Next Level: AI art is almost human, but fueled by unwilling artists
HTML-код
- Опубликовано: 25 янв 2023
- AI art has gotten good enough to mimic real artists, down to stylistic flourishes. But those works are generated from databases of scraped artwork often taken without the original artist's consent. What are the ethical, legal and moral boundaries of creating AI-generated content from other people's works?
-------------------------------------
Scripps News is your source for news and analysis covering the top stories from around the world. With persistent curiosity and no agenda, we strive to fuel meaningful conversations by highlighting multiple sides of every story. It’s news with the why.
Join our newsletter at bit.ly/2q1tepr
See more at www.scrippsnews.com/
Like Newsy on Facebook: / scrippsnews
Follow us on Twitter: / scrippsnews
Sincerely appreciate this piece. More people need to know about this. The current generative models are egregiously unethical and artists everywhere are being exploited. Thank you for making more people aware.❤
Great video!
Best. i love the host ! and also Scripsps news! both of u should get paid more for doing what is right.!
thank you for this :)
THE Aicreators try to use the FAIRUSE to justifiy and copyright infrigement, but COPY an artstyle of an artiste is STEAL. If tomorow I copy perfection the style of Toryiama or Guarnido, I will go to tribunal before the dawn, and it will be totally legit.
Nope, if you will mimic someone style perfectly (which is highly unlikely) it's fine and anyone who would sue you for that, would 100% loose. However it is an entirely different matter when machine do that.
@@iRaven23 Do you draw ?
@@gaho5624 yes
@@iRaven23 I don't understand so, how can you defend AI.
Art styles cant't be copyrighted, But Ai generators certainly violates te Fair use (Fourth factor), among other copyright violations.
Factor 4: The Effect of the Use on the Potential Market for or Value of the Work.
Ai has a strong effect on the illustration market and potentially undervalues original artworks and original art by the massive reproduction of different art styles which some artists are known for. Ai also eliminates most of the demand of comission artists to make a custom art piece, due the huge gap between prices and speed.
This wouldn't be a problem if Ai's were trained and developed ONLY with copyright-free data and public domain data, but they don't . This corporations like Stability Ai and Midjourney relies on "Fair use" and Artists names to make money even if they destroy multiple markets at the same time.
May traditional artists rule the art world once again…might be the only positive.
Us digital artists are niot traditinal artists either right
Dumb. No. The fight is on the internet.
@@abgrafix Yo, I do both digital and traditional art. I do digitally enhanced traditional art as well. I predicted this years ago: that the market would get over saturated with digital art, and tools would improve to the point where the humanity starts to disappear from it, and skill and concept becomes irrelevant, which is great for mass production of design and image products in a world of faster, bigger, better, more, and people would miss the people, the hands, the physicality, the organic, in art. Our whole world is screens now, and art generally follows a pattern of pushing against the norm. It would make more sense to call it AI imaging, but artists also use the technology for their artistic purposes (think collaging and photobashing AI created images), and "art" is not easily defined, so that goes out the window too. So, if someone wants to KNOW for certain that work came from a HUMAN (if that is something they value), the only way is if it is traditional work, and for some people, this aspect is of great importance. (However, I'm sure it will only be a matter of time before we have collage making sculpting androids and robots...I've already seen paint robots.) I have a hard time even appreciating digital art much these days (which I'm pretty sad about), because, I know there are so many tools and tricks that can just basically make art for you, and then you can sign it, because you made the decisions. You don't even need AI. A lot of concept art that looks so amazing and realistic and otherworldy is just photobashing with a lot of filters, stamp tools, and watermark removing...
if artists don't want to be in a dataset they shouldn't be they should delete the old modles and start again with a clean database.
You'll be surprised to find that even if and when this happens, the "threat" for the artists will be the same, their discontent will be the same, but they will have nothing to complain about. I'd say they shouldn't even complain about it now. They're framing their complaints in a way that makes it hard to sympathize and in one way or another they will lose even if they win a law suit or 2.
@@Marian87
what ?
what discount?
and no if someone doesn't what Ai to be trained on there art face voice or written work that the Ai companys should respect that it is simple as it can get.
This will not end well. Making art is a human endeavor and a scared act.
If there is no hardcopy with authenticity, it is not original. People will steal and bootleg anything for personal gain, fame or revenge. Maybe artist should go back to Paper and charge a fee to see it online.
Make the data collecting companies pay for all images their ai uses, simple . Pay each artist per image used. Have it federally regulated just like the music copyright industry.
Make each download a fee and tracked like taxes.
or ban any AI generated works until they are trained on a dataset with no copyrighted works, nevertheless public trained models are outlawed
people who draw in pauper got there art stolen too ... kim jung gi was used to train a ai 1 day after his death master kim was a traditional artist and they made a Ai model to copy his work...
tbh i don't think artists will except the offer artists like posting there art online so there fans and other artists can see.
if real artists stopped posting online Ai art will fill that spot .
The AI in itself doesn't include the dataset. The dataset is only used for the training, and the final model is about 1/1 000th of the size of the dataset.
The machine learning algorithm learns concepts. Nobody can copyright what a dog is - when the ML algorithm go through images it associates texts with pixels. For example when an image is tagged as dog, it will think all of the image is a "dog", even the sky or the grass in the image. It needs more pictures of dogs in different locations or poses to learn what a dog is, and what a dog isn't. Since some pixels seem to change when a dog is mentioned and others aren't, the AI slowly learns the difference between a dog and skies or grass.
