discussion AI CG Art isn't ethical

[MattShizzle]

It isn't itself unethical, but can definitely be used for unethical purposes - like making it look like some actual person did or said something that they didn't. It won't be long until it's very difficult to tell if it's real or AI made.

 

[ZZBB24]

I don't want to come off as pretentious or obtuse but I don't consider the smorgasbord of art thrown into a proverbial vat and then served back to the end user as "art." AI content is content, nothing more nothing less. The only ethical application that I have seen when it comes to AI is shitposting and user-made material. Someone using AI and passing it off as art is entirely fraudulent in my opinion.

 

[lcflex]

I don't want to come off as pretentious or obtuse but I don't consider the smorgasbord of art thrown into a proverbial vat and then served back to the end user as "art." ... Someone using AI and passing it off as art is entirely fraudulent in my opinion.

I'd concur for a lot of what is passed off as "art", but I disagree with the blanket statement. AI is just a tool, and what is or is not art isn't defined by the tool used to make it.

There is a credible argument that all art is just observed prior art mixed with lived experience. No art is created in a vacuum. No artist ever has a truly original idea. AI doesn't operate in a vacuum either, not in its training, nor in its operation. For a real artist - not the casuals or hack jobs like me - AI is just a tool to use, a paintbrush that holds itself. The artist is still directing it.

Process-wise, making images isn't as simple as telling the AI what you want and getting a pretty picture out the other end. That's what the majority of the garbage out there is. Making something that's actually good takes time, effort, and skill. It's a much more hands-on and collaborative (as in, collaborating with the AI) process than most people think.

 

[sitrukpc]

I don't like ai art for many reasons. Some are that it steals from many sources, people are trying to pass it off as real art, and most importantly it could be used to deceive people into believing something that is false.

 

[Ani_Sloven]

I view AI itself as a tool.
The question of ethicality should be applied to the user/developer and their intents and purposes for the AI.

There are a non small amount of people that get tricked daily by AI, but I would blame the person behind that rather than the picture/video they post.
I think the biggest problem is how easy AI is to obtain and use, cause it lessens the effort required for people to do bad things with them.

And about AI using unconsenting artists data. I'd blame the developers of the AI for specifically allowing/choosing their AI to feed and learn of the artist.

 

[MattShizzle]

It's better when it's not from art, but looks so real that it would be very difficult to tell the difference.

 

[s3gfaul7]

Call me morally bankrupt if you will but, to be perfectly honest, I don't particularly care how the AI was trained or whether the content it used to create a given image was licensed or not. All I care about is the end product and whether or not it brings me personal enjoyment. Yes, it's a very selfish position to take but when I sit with a game, I seek to unwind and take a break from IRL choices, dilemmas and issues. I simply refuse to be bothered by whether or not an image was generated by DAZ, Flux (or whatever other AI is popular these days).

*shrug*

 

[Hongbottlo]

Generative AI is corporations harvesting humankinds data and feeding back a mediocre, but convenient version of it back for maybe some profit.

 

[Tigerdiamonds]

Diffusion models don't steal the images they use for training anymore than a human using an image as reference.
Of course something "designed to be able to make anything" COULD be used to make a copyright/forgery/plagiarism
but if that output is used commercially in anyway, it should be covered by existing laws.

I do agree that ethically, any illegal photos probably shouldn't be in the training set

Trying to get them to replace jobs is a different argument and something I am against.

 

[entry1]

Diffusion models don't steal the images they use for training anymore than a human using an image as reference.

until they give you images/text almost identical to the sources. Also would do you really want to compare an algorithm to a person? Because i doubt they are the same at learning through observation.

 

[Tigerdiamonds]

until they give you images/text almost identical to the sources. Also would do you really want to compare an algorithm to a person? Because i doubt they are the same at learning through observation.

The fact that you have any doubts tells me that you don't actually know how AI work and are just guessing.
Diffusion models are basically image recognition run in reverse.

If you look at the visual "mind map" of a diffusion model, you can actually see much better how it works,
It basically starts with analyzing simple shapes and looking for hard edges (characterized by drastic changes in pixel value)
and then if you follow any given direction on the map you can see where it begins to associate specific shapes and colors with specific tokens.
For instance, if you looked along the vector for "fruit" you would see that it associates "small multiple fruit" like grapes or berries on one end and "large singular fruit" like melons on the other.

During it's training it looks at a billion different instances of any given token, and tries to find the common elements that define that token.
much in the same way that a human, over years of their early life, sees different cat or dog shaped things and has to figure out what the differences and similarities are.

