https://thispersondoesnotexist.com/
So I've been looking into GANs (Generative Adversarial Networks) and how it can be used to generate human-like faces
https://skymind.ai/wiki/generative-adversarial-network-gan
The tl;dr is this;
There's a database of say, human faces
Then there's basically a random image generator
The image generator starts spitting images to a neural network that compares the images it spits out to the database of real human faces, the neural network then reports to the image generator how likely or unlikely that image resembles a human face from 0 (not likely at all) to 100 (very likely)
At first the image generator spits out nothing but random noise. The neural network then begins reporting a lot of 0 percents to the generation software
Over time the image generator begins figuring out how to gain higher and higher scores with the neural network. After thousands of generated images the image generator gets closer and closer to 100 percent and the faces now look more human. What was once just a random image generator has turned into something that can literally create photo-realistic faces out of thin air.
My concept for a GAN run CP bot is to reverse the process.
Build a database of CP (controversial, might need some approval body) the train the neural network based on this database like normal. BUT instead of having an image generator, we simply feed it images other people have reported as CP (say, on an imageboard) and then the neural network compares the image, if its greater then 90 percent likely to be CP, then the bot bans the user. If its less then that, the AI flags a human moderator to check on it to see if its a false report and take appropriate action.
Basically use GANs to take some burden off of human moderators, very useful for websites like 8chan itself, without having to have a continuous growing internal CP database which would be more impractical and perhaps more controversial
What do you guys think of my idea?