At first glance, thispersondoesnotexist.com may look like nothing more than just a website that showcases random faces — but it’s not as simple as that. The website is proof of where artificial intelligence stands as of now and a sneak peek at what it may be capable of doing in the near future.
Thispersondoesnotexist.com displays an image of a person that doesn’t actually exist but is generated by a Deep Neural Network. Every time one refreshes or reloads the page, one gets a new “fake” person.
Image of a person who doesn’t exist. As real as it looks, the above image was generated by AI.
The website was created by Philip Wang, a software engineer at Uber, and based on a research paper published by NVIDIA to create an endless stream of fake portraits. It works on an underlying AI technology called GAN developed by Ian Goodfellow in the year 2014. GAN, or the Generative Adversarial Network, is a machine learning model that goes beyond predicting to generating new inputs. NVIDIA’s version of GAN called StyleGAN is what powers the website which trains on billions of images of real people’s faces and learns to generate new images of people who doesn’t really exist.
NVIDIA has recently open-sourced its StyleGAN, which now enables AI enthusiasts around the world to lay their hands on a powerful tool that can let their machines imagine and make up new faces that don’t exist.
The dataset that is used to train StyleGAN called Flickr-Faces-HQ Dataset consists of 70,000 PNG images of real human faces at 1024×1024 resolution. The open-sourced project allows the users to either train their own model or use the pre-trained model to build their face generators. However, before getting excited and going ahead to download the huge data set to train your StyleGAN, make sure to have something powerful inside your CPU cabin. Even the most powerful CPUs and GPUs can take weeks to train the model completely.
Why, Though?
This “real” fakeness is a huge technological advancement that can benefit a lot of sectors, especially in the entertainment industry. We might as well see machines drawing and designing characters for animated movies in the future. The current system is made to generate real-like human faces however research is already proving its worth in creating new Anime characters, font styles and graffiti.
Anime character faces generated with StyleGAN
The Dark Side
Apart from the technical achievement, there is a darker and creepier side to it. Sometimes some of the AI-generated faces look too creepy and scary to be real but the real creepier side comes with its applications. Any technological advancement is not limited to good use. As the GAN evolves, concepts like deepfakes will start to become a threat to the public. Creating fully virtual yet realistic worlds will become possible as well as creating fake video content involving real people, things which will question the ability to trust not to mention the influence it will have on politics.
Overall thispersondoesnotexist.com is proof of the capabilities of Artificial Intelligence and a clear indication of technological advancement.
Into The Future
The human mind is limitless and the capability to think and imagine and create things is a wonderful ability that our brain possesses. The fact that a machine can create human-like faces is as creepy as it is amazing. The warnings and possibly greater threats that it poses cannot be and should not be ignored, considering the fact that the technologies we possess right now are nowhere close to maturity and once we get there, there won’t be a going back.
