For generative art, trademarking your name might help prevent people from using it in prompts, but for general copyright, where does the line stand between someone casually publishing every color in the rainbow, every note combination, every letter in the alphabet, and claiming anyone else is infringing on their copyright?

If someone copies your thesis, abstract, poem word for word, that is clear violation of your IP, but we are all remixing words that everyone uses, colors, brush strokes, API terms, programming language keywords, and notes. Copyright law has the fair use doctrine and transformative use is explicitly allowed to allow iteration. There is some level of granularity that is essential to creativity - otherwise one entity can copyright all possible combinations and prevent any creativity from happening legally. If AI goes below that threshold, all of humanity has a chance to iterate far faster and find new spaces and fill new needs for everyone. Humans have been able to draw in the style of Picasso or Monet for centuries. A program doing it is not infringement, just much faster iteration.

> A program doing it is not infringement, just much faster iteration.

"Much faster" is absolutely relevant, morally and legally. Visiting a website a bunch of times is not illegal, programmatically DDoSing it is. Having a private conversation with someone and writing down what they said afterwards is not illegal, but recording the conversation and perfectly reproducing it without their permission often is. Shouting at someone in public is generally ok, having a drone follow them around anytime they're in public playing a recording of whatever you shouted is probably not ok.

Computers are not people. Just because it's ok for a person to do something, doesn't mean it's ok to have a computer do the same thing a billion times per second.

DDoSing is bad not because of the speed but because it overwhelms the infrastructure a product is designed for. Doing it to your own computer by making it crunch AI models until it runs out of memory is perfectly legal and iterative. Printing pages of a book in seconds vs dedicating lives of people to hand draw each letter in the monastery is iteration.

Computers are not people. Computers are iteration tools people use to free up precious lifetime they have and bring more value to the world. If you are a human who trains 10000 hrs to invest like Paul Graham, or draw like Thomas Kincade, or play the piano, or operate as a top brain surgeon, you have spent a fraction of your life to do this fast and reap the rewards. But that fraction of your life has tremendous cost on society. Many people paid with their time and money to feed you, teach you, house you, during that time and during your upbringing which allowed you to have those 10k hrs to dedicate to this task. Now all of that work can be used by you to do exponentially more with your precious life. Instead of spending days or years making a portrait, you’d spend seconds. Now you can find higher purpose and solve much bigger problems - instead of asking for 100 to hand draw a portrait for a few hundred people in your lifetime, you could create one for every teen who needs a boost in their self esteem and raise their confidence and ability to cope with challenges in their life at massive scale.

More importantly, since there is a huge scarcity of people trained to fulfill each niche need that forms a bottleneck on society’s capacity to use that. Imagine if instead of airplane we counted on a few trained supermen to fly people who needed to cross places fast by hand. How many people would die before they see the world or are taken to a doctor, etc. The world can’t survive on superheroes or super trained people. The world can do more with the time and lives of people in it.

I disagree but this is a reasonable perspective. One specific point:

> instead of asking for 100 to hand draw a portrait for a few hundred people in your lifetime, you could create one for every teen who needs a boost in their self esteem and raise their confidence and ability to cope with challenges in their life at massive scale.

This is a misunderstanding that reminds of those startups that were like "we realized people love getting hand-written cards, so we built a product to learn your handwriting and generate them for you!" No, the effort is the point. For people who like those cards it's not about the aesthetics of handwritten text, it's about knowing somebody dedicated some of their limited time to you personally. A depressed person is not going to be cheered up by an auto-generated portrait, even if it's indistinguishable from one a human artist spent 12 hours on. You can't "scale" human connection like this, unless you hide the fact that robots are involved.

I'm not saying all technology is bad. I think robo-surgeons would be great if they can save more lives, even if they put human surgeons out of work. In this particular domain, right now it seems like these tools have the potential to discourage future generations of artists, which would be self-defeating because the models are not AI and will stagnate without additional training data. I don't think they should be banned, but I think we should take human artists' concerns seriously, not co-opt someone's artistic identity if they ask us not to, and try to make sure we think about the unintended consequences of a powerful new tool.

