Considering every bit of art posted here undergoes compression .. I'd say 100 percent of the art posted has been manipulated by AI. It's amazing how the same panic happens over and over.

That's not really what AI is here, not in this sense. Read below. A compression algorithm is not an AI, it is a mathematical formula with a predictable result. There are "AI" up-scalers, but even that is tenuous at best for what this conversation is revolving around.

I understand what we are referencing for AI I am making the point there is little or no difference between a compression algorithm and a learning algorithm /matrix. They both respond to parameters they are built around. One is directly built by a human directly entering code the second by a learning algorithm that indirectly learns from other humans input.

Yes they both use an algorithm matrix for its solving. But a compression algorithm is static. But I also think the level of complexity is also a factor here. A human wrote the code for compression end to end and can tell you exactly what it does and how it does it. It is pure math from human ingenuity.

An AI was given training data, but we have no idea how it determined it's parameters. We have no knowledge of its function beyond basic structure of what it is doing. We don't know its weights, parameters, etc. All we know is the data we gave it. We do not know how to reverse engineer it to get the same result, even if you fed it the same training data. It is a pure black box. A compression algorithm is not.

Here is the big difference, with everything from Photography to CGI and Digital media, they still required a person to use the tools in a manner that which the tool was under the artists control and the artist could predict the outcome. A tool is completely under our control, and AI generator is not. If I draw a hand, I can tell the exact steps to replicating my process for getting a hand, at the moment, most prompters cannot tell you had to replicate their result. They can tell you the prompts sure, but that is not going to give you any insight into a predictable and deterministic outcome. None of the other tools you mentioned could function without a human controlling it. If given a camera I can tell you how to get very nice night time images, the specific settings, how those settings function, etc. I could give you a list of instructions for how to replicate my process and the photographs I take. Same can be said for CGI, watercolor, digital art, pottery, etc. It is a learnable skill. Learning to get good results from an AI is still something that needs to be learned, but it hardly requires any level of actual skill, high level of knowledge, but low level of skill, and then to get predictable results is another feat all together. Can it develop to the point of that, sure, but right now a vast majority of prompts are barely the length of a good commission description.

AI generators are solely reliant on the samples they are fed, humans are not. When humans create something, through real media or digital tools, those creations are filtered through our experiences, and we then can engage with that, our inspirations, our aspirations, our motivations...all influence what we make. AI generators take that out of the equation, generators do not engage with what they are fed, the companies who have made them are selling this 'tool' for profit, which was fundamentally founded on stolen works that they did not pay licenses for, and this is why we afford humans the ability to draw inspiration from others, because of how we filter it through a cognitive process, actively thinking, reflecting, learning, etc. AI cannot do this. It doesn't actually KNOW anything, all it knows is a series of weighted relations from one pixel to the next once it starts with noise and starts trying to consolidate it. Not to mention that MANY of the coders who have developed these AI don't actually know HOW any of these AI work beyond a surface level of the instructions it gave it. Each model works slightly differently and there are different types of models, but again, all you get is a fairly surface level explanation of how those types of models work.

I can see AI having some good uses, for those that have aphantasia and need something tangible to start working from. People stuck in art block and need to spit something to work from. But all both of those and all of the good uses require AI to be used as an adjunct TO an artist, not a replacement for which is what many people are trying to do with AI.

I highly encourage to fully watch this video from a creator who spent a lot of time in the AI art community, and gathered A LOT of information about it and why it poses a problem. AI CAN be a tool, but that is not how people are treating it, not even the slightest.
https://www.youtube.com/watch?v=9xJCzKdPyCo

And for a bit more about how something like ChatGPT works and generates its answers and why that is problematic
https://www.youtube.com/watch?v=l7tWoPk25yU

Both of these videos are very in-depth, the latter coming from a science communicator who has worked with a number of very smart people and even hosted a mini series after Mythbusters left the air.

