Games

All ai generated images are by definition derivative of other works. No machine learning algorithm can actually create anything, they can only mimic and rework what already exists.

Put a brand new human in an empty cave and they can still create art. An AI is literally incapable without being fed a library of art that people already created

edit: i am so sorry wow you got a lot of other responses while i was writing this

to me the biggest point against generated art is that it's just crap.

A shitty man-made drawing will still have some sort of consistency between each piece, an inherent uniqueness due to the artist behind it, and the mistakes or inaccuracies aren't nearly as jarring or disturbing. It takes some serious dedication to draw or paint something that's unpleasant to look at.

Some schmuck with stable diffusion (or whatever the new favorite slop churner is) could generate a million picturesque landscapes, selecting only the finest of outputs, and they'd still be samey, weirdly framed, and plagued with unpleasant oddities in the details, only enjoyable from a distance and without looking for too long.

In the context of game development and illustration (comics etc), AI falls apart real fast. You often need to depict the same characters, objects, and places, possibly hundreds of times in different poses, actions, settings, etc. Weilding an image generator and sifting through dozens of results at a time, trying to find one that shows your original character doing what you want while looking the same (no sudden changes in outfit or number of fingers), sounds downright sisyphean. Not to mention the lack of intention behind each asset, subtle details or expressions being even more completely unobtainable than reasonable consistency between shots. It will either be somehow more effort than doing it yourself, or look even more low-effort than AI "art" usually is. Both outcomes are inferior to doing it manually.

The only thing i can't yet come up with a real issue against is generating boring repeating textures for 3d scenes. One of the least interesting parts of the process.

So that's why we need real artists. Because AI slop looks like shit.

expressing creativity and inspiring others to do the same is unique to the human experience. AI proponents view art not as an expression of self but as a faceless product to be consumed (and as such are happy to remove the human element entirely) but the latter without the former has no meaning and inspires no meaning in others.

the art we create is also a result of our own study, practice, and effort. much of what makes art worth appreciating isn't just the end result but the process itself. if someone makes ghibli-esque animation with sora or whatever, it looks fine, but i can't look at it in awe of the massive effort and years of study and practice it took for the artists to get to that point and feel inspired to do the same with my own ideas, because it was generated automatically by a machine and i am not a machine.

to see proponents and users of AI try to eliminate this uniquely human experience that has given many of us a life purpose throughout all of our species' history for the sake of quicker and easier consumption is something that must be opposed, and without capitalism i don't think anyone would truly be compelled to try to automate art because art is not and was never about making products more efficiently.

Well the human has to still prompt the program, and determine the level of quality, and create the initial database from which the LIM draws it's programming data from. They haven't eliminated the human, just shifted how much effort and skill is required to produce things at the other end of program, which is why so much of it is horrendous slop, because time, effort, and compromise is how you build a style, genre and taste. And there are plenty of decent LIM and LAM generated things, but it is from people who already had an understanding of taste and genre flexing their use of a new tool.

But even then you still need extremely talented human artists, who have taken extraordinary amounts of time and effort to develop the skills to produce the initial database.

The idea that you can remove humans entirely from the process is genuinely one of the most absurd and misanthropic things I've ever heard, a borderline sociopathic understanding of art. About as bad as any capitalist understanding of creativity and creation.

And none of this requires a "soul". It just requires you to not be a vulgar materialist. We imbue art and animate it with the very product of our labor.

I have yet to hear any argument on why humans are so necessary for art and why AI can't produce it that doesn't rely on some metaphysical and made up bs like "soul".

Art is human expression. Art in any form depicts the interpretation of one's personal experiences, ideas and feelings. That might sound like metaphysical and/or made up bs to you but that's what classifies art as opposed to, like, any depiction of anything. A short story is an artwork, a court transcript isn't. A movie is an artwork, a video recording of a press conference isn't. A child's drawing of a house that's literally just a square with a triangle is an artwork, an architect's sketch of a house they're about to build isn't.

If I ask an artist to draw a house, the result will be that particular artist's subjective interpretation of what a house looks like and every stroke of their brush will be influenced by their subjective experiences and influences, and this subjectivity is what makes it "art". If they just traced over a photograph of a house, even though it might look "better" by some standards, as an artwork it would be significantly lesser. If this is too metaphysical for you, I challenge you to give me a better definition of art. To be fair, I can't claim that AI art is not art at all, since there is still an element of human expression in the prompt that is given. It's just, like tracing over a photograph, significantly lesser. Because whatever prompt you give the AI, the result will by virtue of the technology be the most generic interpretation of that prompt. Generative AI can only ever give you the lowest common denominator. AI art minimizes the user's creative expression and fills the gaps with generic slop, and this is not something that will improve with time, this is literally the core mechanism of any probabilistic model.

So yeah, AI art is lesser art. Any AI artist is a lesser artist. If you wanna consume what Midjourney or Sora have statistically determined to be the most generic slop possible, be my guest. I don't.

