discussion AI CG Art isn't ethical

[Karamchad]

Ethics and morality has nothing to do with it. It is just a tool similar with digital photography, photoshop, graphic tablets and any other around.
The point is that AI art feels empty and does not have the same taste as hand drawn content.

 

[dk84]

I disagree.
I like to draw using real pencils and papper. And I could have said the same thing that u said about photoshop o digital art... but is nonsense

 

[Rebel 357]

I've seen no one mentioning the fact that most of these artists use other IPs for their so-called art. You can see many hentai artists making money off of characters that are not their own. Seen plenty of artists who's majority of work is that of others, that they are making money off of. Pretty double-standard if you ask me.

 

[kurgan]

disagree.
I like to draw using real pencils and papper. And I could have said the same thing that u said about photoshop o digital art... but is nonsense

I share this opinion.

At the end, AI is just a tool. Every artist creates its own art learning from art of other artists or from pictures that may not be their owns. Every writer has learned to write reading other writers. That's how it works, just that now is a software who does it instead of a person. We have been replacing persons by machines from a long time now. Look at items around you and tell us how many of them have been completely handcrafted. AI is just a new step of technological progress.

 

[Trunk Monkey]

Most of the image generation models that are available to the public have been trained on LAION-5B, which is a massive dataset of image/text pairs scraped from the internet. So it's little different than having a person look at images from the internet as they work. And they aren't actually "AI". They're purely deterministic. For any set on input values, there is exactly one output. Many models are open source, and you can read through the source code to see how they work. Stable Diffusion is even written in Python, making it easier for more people to understand.

 

[vixidev525]

Most of the image generation models that are available to the public have been trained on LAION-5B, which is a massive dataset of image/text pairs scraped from the internet. So it's little different than having a person look at images from the internet as they work. And they aren't actually "AI". They're purely deterministic. For any set on input values, there is exactly one output. Many models are open source, and you can read through the source code to see how they work. Stable Diffusion is even written in Python, making it easier for more people to understand.

Aren't humans deterministic as well? You put a person into the exact same situation twice and they will behave the exact same way. Humans have a memory and remember things while current AI does not, but the same human with the same memory will make the same choice, just like an AI would.

 

[Trunk Monkey]

No... humans can learn as they go. current generative "AI" models cant.

 

[vixidev525]

If the human learned, it isn't the same human in the same situation, but a smarter wiser human.

Current AI models aren't great at learning, and that isn't just with learning on the go. When training an AI it can learn from examples, but even the best AIs take far more examples to learn something than a human can. One of the next steps is figuring out why a human can learn from a few examples and how to capture that in the AI models.

But even if they learned on the go, you could always wipe their memory and restart them from scratch in ways you can't do with humans. At least not ethically.

 

[HTYYHRK]

Aren't humans deterministic as well? You put a person into the exact same situation twice and they will behave the exact same way. Humans have a memory and remember things while current AI does not, but the same human with the same memory will make the same choice, just like an AI would.

This isn't a provable hypothesis, since as you say later we can't test if a person given exactly the same "inputs" will take the same action twice because previous results influence future results. There's an argument to be made in favor, but we don't know what every "input" to human choices is which means one could be truly unpredictable. Free will if you want.

As for the OP, none of the points are really limited to or even particularly AI problems. All images have a copyright holder, and we can separately consider if we should strengthen personality/image rights. AI training sets should be required to have license to use any images they use for that purpose. It would be within reason to give people more control over the use of their image, even if they don't own the copyright to the image. This is not a problem limited to AI training though. Laws already grapple with the use of legal images of children to create illegal images of children, AI can simply fall under the same set of rules. And the last one just seems to be a compliant against the efficacy of AI which doesn't have to do with ethics.

On the other hand AI art is a very interesting tool. I've already seen quite a few game projects that use it to create dynamic images for characters based on their changing description in the game. Entertainment companies will probably get to some sort of equilibrium in using it with regular artists to still have quality visuals but take out a lot of the scud work needed frame by frame. To the extent that this will "threaten jobs," the impetus has always been on craftsmen to learn the latest tools of their craft. That's not to say individual projects shouldn't be called out if they are low quality because of bad AI use or that people can decide on their own not to consume anything made with AI, but that's not an indictment of the technology.

