Kickstarter bans AI-generated art enthusiast group "Unstable Diffusion" and refuses to deliver their successfully raised $56k funds

[jagdtigger]

41 minutes ago, 05032-Mendicant-Bias said:

This is an hill I'm willing to defend.

Creating art is a creative process, AI inherently lacks it so its pointless to argue that computer generated images are art.

[Holmes108]

15 minutes ago, jagdtigger said:

Problem is none of your examples ended as much jobs as AI will, you kinda comparing apes to oranges. And that is the essential issue, many will loose their jobs and wont be able to work in their profession because evdry company will use AI........

 

There's always unique circumstances in each of these situations that have come up in history. And they're always "really really important this time", but I just really don't see the difference. 

 

Economically, no matter how bad the outlook is though (I'm more worried about AI automation in general, than AI art specifically), the answer is the same. It's here, we just need to find out how to best deal with it. 

 

But I never think the answer is artificially trying to restrict innovation. If I want to make a game tomorrow with all computer generated speech because it's astronomically cheaper (or indeed just possible for me at all), that shouldn't/doesn't make me a bad person. (edit: and no, I'm not a game maker at all. Just an example). 

 

I'm sure the abacus makers were devastated when calculators came along, but I wouldn't support regulations, or even simple public hatred towards the calculator makers or users either. We just need to move forward. 

 

It's indeed possible (probable?) that handmade art will become as niche as handmade furniture. So be it. I'm not saying anyone has to be happy about it. I just don't know what any type of alternative is. It's the reality of the world we live in. There will be a place for human generated art in the world always, but it will probably be a lot smaller of a market than in the past, as it is with all legacy technology.

 

  

13 minutes ago, jagdtigger said:

Creating art is a creative process, AI inherently lacks it so its pointless to argue that computer generated images are art.

 

It's kind of pointless to argue labels and semantics. Whether or not something is "art" is a big conversation (I'm a huge gamer... I don't think videogames are art. Come at me.) 

 

But at the end of the day, whatever you're talking about, say images for example, they are competing for the same space. There's a market, and the AI image is in direct competition with a hand painted piece of 'art', so who cares what the AI piece is called.

 

If they were truly different things, practically speaking, then there'd be no problem at all then, right?

[Holmes108]

*double post

[05032-Mendicant-Bias]

17 hours ago, jagdtigger said:

Creating art is a creative process, AI inherently lacks it so its pointless to argue that computer generated images are art.

I wouldn't say a generative model lacks creativity. Diffusion models have distilled an incomplete simulacrum of human creativity in their weights. They had to do it in order to minimize their error function. I find it amazing how Diffusion models move backward through entropy, and fill the blanks left by entropy according to a statistical model, turning noise into recognizable images guided by their latent space. It's like seeing shapes in the cloud and slowly morphing the clouds to become the shape.

 

At their best Diffusion Models understand what the statistical relationship between pixels look like a Picasso, and are able to apply that style to something Picasso never drew to a convincing degree. That is a form of creativity.

 

Also, I'm not convinced Art requires intent, or creativity or anything. I think Art is a two way process, the ones observing will have their own subjective interpretation of a piece of Art and not even consider it Art. E.g. I can't understand modern Art for the life of me, I personally like works that are less abstract and more close to earth.

It feels to me some metrics that current technology is bad at, is arbitrarly choosen to explain why that technology can't produce art, just like photography fundamentally couldn't be art in the 1900s.

  

17 hours ago, Holmes108 said:

Economically, no matter how bad the outlook is though (I'm more worried about AI automation in general, than AI art specifically), the answer is the same. It's here, we just need to find out how to best deal with it. 

Generative Models are such a game changer! LLMs are good enough that I can entrust them with commenting and  writing small well defined functions saving me lots of time right now. I get rare generations that save me lots of time on bigger functions by getting me 80% of the way there.

 

Our civilization has some humongous problem to solve. Climate change, fusion power, personalized medicine, circular economy... There is more work to be done than ever, and AI is the only tool fit for the job.

The role of LLM in Education is a big one, I'm convinced it'll level the playing field a bit between advanced nations and developing ones. If only I had an LLM like GPT4 when I was learning how to code. It's surprisingly good ar suggesting new libraries, new languages, new code and more.

