But it does not exist in real life. She is a person created on a website – appropriately titled thispersondonesnotexist – by artificial intelligence. If you reload the page, it will be replaced by another person who is just as attractive, but equally unreal.
Launched earlier this month by software engineer Philip Wang as a personal project, the site uses the newly released artificial intelligence system. developed by researchers at the Nvidia computer chip manufacturer. Called StyleGAN, AI is able to find some of the most realistic looking faces of non-existent people that the machines have produced so far.
Thispersondoesnotexist is one of several websites that have appeared in recent weeks using StyleGAN to produce images of people, cats, anime characters, and holiday homes that look closer to reality, and in some cases indistinguishable from the average viewer. These sites show how easy it is for people to create fake images that look plausible ̵
1; good or bad. Wang, like many other researchers and enthusiasts, is fascinated by the potential for this type of AI. So much that he created a second site called thiscatdoesnotexist that generates artificial cats.
He's also concerned about how he can be abused
This makes sense as the AI tactic underlying StyleGAN has also been used to create the so-called "deep funky," which are convincing (but fake) video and audio files that claim to show to the real person something they have not done or have said.
These worries are reflected by prominent voices in the industry. Earlier this month, OpenAI, the non-governmental intellectual property research firm, decided not to run an AI system it created, citing fears it was so good to compose a text that could be abused.
But although the images that appear on Wan's website can be used to say to help a scammer to create realistic online characters, he hopes that this will make people more aware of emerging abilities of AI. those who are ignorant of technology are the most vulnerable, "he said. "It's like phishing – if you do not know about it, you can come across it."
The bait (and tells) about fake people
Many people are not quite sure how to feel about such easy access to false faces. But they are interested in them.
Wang, who had previously been a software engineer at Uber, studied AI alone for six months when he launched his site in February – shortly after Nvidia made publicly available StyleGAN. He posted on the AI Facebook group site on February 11th. In the weeks after that, about 8 million people visited it.
"I think for many people out there, they watch this and go" Is this a simulation? Are people really in the computer? "Wang said.
The generator creates a new face every two seconds, says Wang, which you will see when you refresh the page.
"You may think that AI dreams of a new face every two seconds of
The faces that visitors see vary endlessly, with many eye colors, face shapes and skin tones. eye shadow, a handful of sports glasses, and sometimes a person with facial hair appears
They have all sorts of facial expressions, some smile, others swell or look serious. (19659022) As these faces may appear, there are still many details that give away that they are not real people, For example, teeth often do not appear to be older than middle age. they seem a little strange and seem to be in need of brackets, and accessories such as earrings can only appear in one ear, often seeming to have unhealthy skin conditions or serious scars on the face. Clothes can look blurry, have swirls of colors or just some kind, well, weird.
How To Make Faces
In order to generate such images, StyleGAN uses a machine-learning method known as GAN or Generative Racing Network. GANs consist of two neural networks – which are algorithms modeled on neurons in the brain – facing each other to get real images of everything from human faces to impressionistic pictures. One of the neural nets generates images (for example, the woman's face), while the other tries to determine whether that image is a fake or a real face.
Although the AI area covers decades, GANs are only around since 2014, when tactics was invented by Google's researcher Ian Goodfeld. They quickly gained prominence among many researchers as a great advance in this area.
StyleGAN is especially good at identifying different features in images – such as hair, eyes and face shape – which allows people who use it to have more control. over the people with whom it comes. This can also lead to better-looking images.
GAN-made counterfeits can be fun – if you know what you're looking for – and potentially big business. Starter called Tangent, for example, says it uses GANs to change real-life models, so online merchants can quickly (and realistically) customize images in buyers' catalogs across countries instead of using different models or Photoshop. A videogame company could use GANs to help make new symbols or repeat existing ones. This was not Airbnb
Christopher Schmidt, a Google software engineer, was one of the millions who saw Wang's site soon after its launch. He noticed that Nvidia researchers also trained StyleGAN to come up with realistic images of the bedrooms and had the idea of building their own site to combine the room images called by AI with text generated by AI. The text generator he used was trained by Airbnb.
Nvidia declined to comment on this story. A spokeswoman said this was due to the fact that the company's StyleGAN research is currently under peer review.
Looking and sounding like strange, confused versions of holiday ads, Schmidt's results from AI are far less plausible than the Wang site's faces. (One of them included an image of Dali's dining table, the other included the line, "Minutes from Woods and having a garden or summer or a relaxing glow of all electrical products.")
But also Schmidt hopes sites like his people to doubt what they see online.