 

[zark]

For years now game have (not just adult) been push to put more and more of there time in to the art and sacrificing the time needed to program them. giving us game that look good, but are junk! A.I. art at this time is not up to fixing this imbalance, but it is improving fast. Is a lot of it "soulless", yes.... for now, but just look and the new game Brainstorm(on LC) and tell me if that is not a move to being less so that the stuff we have seen in the last year.

​

 

[Trunk Monkey]

If the human learned, it isn't the same human in the same situation, but a smarter wiser human.

It's the same human, the fact that the individual can draw from experience is what differentiates humans from generative models

Current AI models aren't great at learning, and that isn't just with learning on the go. When training an AI it can learn from examples, but even the best AIs take far more examples to learn something than a human can. One of the next steps is figuring out why a human can learn from a few examples and how to capture that in the AI models.

Current models aren't just not great at it, they're completely devoid of any feedback system and can't learn at all. They can be 'trained', but it's a fairly limited type of training, they're getting image/text pairs, but not getting any sort of rating... like 'This image is a food representation of these words, but the other image isn't'.

And, honestly, you wouldn't want them to learn. At least not if/when they are publicly accessible. Imagine a model that learns from it's users... and the bulk of them are, at best, random 'average' people on the internet. The model would get worse over time!

 

[tyktamsyl]

Get off your high horse. Do not tell us what to do, or at least, do not expect people to obey you.

If people want to look at AI art, they can do whatever they please.

Also what part of AI art is hurting people? People are still going to do fucked up shit, they don’t do fucked up shit so that an AI model can take it and generate stuff.

You remind me of religious people and how they used to tell you that masturbation is bad for your soul or some shit.

Respectfully, mind your own business.

 

[vixidev525]

Current models aren't just not great at it, they're completely devoid of any feedback system and can't learn at all. They can be 'trained', but it's a fairly limited type of training, they're getting image/text pairs, but not getting any sort of rating... like 'This image is a food representation of these words, but the other image isn't'.

The new models can be trained on a deeper level. It isn't just input pairs. They have begun to classify things in self taught categories. Often these tie back to categories humans recognize and we can train them on these category levels. When they don't tie back to what humans recognize, it indicates a higher level pattern we haven't discovered yet. Or a bias in the training data.

This is part of the reason they can generate images that are of concepts they haven't been trained on by combining individual concepts they have learned from the training data.

 

[Trunk Monkey]

That's not entirely accurate. The training data is still just image/text pairs. During training, they're just using forward prediction weighted by probability of words being paired and occurring in proximity to each other. It's the same effect as doing a complex database query. The only difference, is that a compute can handle parsing much larger datasets to identify colorations between words, and do it much faster, than a person ever could. Eake FLUX.1, for example, released about a week ago... trained on LAION-5B. Which is just a large set of image/text pairs.

 

[floydrowx77]

It just depends on how you use Ai, the same thing can be said by an artist who draws hentai. Just look at Shadman he literally made a comic of Hillary as a Loli.

 

[dereguedeixe]

I respect ai on a lot of stuff, but on art is a really turn off

 

[48e1-920b-9a4ff0ce20]

except that it certainly will have an impact on the porn industry

expectation: porn industry collapses as fake content overwhelms it
reality: porn industry expands as goonwhales subscribe not only to girls who pretend to like them, but also programs vaguely imitating girls who pretend to like them

They're purely deterministic. For any set on input values, there is exactly one output.

maybe, but since you start with a random base for any given image, which is iteratively transformed into the final image.
since the noise is always different, the images are always different

Stable Diffusion is even written in Python, making it easier for more people to understand.

python is used to tie the shit together. the hard bits are all just shit python calls out to

 

[Wrazzer]

Human generated content would have sources of inspiration for creating the final product. Perhaps based on poeple in real life. Is AI so different to what we are. Not a big fan of the current content style of it however the potential is difficult to fathom.

 

[CozeeBeaver]

One of my pass times is to create AI enhanced pics of some of my favourite characters from VNs.
If posted to LC I fail to see how something like this can upset anyone or have it removed by moderators when its a so called "RL" version of a character created in artwork software such as Daz.
The attached files (which will probably be removed!) show a couple of example of what I'm talking about. Both pics show a screenshot on the left and the AI enhanced pic on the right. (Models are Sister and Mum from Life in Santa County by Bold Bash Studios)
What is the harm in this?

 

[CozeeBeaver]

Following my last post, I just wanted to add that based on what I posted makes USER 62433's ( or whoever he is) original post and bullet points a load of bollocks!