I'm really optimistic about the impact of more advanced generative models to our civilization and way of life.

[Kisai]

2 hours ago, 05032-Mendicant-Bias said:

I wouldn't say a generative model lacks creativity. 

It lacks creativity in a way that makes aphantasia seem like a blessing.

 

2 hours ago, 05032-Mendicant-Bias said:

 

At their best Diffusion Models understand what the statistical relationship between pixels look like a Picasso, and are able to apply that style to something Picasso never drew to a convincing degree. That is a form of creativity.

No, that is a called kitbashing. Taking existing things and glueing them together.

 

What an AI is doing is a bit worse than aphantasia, it's basically doing this:

 

So it knows these words go in the image and go together, but doesn't know how they fit.

 

2 hours ago, 05032-Mendicant-Bias said:

Also, I'm not convinced Art requires intent, or creativity or anything.

 

 

Art requires intent, otherwise all you have is a machine spitting out garbage. Techbros want that because they can generate passive revenue from creating endless low effort garbage and sticking ads on it, contaminating the internet with content that is worthless.

 

 

If I ask someone to "draw a cat walking", a typical artist would draw something like this:

 

But someone who lacks this mind's eye, would probably think of 

This is like a 6th generation copy from the source video of a cat named Momotoro, played backwards:

https://soranews24.com/2014/10/18/gripping-footage-of-a-cat-walking-backwards-on-two-legs-is-todays-must-see-cat-video/

 

Stable diffusion however comes up with:

 

Yeah no, none of these even resemble a cat walking like a human, and most of them are cronenburg cat shapes.

that bottom-middle image, looks like something that google maps has:

 

Under normal circumstances a human would not think of these kinds of images, because they make no sense.

 

This is why generative AI are basically "infinite meme machine" , they create low quality garbage, that occasionally results in a laugh, but you're not going to generate much that actually is good, just a lot of weird kitbashed garbage that you have to spend a lot of time sifting through to find anything good.

 

AI's do not understand the prompt. They are "the chinese room" 

Quote

(1) Some critics concede that the man in the room doesn’t understand Chinese, but hold that nevertheless running the program may create comprehension of Chinese by something other than the room operator. These critics object to the inference from the claim that the man in the room does not understand Chinese to the conclusion that no understanding has been created. There might be understanding by a larger, smaller, or different, entity. This is the strategy of The Systems Reply and the Virtual Mind Reply. These replies hold that the output of the room might reflect real understanding of Chinese, but the understanding would not be that of the room operator. Thus Searle’s claim that he doesn’t understand Chinese while running the room is conceded, but his claim that there is no understanding of the questions in Chinese, and that computationalism is false, is denied.

 

2 hours ago, 05032-Mendicant-Bias said:

The role of LLM in Education is a big one, I'm convinced it'll level the playing field a bit between advanced nations and developing ones. If only I had an LLM like GPT4 when I was learning how to code. It's surprisingly good ar suggesting new libraries, new languages, new code and more.

I'm really optimistic about the impact of more advanced generative models to our civilization and way of life.

 

No, the direction things are going, AI generated content is degenerating the internet as a research tool. AI's ripping the text from legitimate sites, laundering it through ChatGPT and then posting it as "new content" 

 

This is also going on for music. To a lesser degree for artwork itself, because people don't want to consume garbage when they know there is no intent behind it.

 

The solution that needs to come down the pipe is the watermarking and metadata requirements that hardware shows the entire chain of custody, and is not stripped off by intermediate software or websites in the name of saving a few bytes or "protecting privacy"

 

A drawing will enter the chain of custody from being scanned, or by being sketched in photoshop or CSP or something. The software should keep the metadata for every software it passes through that changes the file. AI metadata should retain the prompt and a hash of the model it was generated from so that someone could "re-create" the image on a later version of the model in theory.

 

In practice, this loss of metadata keeps happening because of the fight between privacy advocates and commercial interests. Privacy advocates want all the metadata stripped. Commercial interests want all the metadata to be retained. From a legal standpoint, if you've stripped the metadata, as the host, then you've "damaged" the copyright of the file, because now the host is claiming the copyright instead of the uploader. 

