why and how Loab exists. She eventually employs a method called AI Image Crossbreeding to combine one of Loab's photos with a completely separate one her friend created, however, instead of this combination resulting in Loab's disappearance, or even innocuous photos involving this tunnel , what comes true is something a little more grotesque. Through some sort of emerging statistical accident, something about this woman is adjacent to extremely gory and macabre images in the distribution of global AI knowledge. Since Loab was found to use negative warning weights, her gestalt is made up of a collection of traits that are equally distant from something.

But her combined traits remain a coherent concept for the AI, and almost every offspring image contains a recognizable Loab. In almost every image, Loab is distinct and embodies the physical manifestation of the repulsive and sinister. This AI is somehow creating images that perfectly encapsulate what terrifies us, embroiled in an infinite number of situations such as family portraits, lurking inside an office building, hovering over children like a reincarnation of Jesus Christ, and even waiting after what appears to be a murder by his own hand. But it doesn't end there, since the correlation between Loab and the idea of ​​malformed and mutilated images was just beginning to take root.

The Angel Hall plus Loab also produced art with so much blood that probably very few people want to see them, but here are two. I don't feel comfortable posting the most

disturbing

ones. They are dubious images of dismembered and screaming children. There is something moving to me about these grotesque scenes and the desperation, panic and sadness they convey. Again, these are produced with other images as inputs and no text. They are the result of crossing images of Loab with images of other things. The images that result from the crossing of Loab with other images can, in turn, be crossed with other images.

The AI ​​can hold onto Loab's idea so well that it can persist across generations of this kind of crossbreeding, even without using the original image. From here you continue your journey down the Loab rabbit hole. As mentioned, her persistence in existing within each resulting photo is astonishing, and even when they try to get rid of her, distort her to eliminate her, she remains in immensely uncomfortable scenarios. With a child, clinging to the human form. Locked in time, like a clay statue. Warped by unknown technology. Lurking, watching from the darkness. Trapped inside a Stephen Gammell canvas, painstakingly out of place.

Abandoned like a lifeless child. Trapped in the deep blue. And frozen in eternal pain. (atmosphere humming) I was tearing Loab apart and putting her back together. It is an emerging island in latent space that we do not know how to locate with text queries. But for AI, Loab was an equally strong point of convergence as a verbal concept. The big lesson for me here with Loab is that image cues can essentially be used as a custom parameter to challenge the AI's way of thinking. You can produce original styles and characters that you literally discover. Negative cues can help you find emergent accidents, as does Loab.

This is largely where she ends her thread. She states that she will update you with more photographs of Loab as they are created, and that is a promise that is kept. To this day, they're still adding things, and this thread, in its entirety, serves as an interesting dive down a surreal, strange, dreamlike rabbit hole regarding this undisclosed AI art generator and its strange connection to this specific face. . As we can remember, the original message referred to the inverse of Marlon Brando, and the resulting image looked nothing like what any human would logically expect it to be.

It was an abstract horizon, so how did this come about? (upbeat guitar music) (clips playing) (ominous music) Why are AI creations so

disturbing

? What is it about these images that causes this inexplicable fear? The program I was using to make this slideshow and throughout this video was Stable Diffusion, and in the process, I learned quite a few things about it. When you make your entry too specific, you may find exactly what you're looking for. AI is, in some ways, limited by what you feed it, however, if you remove the details in your message and really allow it to spread its wings and create something for you, well, that's when things get interesting. . (clips fast forward) Anomalies, much like Loab, await, hidden within the confines of something called Latent Space, or the medium in which the AI ​​recreates the characteristics it has received.

You may have seen other examples of this idea before, notably with a Twitter thread from June in which this person discovers what appears to be a demon after entering a completely made-up word. (clip reels) Crungus, they called him. Which is pretty creepy and is even something you can discover yourself. But this system doesn't just excel at disturbing, unpleasant creatures and faces, and the breadth of AI weirdness isn't always as simple as entering a text message that we know will give us something scary. In fact, one of my favorite examples of artificial strangeness goes back years. So, let's pause. (The movie plays) And let's go back. (film whirring rapidly) (film thuds) (ominous music) In 2019, an anonymous programmer calling herself none other than "The Girl Who Talks to AI" set out to create a simulation of the human brain using a well-known system. as Generative.

Adversarial Network, or GAN for short. Basically, GANs are trained using machine learning to absorb millions of existing stimuli in order to recreate an idea. In this case, the system was tasked with creating a face and, after sifting through millions of existing photos, it spit out this one. The programmer, however, had a curious idea. Since the AI's entire worldview is limited by the sum of its inputs, what would happen if it slowly forgot what it was trained to do? And so, with the AI ​​neurons displayed on the right, it turns them off one by one, until this face distorts into something completely unrecognizable. (tense music) (mouse click) The parallels between synthetic deconstruction and the afflictions of the human mind are clear, and this project serves as a curious mirror into the realm of neuroscience and how we, as individuals, perceive the life we around us.

The way that face slowly distorts into a faded memory, leaving the remaining neurons struggling to fill the void, leaves us with an end result that is disturbing and, to this day, remains one of the curiosities. of most fascinating AIs I have ever encountered. never seen (tense music fades) (somber music) Loab is a haunting case study. Someone who has successfully established himself in the online sphere, and although the reason behind his conception and why this AI somehow latched onto his strangely specific facial features and linked them to macabre and disturbing images is currently unclear. The truth is that its persistence has left us with an interesting revelation.

We know that AI knows how to create subjectively beautiful synthetic art through mere text prompts, generating some of the most striking examples of the great potential of modern technology. That said, who's to say you can't excel at the opposite? Relatively, the adoption and potential of synthetic art is just beginning. And with the arrival of Loab scaring many of us, what else is hidden within the vast abyss of AI and how will this affect us when we find them? (somber music) (footsteps echoing) (hello, old friend) (atmosphere humming) (parting tone)