But it does not exist in real life. It is a face created on a website based on intellectual aptitude. If you reload the page, it will have another face, which is equally credible but real.
Phillip Wang engineer software was deployed as a personal project a month before, this site uses a new AI system developed by Nvidia computer chip computers. Style Style, AI is one of the most realistic real-world faces of people who have not yet developed machines.
Thispersondoesnotexist is one of the websites of the latest websites created. Thus, with StyleGAN, people, cats, anime characters and houses of real-life houses, and especially the viewers become invisible in some cases. These sites show how easy it is to make people create false images that look like real-life, better or worse.
Wang, like many AI researchers and enthusiasts, is fascinated by these AI types. However, he created a second site called thiscatdoesnotexist that produces faux felines. But Likewise, it is worried about how bad it can be used.
This makes sense to create a deep-fake AI based style style that can be used to create a real person who does or does not say something that is "conventional" (but false).
There are voices that concern the industry. Earlier this month, the non-profit AI research firm decided that it did not want to release an AI system created by OpenAI, because it was afraid of completing the misused text.
But images that can be used in Wang sites can be used, for example, to help create a scammer in real people, hoping people will learn to create new AI abilities.
"I think those technologies are not the safest," he said. "It's kind of like phishing: if you do not know about it, you can fall."
To listen to land (and tells it)
Many people are not sure about how to get fake faces. But they are interested in seeing them.
Wang, a Uber software engineer, analyzed AI himself for six months in February when he made his website on February – Nvidia, shortly after its release by StyleGAN. He published on February 11th the AI Facebook group website. In weeks, they visited about 8 million people.
"I think there are a lot of people, they look at this and go" Wow, The Matrix! Is this simulation? Are people really computer? "Wang said.
The generator creates a new face every two seconds, said Wang, when you refresh the page.
"In AI, AI feels a new face every two seconds on the server and in the world," he said.
Visitor faces vary infinitely, with a lot of different shades of eye, face, and skin. Some wear lipstick or eyeshadow; A handful of sports goggles. Sometimes a facial hair boy appears; He even looked at his nose.
They have all kinds of facial expressions. Some smiles, others pout or look serious. The youngest faces are small babysitters, but there is no elderly person.
As realistic As these faces may appear, there are still not many details that give real people. For example, teeth are often strange and have the keys as they are, and they are like accessories like earrings. Often, a person may have a different skin condition or facial faces. Clothing can look blurry, swirls in colors, or just kind, well, strangely.
How to make faces
To create these images, StyleGAN uses a machine learning method known as a generic GAN or a counter-generative model. The GAN consists of two neural networks; They are algorithms based on brain neurons, looking at each other, images of human faces and impressionistic paintings to look for images. One of the neuronal networks creates images (such as a woman's face), while the other tries to determine whether this image is a false or real face.
Although the AI has decades, GANs have only been in existence since 2014, since Ian Goodfellow invented the invention of scientist research. Many researchers have been very important in the area.
StyleGAN is particularly good for identifying the features inside the images, such as hair, eyes and faces, which allow people to have more control over their faces. It can also create better images.
A GAN-produced fakery can be fun, you know what you are looking for – and it can be great business. A Tangent-like launcher, for example, asserts that GANs change real-life patterns, so online merchants can quickly (and realistic) buy catalog images from buyers from different countries using different models or Photoshop. The video game company can use GANs to face new characters or with existing iterations.
This is not an Airbnb
Christopher Schmidt, a Google software engineer, watched Wang's website from millions of people soon after it was launched. Nvidia researchers have suggested that StyleGAN's style is realistic and has the idea of building a site, that is to say, it does not exist, combined with a combination of ersat room combined with AI's text with AI. The creation of text used was in the Airbnb charts.
Nvidia has commented on this story. The spokeswoman said the company's StyleGAN investigations are conducting peer reviews.
Looks like terrific and confusing versions of vacation rentals, and look at Schmidt's AI results are less credible than the faces of Wang's site. (The image of a dining table in Dali-esque was introduced, another came in line: "The Woods Minutes, and in the garden or in the summer, or the relaxing shine of all electric products").
However, Schmidt, hopefully like its sites, is questioning what they see online.
"Perhaps we should all think about the second half before thinking about something real," he said.