 

If a regulation came down the pipe that images to the internet must have proper copyright metadata or visual watermarks in order to challenge copyright infringement claims. This would encourage artists who want to profit from their artwork to retain the metadata data. In a situation where two artists claim to have created the same image, the metadata could be compared, and determined that the images do not infringe because they started from different sources. But AI? If the prompt and model type was listed in the metadata, it could be decided that an image is "different enough" from the one generated from the model to have a copyright, or maybe that the image changes are not significantly creative.

 

I do not see a future where generative AI plays anything more than an assistive-technology role. We are not going in the direction of the holodeck, we are going in the direction of the "infinite meme machine", which takes an input and tries to generate something to think about.

 

Sometimes, and we see this with TTS primarily, is that a generative output on a limited training set does amplify weaknesses in the source material. For example, you may not realize you have an accent, but if you train a TTS on your voice alone, you will hear your accent amplified in the generated output. But if you "fine tune" a large model with your voice, it won't sound "like you", and people who know you won't be tricked by it. It will be the large model that has picked up how to "sound like" you, but not "speak like" you. It doesn't know what words you'd change, what kind of slang you'd use.

 

This is why we have stereotypes of "robots" speaking stiffly, because, this is what we "expect" from a computer. If that robot starts using your voice, it will still sound stilted using the same diphone TTS assembly.  Where AI's have improved here is learning how you say "phrases". So if say "What the ever loving ****?" a certain way, it will learn to say THAT phrase, by overfitting it. It will sound like you, because it has no samples of you saying it any other way. So like how Generative Text AI LLM's predict text, TTS's designed this way predict the sound based on the text.

 

So as an assistive technology, I feel the good outweighs the bad, especially in translation. Being able to speak English to someone who doesn't speak english, and have the machine both translate it accurately, but also retaining YOUR voice, makes it easier to follow a conversation, than some robotic voice that doesn't match the sound and pacing of your voice.

 

But as a source of truth, The AI's do not understand truth. It only understands relationships between words.

[SpaceGhostC2C]

On 12/25/2022 at 8:15 AM, Thaldor said:

So this is the kind of project Kickfarter does act upon.

Ultimately, the problem is not so much what Kickstarter chooses and what it doesn't, but the fact that it gets to choose, at any point in time, with no restrictions and no accountability. It's just a terrible institutional design that results in terrible contracts between startups and Kickstarter, and between Kickstarter and backers. AS a backer, you retain all risks, no guarantees. As a startup, you retain all costs and receive no property rights. It's a miracle it's able to exist. Makes you wonder whether it's a question of people being stupid or a question of financial markets being too dysfunctional that such a terrible idea gets to exist.

 

[Kisai]

I swear I do not have a crystal ball.

 

https://news.bloomberglaw.com/us-law-week/california-lawmakers-push-for-watermarks-on-ai-made-photo-video

Quote

California lawmakers are drawing up multiple plans to require watermarks on content created by artificial intelligence to curb the abuses within the emerging technology, which has affected sectors from political races to the stock market.

 

At least five lawmakers have promised or are considering different proposals that would require AI companies to implement some type of verification that a video, photo, or written work was made by the technology. The activity comes as advanced AI has rapidly evolved to create realistic images or audio on an unprecedented level.

 

Advocates worry the technology could be ripe for abuse and lead to a wider proliferation of deepfakes, where a person’s likeness is digitally manipulated to typically misrepresent them—with it already being used in the presidential race. But such measures are likely to face scrutiny by the tech sector.

 

This article reached me AFTER I had written the post above. It's not the first time I've seen proposals for AI watermarking.

https://www.bloomberg.com/news/articles/2024-01-18/meta-s-clegg-wants-standards-on-watermarking-for-generative-ai

 

And we've been hearing about it since middle of 2023

https://deepmind.google/technologies/synthid/

 

The problem isn't really a tech problem, it's a regulatory problem. If you require one watermark to be present for AI and another for non AI generated content, you use the lever of Copyright for how to protect artwork from humans from generated artwork by computers. 

 

ISRC records for music should be required in order to claim them on Youtube, however users aren't required to tell youtube the ISRC records of music in the video they upload. So perhaps if it became a requirement in order monetize the work, that would get people to do so. And that should go for stock images and AI artwork.