I guess there is an amazing technical reason for why scientists at Nvidia–the manufacturer of mobile and high-end graphics processors–decided to train two neural networks to create synthetic celebrities but, looking at the results in this one-hour YouTube video, perhaps they should have used that time in doing something good for humanity. Like crochet. Or giant rubber band balls. Do we really need celebrities who simultaneously remind you of Ben Affleck and Bill Cosby? Check out the video:
The system uses two neural networks, one to generate faces from low-resolution photos of actual famous people and the other one to evaluate that result. As the neural networks continue this creation and evaluation process, the image resolution increases, “adding new layers that deal with higher resolution details as the training progresses.” According to the Nvidia research paper, the system “greatly stabilizes the training and allows [it] to produce images of unprecedented quality.”
The resulting images are surprisingly realistic, adding another example to the many cases in which artificial intelligence is blurring the frontier between fantasy and reality. The capacity of machines to learn about our physical reality and fully understand it seem to increase exponentially with each research announcement. There is no doubt in my mind that their ability to generate new realities will match their cognitive power and, eventually, they will destroy the divide between synthetic and real dimensions.
I can see this AI research being used to create believable people in synthetic 3D environments in a completely automated way. These procedural humans–whose characteristics are generated randomly from various data sets under specific rules–would have a good use in future games and virtual reality worlds. I can imagine Nvidia incorporating this technology into future chips to produce believable synthetic characters in real time. Until then, we will have to keep entertaining ourselves with living fictional celebrities, like Meth Damon.