Here’s a fun fact: everyone you’ve ever dreamt of is a real person from your memory. Even strangers, who you probably don’t explicitly remember but have stored images of their profiles in your subconscious. It’s actually humanly impossible to think up of a face completely imagined without having seen that person before.
Several days ago, however, a video showcasing the abilities of a new software called Generative Adversarial Network (GAN) reveals how AI has been utilised to create a series of random “celebrity” faces of people who don’t really exist in real life (at least not yet).
Trippy, huh?
While attempting to create realistic images of celebrities isn’t a new technology, GAN has produced the most successful method of doing so thus far. The software detects however many images of celebrities at one go, filtering through other fake images and randomly creates facial features based on existing ones, producing realistic images of humans who aren’t real.
Some of GAN’s shortcomings include only developing images of a maximum size of 1024×1024 pixels and occasionally distributing facial features unevenly, resulting in wonky eyes or droopy earlobes. Purely by chance some images look too similar to real celebrities, rendering it ineffective.
While this can foreseeably boost the entertainment industry regarding by more quickly synthesised fantasy universes in games or movies, some wonder if such photorealistic AI evolution could become complicit in propaganda and misrepresentation. Speaking of which, just a few days ago, Saudi Arabia granted an AI robot citizenship, a decision which has left many questioning the boundaries of AI in our future lives.
Regardless, it’s certainly an impressive step in technology. In fact, the software goes beyond generating fake human faces. Depending on its use, it can create any fake object — a car, a building, an animal — just by sifting through similar images and piecing random traits till it becomes something of its own.
Header image: The Verge