People who love to walk will continue to walk even when there are bikes, cars, airplanes, self driving tech, and teleportation etc, available to them. AI art does not discourage artists any more than restaurants discourage home cooks who enjoy cooking. In all of those scenarios the tech caters to people who are in need of a task done and not in desire to spend their life on the craft.

There will always be unintended consequences and some will be severe. But what is happening now is pent up demand that finally found an outlet - like a bunch of high pressure mountain water that found a hole through a cave wall and into the ocean - it is gushing.

Instead of blind fear of change, I try to see the value previously unseen and it is tremendous. As the creator in the article said she does not see her true art in the stable diffusion creations: the eyes that speak to character in each character, the poses that show confidence or query, or passion. Instead she sees images that mimic her style of drawing.

I do see how someone may choose to not become an artist for a living because AI art becomes so ubiquitous that they could never make a living with a paintbrush. BUT, with so much 80-90% of desired art generated by AI, there will now be huge demand for skilled artists who can take a generated image the rest of the way to desired results. I trust that human ambition and taste always expands past superhuman capabilities. The artists if tomorrow will have much different brushes than those of yesterday and be far better and more productive than those of yesterday. There will likely be 3D art in the real world and universes to explore in the virtual one. I’m more concerned we will and are running out of space to store our contraptions. Data storage manufacturers will be thriving.

That is a fake argument. It has been proven wrong on every one of these articles, yet pro-stealing people like yourself keep posting it. You can't copyright such works, only an 'installation' of those works. If you want to talk about copyright, maybe educate yourself on copyright. It's Title 17. https://www.copyright.gov/title17/

Which argument, which articles - be specific. AI content is not copyrightable at this stage. Drawing styles are not copyrightable. Name calling and labeling are ad-hominem attacks however and that has no place on HN.

Edit: First and foremost, I'm sorry you felt attacked. That is never OK. I need to step back from posting anything until I can be a decent human being. I regret not doing that before I posted.

----- People claiming you can copyright a canvas that is just a color (when you can't, you can only copyright the art installation/display). People claiming that you can copyright the alphabet (when you can't). It's just frustrating that HN wants to keep having this discussion but with ZERO basis in actual copyright law, and people making factually inaccurate claims as if they understand it. I had the same issue on a criminal law post. Someone posted completely factually inaccurate information regarding title 18 statutes, and HN blocking me from responding in a timely manner while people were reading the post. This just isn't a forum for informed discussion I guess, but peoples gut feelings and what they THINK copyright law is. Opinions are great, and needed, but established law is a real thing and should be part of a discussion that at it's core is about copyright.

Look at the inputs and outputs of generative AI art - specifically the ones shown in the post and others. The disconnect here is not our insights on copyright law (At least not for generative art. Maybe for copilot, code, and GPT3, but not for the art.) Hollie is a Character Artist. A Character has a combination of appearance, emotional radiance, mood, behavior, that have to fit in to create a unique persona and trigger an emotional response to that character from viewers. Stable Diffusion does not take or copy her characters - it mimics her drawing style in colors, hues (it reads what pixels are proximate to what other pixels and tries to apply that proximity to similar colors, line patterns in the future). When you create a stable diffusion image, you set a parameter of how much randomness to apply and how much variety, and what else to mix in. So any content it's trained on becomes a soup of colors and lines and hues and saturation. So if Hollie draws with de-saturated colors, and favors pinks and greens, an image produced with her name will probably have that, but not more than a soup mixed with objects searched by the words of the generator human. Code generation is different as a code has to appear in some form to work properly and that happens with a lot less variance, so generators there would likely remix a LOT more of the original, but images have a much much wider variance unless you ask specifically for as many keywords as possible (aka Mona Lisa by Leonardo Da Vinci) in which case you'd fall under trademark law as well probably (Which is also why Microsoft added trademarked images on their CDs so they could prosecute infringers easier).

