The faces pictured above may look real, but they’ve all been generated by a website designed to demonstrate the creepy applications of artificial intelligence.
The website, thispersondoesnotexist.com, generates a synthetic face every two seconds. It can produce images of men, women, and even children, all with lifelike authenticity. But none of the people are real.
The site’s creator, software engineer Phillip Wang, told PCMag his goal was to raise awareness about the AI technology and its implications on society.
“Once they see that they (the faces) are truly generated, I think that shifts a lot of people’s worldviews about what is possible, which is the intention of the site,” he said in an email. “People start speculating about the synthesized reality of the future.”
Indeed, the website has both impressed and freaked out the public. “Imagine you are having a nice online chat with someone, with pictures and videos to back up the claim that they are real… But in fact, all computer generated,” wrote one Reddit user. “I hope my kid doesn’t accidentally fall in love and marry an AI.”
“This is a gold mine for fake social media accounts,” wrote another. “What’s a positive use case? I can’t help but lean into the negative.”
Wang built his website using technology from Nvidia, which has been researching face-generating AI. The company showed the technology back in 2017 by demonstrating how computer software can create synthetic human faces constructed out of celebrity images.
Nvidia’s AI uses a computing approach called Generative Adversarial Networks or GANs. Essentially, the approach pits two computing systems against each other. In this case, one system is devoted to generating the synthetic faces by modeling them on existing photos of people. The other acts as a critic and filters the low-quality images out from the good.
Since then, Nvidia has been fine-tuning the system so that it can take different attributes of real-life photos and blend them together to create life-like, but ultimately synthetic faces. The quality is good enough that the synthesized data is “indistinguishable” from the old data, Wang said.
“This has big implications on our future world, especially as it grows increasingly digital,” Wang told PCMag.
Although the technology could benefit computer graphics and video-editing software, it nevertheless has some worried. Other AI systems have proven to be adept at creating lifelike, but fake videos. In response, last year a group of US lawmakers called on US intelligence to evaluate how the country might stop “deepfake” technology from spreading misinformation on the internet.
Still, the technology isn’t flawless. Wang’s own website will occasionally generate some odd-looking and even nightmarish photos of faces that don’t look fully human. But it’s clear the AI-powered tech is already starting to blur the line between what’s real and what’s not.
“An assault on people’s sense of reality may be possible on a large scale,” Wang added. Those who are unaware are most vulnerable. That is why I made the site.”