Human art is a communication of said human's (or set of humans') life, experiences, etc. It tells a story that's true to life and insightful into another mind. I learn about those lives and improve my understanding of reality.

AI art looks like the above, but instead doesn't comport to the reality of a human's experience. Instead I'm tricked into losing insight into the minds of others because it looks real but isn't the product of said minds. I am then left more alienated and inevitably misled about reality.

Hope that helps. Purely rationalist. No soul needed.

It doesn't help that most ai art is nothing more than a prompt thrown into a commercial LLM or stable diffusion workflow, with hardly any further human input.

I have yet to hear any argument on why humans are so necessary for art and why AI can’t produce it that doesn’t rely on some metaphysical and made up bs like “soul”.

I don't believe you

Well this might help:

Can you explain why a beautiful sunset isn't art?

Computers are binary state machines with no reciprocal feedback reinforcement built in. A biological neurological system forms associations with feedback mechanisms reinforced by chemical frequency, gradient and multiple inputs that don't have to be classified or sorted into appropriate i/o setups made by people. A nerve from the back of your brain that associates with your breathing can associate with a group of nerves in the front that associate with smell and memory so your breathing might reflexively shift a little bit if you think of the chlorine smell from swimming in a pool hard enough.

That's to say: being alive over the years creates associations between ideas on ways that aren't entirely logical, and in ways that maybe people don't discuss or think of often. A kind of latent, collective unconscious about experience.

An LLM does not experience (if it experiences at all) associations from weak -> strong. It has no associative history from it's own life what so ever. It takes input, references a data set, picks a response based on probalistic algorithm, and presents it. It doesn't understand how to convey meaning through associating it with its experience like a person. It can be taught /why/ a brush stroke should look the way it does but it doesn't have any capacity to feel that it's correct, and, likewise, capacity to understand how to make it wrong or different on purpose to convey a meaning.

To us, a tree is green because we've seen green trees. We remember the light at the top, the shade at the bottom. We also, based on our own lives, know what the leaves and branches should look like. In a way, an LLM can learn that too from data set, sure. BUT an LLM can't remember it's first tree, it can't tell you that all the pictures of trees it saw in children's books when it grew up looked so much different than the ones it saw outside because it lived somewhere where the trees were Eucalyptus and the rest of the ground was dry.

An LLM won't paint a tree thinking about what it would feel like to touch it, how the wind on its skin feels in the breeze and how the memory of that leads it to construct the subtle splitting of the tree's canopy, how the sound of creaking branches makes it remember to make the branches look weak and spindly like it used to see on walks back from the corner store because this painting is about autumn when the wind picks up when the smell of leaves fills your nose. An LLM might make the sky cloudy because it feels, statistically, the prompt that requested the drawing has clouds but not because thinking of the cool fall air makes it also think of rain clouds and rushing home. And through rushing home you think of deer and add that to the landscape and have it running. And then maybe someone sees your painting with the wind and the storm and the running deer and they get the impression, the vague impression, the ghost of a feeling... "oh boy this painting reminds me of needing to get home before it rains, I better stop by the corner store on my way home" and BAM that's it.

I'm sorry if this is meandering but you're seeing it right? Maybe you've seen people say "soulless" a lot, but I think that's because to people the soul is the sum of our experiences, and there's an inherent universality to certain ideas you just /have/ to be alive for both in the past to experience and in the present to need to use those ideas to stay alive and do things you want. There's certain patterns and associations that always form throughout living and that's the secret sauce to any art imo. Art is a language first and foremost, it conveys meaning, but it relies on these ideas that need time and associations and pressure and varience to build. An LLM doesn't have that capability.

The part that makes art art isn't just the product, it's the fact that a living person with experience made it. Without the fact that another human being intentionally composed the work, there is, automatically, the destruction of linguistic or relational capacity of the work. LLMs do not EXPRESS their ideas because they do not HAVE ideas. Though you /may/ relate with a generated work in some way, there is no connection to anyone who made it. It's factually lesser because it no longer expresses an intention or carries a conveyed meaning. It isn't art anymore because art is expression. LLMs do not express, feel the desire to express or have any experiencial touchstones with which to form expression. It's a math problem and that's it.

Maybe, sometime in the future, they can use software to simulate a brain and the proper neural connections. Maybe even with hardware. Maybe they can scan a human brain down to sub-atomic resolution and run it and it makes art. But AI bros fucking /wish/ their silly little calculators could do something as elegant as hearing a door creak and being reminded of their grandma's house and then thinking "hey I want to make a painting of a dingy wooden house to express to someone that experience".

Large data models need constant external, obscured input to operate; without it, the model starts eroding. If they become prominent enough, they start to have an accentuated feedback loop. Iterated Stable Diffusion, Midjourney, DALL-E, etc, will tend to converge on making these grandiose palace halls wih the same lighting. Humans won't do that. That's similar to why large data predictive/associative models cannot invent a new genre (like Impressionism).

This is leaving aside the process of making art with all its potential divergences, and the accessibility of art as a thing that people can do, and the externalities of using LLMs.