I disagree that creative individuals have to do anything explicit here: copyright law is pretty clear that the burden of proof of right is with the copier, not the copied. I expect most artists won't be sending invoices for licensing fees just yet, but corps surely will bleed dry anyone that produce unlicensed derivative works that generates any income.

At the moment there's no legal protection for style in an of its self. Additionally there maybe (and should be) if this style-capture actually displaces the artists they're apeing. But I don't see that happening. IMHO its a tempest in a teacup.

Why? Because Ai generated "art" is a soupy mess and real life human artists can speak and understand colloquial language, work quickly, and develop new styles based on new direction.

But then again maybe we're looking at the death of a widespread industry like when gigantic industrial looms came on the scene, but I highly doubt it.

Then again, last of all, I do see a future where AIs generate full feature length photo-real movies in minutes based on prompts and cheaply.

Exactly, any coder and artist should learn from scratch without absolutely any exposure to existing code, work of art or even idea. Anything else is outright STEALING!

Excuse me when I make an apple pie from scratch.

Have you maybe possibly previously been exposed to the concept of an argumentation straw man? Feeding actual works of art into an approximation machine, and no expecting the output of said machine to not be owned by the author of the art is making a big assumption I think. There is the word copy in "copyright" and the model did definitely got a copy of the original at source. No matter the dilution, copyright is being breached, as I understand it.

Thats a reducto ad absrudium at best. While you have a point, schools and even museums are generally compensated for providing these training models to the public, to look at it in a ml way.

Sure, I also had to pay for the books I studied from, but Dr. Tanenbaum is yet to knock at my door to assert copyright on all the code I have written.

Out of curiosity, how would you feel if someone fed your HN comment history into a ML model, then used that to respond on every HN topic and conversation under the username "othergpderetta"?

My HN history is public, so I wouldn't have a problem with training a model with it. I would have a problem with the model attempting to pass as my self of course.

Interesting argument, but it'd be outputting fully sounding word salad garbage.

So exactly like my comments!!

This is a perfect example because, depending on the apples you're using, growing them may have required a license and adherence to licensing requirements.

https://mnhardy.umn.edu/apples/licensing https://provarmanagement.com/cosmic-crisp/

>Excuse me when I make an apple pie from scratch.

You’ll have to invent the universe first.

I expect a big bang anytime now!

As the artist in the article points out, the artwork in the model doesn't belong to her and by current legal standards she has no authority to give permission; of course the corporate owners do have authority, and I'm not even sure you need new laws to enforce the copyright complaint.

I was complaining about all of this when the derivation was based on "the internet" and everyone was being ripped off at once. All the AI-generated art out there is doing the same thing.

Of course most of this is being used to create derivations of trendy pop art, so are we really losing anything? Was there ever any hope for artistic capitalism as something that communicates in meaningful ways beyond the most local of scale?

you can get copilot to regurgitate copyrighted code verbatim, but I haven't seen stable diffusion recreating copyright works yet, which is quite an important difference.

Wasn't it regularly spitting out whole watermarks?

AI acts like an alien more than a copier in that case. If you tell stable diffusion to make you clip-art of people in a conference room, you will probably see humans that don't have noses or fingers, or have 4 arms and some unreadable text that looks like a watermark going across them. The AI parameters have been trained on millions of clip-art images and assume the image should have whatever statistically applies to most images that have the same keywords on them. You can't make it copy an image, even if you tried. You can't even get it to fix the face or hands without additional processing with differently trained models. It sees as an alien would.

was the image underneath watermarked, or it just reproduced the watermark style over an unrelated image?

NFTs to the rescue.

Serious question: What is the difference between a human intelligence looking at a work and using concepts from it in their own, compared to an artificial intelligence doing it?

If the copyright violation occured by the AI's inputs looking at the work... how is that different than an image of the work landing on a human's retinas?