This website uses AI to generate realistic faces and I honestly hate it
No deepfakes here, just fakes.
Well, this is creepy. Ever needed a fake persona for social media? No? Me neither but if you did, this website would be what you’d use. ThisPersonDoesNotExist.com serves as a chilling reminder of the power of AI when it’s trained correctly.
Using research that Nvidia released last year, Phillip Wang, a software engineer at Uber, created the site, which creates a new fake portrait every time the page is refreshed. The AI powering it was trained on a huge dataset of images of real faces, then fed through a generative adversarial network (GAN), to generate new faces.
Here’s how it works
Think of the GAN as the relationship between a writer and their editor, except in this case it uses images instead of words. Two trained neural networks work in tandem to create the final images.
The first, the generator, creates what it thinks is a face from the trained dataset, then sends it to the second network, the discriminator, which checks that generated image against the trained dataset to determine if it’s accurate. Just like a finished manuscript in the hands of an editor, the GAN won’t output the final image if it’s not satisfied with the outcome.
You can see a trained GAN in action in the above video, showing just how powerful this type of network is in creating faces of people that were never born. What hope do we have protecting ourselves against fake images now that a computer can finally create human faces that don’t fall into the uncanny valley?
Where did this all come from?
The original AI framework underpinning all this weirdness was created by Ian Goodfellow, a research scientist at Google. Like many sparks of genius, the idea came to him while in a pub. This type of algorithm could be turned to any source, from anime to fonts to graffiti, and potentially other uses like drug research.
This is just the beginning of this type of algorithm. In the future, the technology will get so advanced that we won’t be able to tell it apart from the real thing anymore. The future needs to find a way to discern the fakes or nothing we see online can be trusted.
- Ring’s Neighborhoods Portal gives police access to loads of recorded footage
- App pirates are gaming Apple’s Developer program to distribute hacked versions of apps
- Kiwi for Gmail is a thing that exists that may assuage your email woes
- Twitter is testing a new cam feature that pretty much looks like everyone else’s
- For some reason, MacBook Pros with Touch Bars can still skip unskippable YouTube ads