Artificial Intelligence is, I think, more artificial than intelligent, sometimes. And that’s why it’s so fascinating. AI networks that use massive datasets of images to work with one another (or against one another, in the case of generative adversarial networks (GAN), somehow) have been shown to be capable of creating alarmingly-natural-looking human faces, for example. I’ve played with and written about (easily acessible) AI generative systems and cars before, but there’s something about the dynamically-morphing quality of this video that just kind of pushes it to another, uneasy level. Let’s look at some of the non-existent cars this algorithm created.
Oh, first, you should see the video! This was sent to me by my weirdo friend T.Mike, who found it on this Reddit post.
The original poster, Madeofants, describes the video as being an “Audio Reactive Interpolation using ‘cars’ StyleGan2” which means that the images change based on cues or patterns from the audio, and it’s using StyleGAN2, a generative adversarial network developed by Nvidia, and it was fed a dataset with pictures of cars.