If the extent of your ability to use the AI is "throw in a few tokens and make it spit out 1,000 results so I can pick one I like" then true it might not have enough data to make a completely unique version of that thing every time.
but like "The machine that can make any image has the capability to produce forgeries" isn't the mind-blowing gotcha you seem to think it is.

Also, your brain is just a really complex electro-chemical algorithm. (Well, really like 500 different algorithms piled on top of each other but still.)

 

[entry1]

The fact that you have any doubts tells me that you don't actually know how AI work and are just guessing.
Diffusion models are basically image recognition run in reverse.

If you look at the visual "mind map" of a diffusion model, you can actually see much better how it works,
It basically starts with analyzing simple shapes and looking for hard edges (characterized by drastic changes in pixel value)
and then if you follow any given direction on the map you can see where it begins to associate specific shapes and colors with specific tokens.
For instance, if you looked along the vector for "fruit" you would see that it associates "small multiple fruit" like grapes or berries on one end and "large singular fruit" like melons on the other.

During it's training it looks at a billion different instances of any given token, and tries to find the common elements that define that token.
much in the same way that a human, over years of their early life, sees different cat or dog shaped things and has to figure out what the differences and similarities are.

If the extent of your ability to use the AI is "throw in a few tokens and make it spit out 1,000 results so I can pick one I like" then true it might not have enough data to make a completely unique version of that thing every time.
but like "The machine that can make any image has the capability to produce forgeries" isn't the mind-blowing gotcha you seem to think it is.

Also, your brain is just a really complex electro-chemical algorithm. (Well, really like 500 different algorithms piled on top of each other but still.)

my point is beside how you say it works, it's just a dumb algorithm it should not be held on a person standard ever. Right now it's a tool that ingest various works and spit out results derived from that work and the people in charge of it have all the interest to not share any profit gained from others work

 

[TheVampirate]

Honestly you can have a pretty good argument that AI is 'fair use' under us copyright law due to it's transformative nature and there is actually case law pertaining to humans 'stealing' other people's works by tracing being not illegal. Until a country writes a good enough law about AI its pretty much legal unless it violates other laws on the books. There's stuff against deepfakes but that's different than what we're talking about. As far as ethics go, who cares about ethics especially when it does not hurt anyone.

 

[kakarottooalgo]

I'm not into AI art but the art is not the problem by itself
Making a profit by ripping someone else's work is fucked up but just producing an image for yourself is not bad

 

[Tigerdiamonds]

my point is beside how you say it works, it's just a dumb algorithm it should not be held on a person standard ever. Right now it's a tool that ingest various works and spit out results derived from that work and the people in charge of it have all the interest to not share any profit gained from others work

that is literally not how it works.
You continue to spread misinformation while saying how you don't care how it works.

 

[lcflex]

Diffusion models do not construct images using reference material. They literally can't do that; they don't contain the reference material in any way. The training images are terabytes of data (in compressed image formats); the models are gigabytes in size, they're 1000x too small to actually contain the source images. At best, you can say they have a vague memory of having seen the material they were trained on, embedded in a mapping of tokens to edges and colors. You can ask an AI for a reproduction of the Mona Lisa, and the best you'll get is something that sort of looks like it, but only at the surface level with the significant components included. Details will be wrong, background will be wrong, relative positions of things will be wrong. You'll never get something that is a credible forgery, except by complete accident.

 

[SirBanana]

The only part of AI art I find overly unethical is when people pass off their AI art as their own handmade creations from models trained off other peoples hard work, there is massive skill difference between typing prompts and studying anatomy to make an accurate picture and when AI art creators price it at the same value it feels rather unethical but who am I to say, I can't draw for shit.

 

[lcflex]

The only part of AI art I find overly unethical is when people pass off their AI art as their own handmade creations from models trained off other peoples hard work, there is massive skill difference between typing prompts and studying anatomy to make an accurate picture and when AI art creators price it at the same value it feels rather unethical but who am I to say, I can't draw for shit.

Agreed, it should never be passed off as a human creation. Pricing... really that's something for the buyers and sellers to work out for themselves. Who am I to say what someone else should or should not pay for an image?

 

[PervyGod]

Don't forget that AI is also using fuck tons of resources, those data centers are real hogs for energy and water.

 

[ChaosLoli]

I think that a lot of the ethical issues are overblown because this is a new technology. For the most part, if porn doesn't hurt anyone, then you're good IMO