US graphics card maker NVIDIA NVDA 614,47, -3,46% trained an AI to generated hundreds of human faces, bringing the line between computer reality and real life one step closer.
To do this, the company’s engineers fed the AI hundreds of thousands of images. Then they began training it to generate images of its own, starting from very low resolution and adding “new layers that model increasingly fine details as training progresses”, the company explained in a newly-published paper.
The model including training two parallel networks, whereby one was generating the images and the second one, called “discriminator”, was acting like a critic.
By growing the two progressively, they were able to speed the training up and stabilize the results, producing higher and higher quality images.
The two-network model, known as “generative adversarial network” (GAN), are widely used across a variety of industries. They can help produce realistic 3D models of different objects, interior design, maps, etc. Facebook FB 302,24, -1,29% is using adversarial training to analyze user data in a totally different way. The company’s AI efforts are intended to construct patterns capable of anticipating human needs, interests and behavior without humans necessarily being aware of them. Furthermore, they will be able to interpret those patterns and predict future events based on the input they receive during the training.
It is up to everyone’s imagination to construe the scale of power enterprises will acquire once AI receives and processes sufficient data to analyze, store and predict behavioral patterns. It is both fascinating and scary.
Thanks to today’s technology however, many of the latest advancements are available equality to large corporations and two individual entrepreneurs.
You can check out NVIDIA’s project on GitHub, available for free for non-commercial use.
You can watch an hour long of fictional celebrity faces here. And if you feel uncomfortable in a way, don’t panic. We certainly were a little spooked out, to say the least. Some of those faces looked creepily like real humans.