Add a decent randomiser salt into the learning . Boom .

That doesn't fix any of the issues I mentioned. If anything it just makes it even less in our control.

It definatly does not fix any of the things you mention. Adding in a seed definatly brings the stuff another step.closer to being autonomous tho.

which is a scary thought, though....there is now so much AI art out there, that some AI datasets are learning from other AI art and it is making newer models worse, so....effectively inbreeding.

It is like a photocopy of a photocopy of a photocopy....ad nauseum.

It may be good for establishing a prototype or concept, but as is I still think AI has a long way to go. Especially with AI art. There's still plenty it can fail to do right like hands, toes, and heads of smaller creatures like snakes. I'd you add tattoos forget about it. They come out pretty much a mess on most AI art that use them.

I still think we can see AI used in a more limited context. I just don't think it can compete with traditional and digital art just yet other than through sheer volume. I could actually see it in analyzing damaged historical art pieces and creating a digital copy of what it most likely looked like when it was originally created, but that's more historical research than selling commercial commission art.

Basically yeah, as I said, there are perfectly fine ways that it could be used, especially for quickly iterating on ideas to get the creative parts flowing.

IMO, people aren't upset at the tool that is A.I. art. They are upset at how the tool is being used. The fact that the A.I. art systems are trained off of other peoples artwork, or even voices now. I have heard some shockingly decently A.I. "cover songs" of famous youtubers and etc and...that's the scummy nature of it, that's what, IMO, people are upset over, it is being used to effectively steal from others, or others have stolen artwork and fed it into an A.I. to train it to draw better.

I tink you are spot on... eraseeraseerase-darned computer... I think you are spot on...

Vix

The problem is not that AI image generation exists, its a very powerful tool

the problem is all the main AI image generation tools use unethical datasets copying the styles and skill other artists poured their hearts in to without their consent, often allowing anyone to replicate the style of another person with a simple token, this is a violation of one's copyright. AI image generators do not learn like humans learn, its job during training is to replicate a given image with a given series of tokens by manipulating a neural network to take randomly generated noise and denoise it in to a replication of the original work. When you ask it to generate something its effectively selectively tracing the works of others. It has no ability to think or problem solve and has no concept of what art even is, there is simply no creative input. this is like commissioning an artist and they trace the head of someone and the body of someone else and calling it original, you cannot claim that is an original piece even if the request itself was a creative expression. Websites that allow AI generated images and those who post them are just setting themselves up for a massive headache when these unethically generated images are inevitably declared unoriginal reproductions and they all have to go. Talking about history we had this problem in the late 90s and early 00s with ripped MP3s, it was not clear that copying a CD and reposting the MP3 online was a crime, no court had ruled about this and some BBS and forms allowed ripped MP3s and others did not. When the series of court rulings came down in 99, 00 and 01, websites had no time at all to remove all of the copied content before the lawsuits poured in. You get no grandfather protections as no new law was passed and many websites had to completely remove all user posted content regardless of when it was posted to prevent lawsuits or shutdown completely. the same thing happened in the mid 90s with libelous content on forums, once again in 2007 with TV streaming and 2012 with hosting just links to copyrighted content, Websites who remember this era in US history are the ones banning AI generated images, websites who dont need to get the inevitable reminder that will last another 10 years before we do this crap again.

EDIT: You 100% can make ethical datasets, Nvidia has done this with their own internal AI tools such as DLSS and audio denoisers along with their own paint art tool. DLSS is trained on gameplay with the consent of game makers, the audio denoiser was trained in house and the art tool was trained on images they took themself or purchased with express consent. That takes time and money but thats the only way this will survive.

Tell that to all the people that have decades of work taken without their permission to feed an ai dataset. The problem is not the technological advancement itself but the fact that there is this double speak in ai image generation communities as if the original artist doesn't matter while at the same time making predatory use of their work. The current usage of ai art is based entirely on this premise.