Is AI generated content (images/videos) REAL? Is there reason for ethical panic?

[Sethioz]

it depends how you define "real". i have seen lot of people already rolling their eyes and panicking about AI generated images, but in my opinion they are close-minded idiots.
still, let's get more into detail and explain things so that everyone understand what is going on.
technically everything in universe is real. images exist, they have particles. computer data is all real, because it's particles moving inside of your computer, but it happens on sub-atomic levels and most people don't understand the concept, so they say "it's not real".
it's as real as anything else, it's just different. If you are looking at a photo on computer screen, it's real, but different. that photo is stored as digital data, which consists of particles, which makes it as real as anything else.
think of it as taking object apart on sub-atomic level and putting it inside of tiny box, it's still same object, but in different form.
if you had nano-tech (nano crafters) that are capable of doing that, then you could store your entire car inside small box, however that box would still have weight of a car, so you couldn't just put it in your pocket.

anyway to better understand how AI generation works, i have to explain different things.

How Artificial Intelligence works:
first of, AI generated content is only as good as it's model and model is only as good as it's training data.
AI does not store images or other content the way you think, AI is not wikipedia that is just collection of information, AI works similar to human brain.
think of AI generated content as artist, if you ask a good artist to draw realistic image of human without directly looking at any person, then good artists can draw very realistic photos, however they can heavily rely on someone who's face and body they remember the best, while AI uses more general style and doesn't "copy" things off actual people.

Think of AI as blind person who has been teached to draw real humans. blind person can't really see, but if you guide blind person, then he/she can draw very realistic images. AI generated images/videos are kind of like imitations that don't include any real bits or pieces, but since topic is "real human" then there aren't that many ways you can put human together, which is where misconception is coming from -- lot of people think that AI copies and merges real photos, but that's not the case at all.

AI generated content is 100% fiction and does not include anything from real photos at all, any resemblance to actual person is purely co-incidence.


How AI training works / How AI models are created:
AI's that can generate content from scratch are still trained on real data, which is also where misconception is coming from. For example "stable diffusion" image generator with "cyberrealistic" model can generate photo-realistic images of humans, that's because that model is trained on millions of real photos and lot of people think that AI will save those photos and then use them to generate images, but this is NOT HOW IT WORKS. AI models DO NOT save any images/videos, they only use those images as guides on how a human looks like, but they are fully generated by AI without any kind of copy/paste/merge, because as i said, AI does NOT save any images/videos during its training.

AI training is where creator of AI and/or AI model, shows AI lot of images/videos of whatever they want that AI to learn, then AI "looks" at this data and memorizes how things look like in general. for example if you show it 1 million random photos of monkeys/apes and say "these are monkeys and apes" and then you show AI 1 million photos of humans and say "these are humans", then AI will be able to make clear difference between apes/monkey and humans, but it won't be able to tell difference between monkey/ape species nor is it able to tell difference between male/female nor human ages.

This is where you need to add more training data with specifics, saying "these are females" and "these are males". For better accuracy, you have to also put people into age groups and explain to AI during training, which age is which.

That's why AI training takes significantly longer than generating content, AI training can take months even on powerful computers, which is why most AI models are made for specific tasks. AI models usually take 5-15gb space each. as explained before, that data does NOT contain any real images/videos. if you show human faces to AI, it won't store any of them, it just makes a "guide" on how to draw human face.


BUT is AI generated content REAL???
again, it depends how you define "real". everything is real, but if you ask "is AI generated human real?" then answer is NO!
AI generated content is always fully fictional and nothing is copied from training data.
it can confuse average human brain, because it looks photo-realistic, but that's the whole idea! idea is to make AI generate images that look photo-realistic, but are not taken from real life.
With fine-tuning, you can make AI generate images that are indistinguishable from real photos, but they're still just a fiction and NOT a copy. Even if AI generated images look exactly like someone you know, it's still not real human, it's purely co-incidence.

If AI was trained on limited data, then it can make images look a lot like same person. for example when making CGI for movies, they can train certain AI model only on one person, which means this AI model can only draw that one specific person, but it's still not same as copy, it won't change the fact that whatever AI generates, is fully synthetic and not a copy.
It's kind of like lab-grown meat, is it real meat or not? answer is yes/no. it tastes like meat, but it doesn't come from actual animal.
or same for synthetic cannabis, is it real or not? laws about cannabis are getting confusing aswell, because there's no law for synthetic cannabis, altho lot of countries, out of panic, trying to ban innovation and evolution.

Only difference between artist and AI, is that AI is capable of extreme precision and speed.

WHY is CGI / MOVIES OK, but AI images are NOT OK?:
Here's something that will seriously make all of those screamers think.
WHY is it ok for movie creators to use CGI and other "movie magic" to make content where visibly underage girls appear nude, babies and children getting gored, but AI generated content within same topic is not ok?

Here are some examples:
1. in alien ressurrection aka alien 4 there is scene of completely nude girl who appears to be around age of 10. Read more HERE (including photos) = viewtopic.php?t=3232

2. in Terrifier 3 (i think it was 3rd part? maybe 2nd?) you can see children getting axe murdered, there's lot of gore!
3. In Feast 2 - https://www.imdb.com/title/tt1104835 - you can see a baby thrown into the air and it splatters all over the ground!

WHY is that OK? but AI generated images to replicate those scenes are NOT OK?

all those who scream nonsense like "animal abuse", "child abuse" or "child porn" over AI generated content, are nothing more than close-minded idiots. there are NO real animals or humans involved, therefore you can't say those things, because no real biological beings were used, therefore it can't be abuse nor porn. It's exactly same as with movies, if they use a doll or CGI, then it's fair game.

Why is it that when something goes according to plan, like in a movie you see a nude person who visually looks under 18 or you see kids being murdered with axe, then nobody cares, people watch and enjoy, but when someone uses AI image generator to make a nude of girl who looks under 18, then everybody freaks out loses their minds?

Saying that AI generated content should be restricted, is as stupid as saying that movies should be illegal and human imagination should be limited. without imagination, you'd still be living in caves!