Credit: Google News
Last weekend during a few minutes of idle time, I once again found myself flipping through online photos of strangers.
The first was a dark-haired girl around 10 years old. Her gaze was cast downward and her lips formed a slight smile, as if she were softly petting a sleeping kitten.
The next was a woman. She was young — maybe in her late teens. Her strawberry blonde hair was tied in a loose ponytail, a few strands framing her face. She smiled at the camera, her eyes an arresting shade of blue-green.
Then there was a trim, brown-eyed man wearing a business suit. He looked to be in his 40s — there were flecks of gray in his dark brown hair and the beginnings of crow’s feet at his temples. He smiled broadly, bearing two rows of off-white teeth.
There are more where these came from in the seemingly endless repository of faces. They look just like the people I see every day in my interactions, on social media, on the street or in the grocery store. Their expressions are familiar, their features imperfect, their authenticity captivating. They look like anyone else: ordinary people with dreams and families and lives.
But they’re not. Because they don’t exist.
Visit the website ThisPersonDoesNotExist.com, and you’ll find a project created by Uber software engineer Phillip Wang, which utilizes artificial intelligence research to create the fake portraits from a massive data set of real images.
In a February Facebook post, Wang explains that he made the site using StyleGAN, an open source generative adversarial network (GAN) developed by researchers at the graphics chip company Nvidia, which has been pre-trained to amalgamate human faces. Every time you refresh the page, the network generates a new image from scratch.
The faces of the dark-haired girl and the blue-eyed teen and the man with the salt-and-pepper hair were made just for me. No one will ever see them again.
This technology is not unlike our subconscious mind’s ability to conjure elaborate fictions in dreams. We sometimes dream of people and places which seem intensely familiar — my father’s brother, Uncle Rick, will be driving me to our favorite thrift store in nearby Dangleston, for example. But then I wake up and realize: 1. I have never even heard of a place called Dangleston and, in fact, no such city exists; 2. Though I seemed to know my way around the small-town storefront of oddities, I have never actually been there; 3. The special bond I shared with this kindly man during REM sleep could not be real because my father is an only child.
What StyleGAN demonstrates is that, with the aid of machine learning, artificial intelligence is capable of legitimate creativity with compelling results. And while we may have worlds rattling around in our subconscious, our capacity to remember with clarity, achieve fully-formed results on a first try, and produce at scale is decidedly limited.
Wallace Stevens, the American poet, said, “The imagination is man’s power over nature.”
If this is true, the question must be asked: Is superior imagination technology’s power over man?
Aleshia Howell is a freelance writer and digital entrepreneur. She writes about technology.
Credit: Google News