I just came across a highly interesting paper published by NVidia, with the presumably off-putting title Progressive Growing of GANs for Improved Quality, Stability, and Variation. Bear with me, though, because the content is quite fascinating. What NVidia has managed is to randomly generate, among others, realistic human faces.
Have a look at this:
Tastes differ, but the face on the upper left would not be out of place on the cover of a glossy magazine. To conform to contemporary tastes of the political left, though, we’d have to fatten her up a little bit, add a nose piercing and give her pink hair, but apart from that, I’d say the result is a resounding success. The pictures generated are of a size of 1024 x 1024 pixels, which is probably less than the resolution of your smartphone. Yet, from here on we just need to scale this by throwing more computational power at the problem.
In case it has not been clear: those pictures don’t show real humans. They were completely computer-generated, and it took just a few dozen hours to do so. Now think a bit bigger! Right now we can generate 2D model faces with the computer. The next step would be 2D human bodies. A bigger step are randomly generated 3D models. Heck, in ten years we may have randomly generated porn. You tell the machine what you like and it spits out a never-ending library of virtual whores. What a time to be alive that would be!
We are not quite there yet. However, it is safe to assume that models will start to get a bit nervous, and not because they wonder whether the photographer was happy with their blowjob skills. In advertising, we don’t necessarily need a well-known pretty face. No, instead any pretty face would normally do. In the very likely case that the computer dreams up an especially attractive woman, she would likely become a virtual star anyway, and if she gets too expensive to license, we’ll just generate 256 new ones.
Imagine that, instead of hiring a capricious model who needs to take coke breaks and can only take so many dicks a day, the art director of a magazine could simply let a data center randomly generate a few dozen pretty faces and pick one in the end. They don’t even have to bother with photoshopping the images afterwards, so cost-savings would snowball through every stage of the production. One barrier to widespread industry adoption would be that computers don’t suck dick, which surely is a competitive advantage models have over those computer-generated images. However, I wouldn’t be surprised if some Japanese researchers are already working overtime to produce a blowjob robot. We already have VR. Now we just need to get a few hardware extensions.