Nobody owns the concept of a dog, or a sky or grass. Nobody owns how the style of how to draw that. It's ridiculous to fight for a copyright for concepts or styles.
13:00
"Nothing was regulated"
That is completely false. LAION 5B is located in Germany, which is under EU jurisdiction. The regulation for it is in Article 2 (1) and Article 3 of the directive 790/2019 form 2019 and Germany's copyright law Section 23 (3), Section 44b (2) and Section 60c from 2021.
All recent laws.
Except LAION itself doesn't break any laws because it's just a image dataset for "investigation" purposes. The problem is that Stability AI, the founder of LAION and Stable Diffusion, took LAION dataset to train different diffusion models (besides the open source Stable diffusion) to make profit with Tokens.
Considering LAION uses Fair use, it's uses and applications shouldn't be breaking four factors the fair use, but it actually does, it breaks the Fourth factor:
The Effect of the Use on the Potential Market for or Value of the Work.
And not only that, diffusion processes tend to copy in certain situations (This said by Stability Ai when launched it's own music model) , violating copyright laws, since: 1- Copyright author didn't authorize the partial use of it's work. 2- Makes profit of it.
@@jorgedavidruagarcia6838 Stability provides a software that is trained according to section 23 (3) of the German copyright law. This legal. A software that is in itself not in competition with an artist as one is telling a software the other art pieces as their product.
If the output of the software is in competition with a NAMED product of the plaintiff they could argue that fair use does not apply. At this moment there don't exist outputs that would infringe on the plaintiff's copyright. They didn't even name any, so they lack standing. Also when they find something it needs to be directed at the user of the software, since the software overwhelmingly is used for the creation of new artwork not comparable to Napster which was used overwhelmingly to copy.
@@anonnymous7009 The problem is that this unregulated software doesn't compete with an specific painting/work, it replaces a whole service given by the plaintiffs.
Also, in the process of creating an AI painting with a determinated style, the software needs to be provided with an imput were the consumer gives specific instructions and names, names that didn't consent to be in this program in the first place.
Law says you can't copyright an style, defending human creativity. But in this case the style is not the main concern, since you can't really fight for it, the main concern is the unfair use of the trained datasets and their names to create a replacement for their services with no previous authorization or compensation.
Also, Section 23 only applies to LAION itself, not to the whole Stability Ai family and other AI Image generators.
*Section 23:*
(3) Subsections (1) and (2) do not apply to alterations of a work for exclusively technical reasons in the case of uses in accordance with section *44b (2), 60d (1), section 60e (1) and section 60f (2).*
*44b* (2) It is permitted to reproduce lawfully accessible works in order to carry out text and data mining. Copies are to be deleted when they are no longer needed to carry out text and data mining.
*60d* (1) It is permitted to make reproductions to carry out text and data mining (section 44b (1) and (2) sentence 1) for scientific research purposes in accordance with the following provisions.
(2) Research organisations are authorised to make reproductions. ʻResearch organisationsʼ means universities, research institutes and other establishments conducting scientific research if they
1. *pursue non-commercial purposes,*
2. *reinvest all their profits in scientific research or*
3. act in the public interest based on a state-approved mandate.
*The authorisation under sentence 1 does not extend to research organisations cooperating with a private enterprise which exerts a certain degree of influence on the research organisation and has preferential access to the findings of its scientific research.*
(1) The rightholder’s consent is not required where reproduction of the code or translation of its form within the meaning of section 69c nos. 1 and 2 is indispensable to obtain the information necessary to achieve the interoperability of an independently created computer program with other programs, provided that the following conditions are met:
1. the acts are performed by the licensee or by another person authorised to use a copy of a program or on their behalf by a person empowered to do so;
2. the information necessary to achieve interoperability has not previously been made readily available to the persons referred to in no. 1;
3. the acts are confined to those parts of the original program which are necessary to achieve interoperability.
(2) *Information obtained through acts as referred to in subsection (1) may not be*
1. *used for purposes other than to achieve the interoperability of the independently created program,*
2. *given to third parties, except when necessary for the interoperability of the independently created program,*
3. *used for the development, production or marketing of a computer program which is substantially similar in its expression or for any other acts which infringe copyright.*
(3) *Subsections (1) and (2) are to be interpreted such that their application neither impairs the normal exploitation of the work nor unreasonably impairs the rightholder’s legitimate interests.*
Again, LAION itself is harmless, BUT Stability and other Ai generators are certainly violating copyright laws.
@@jorgedavidruagarcia6838
Stability does not influence or get preferential access. LAION releases its research open source and is open to investigations of influence at any point in time.
> it replaces a whole service given by the plaintiffs.
A service is not protected under copyright. Only SPECIFIC expressions of ideas maybe copyrighted. Nobody has the right to sue a company for creating a product that is disrupting a market - see google vs authors. Plaintiffs have to provide a specific expression or they do not have a standing to sue against a violation of their copyright.
Second point: Stability does not provide pictures. It provides a software that can be changed by the user - as seen in the hundreds of unique models that are based on Stability's product. After that, a user has to type in a prompt that may use an artist's name. That output has to be proven to directly compete with a plaintiff's copyright and lead to a substantial impact on their profitability.
Plaintiff does not own the right to create art in a certain style though. Plaintiff can't demand that no computer is allowed to compete in the general art market. And the plaintiff must prove that a specific output is hurting a specific product of theirs also based on visual similarity.
There is no standing in this case because the plaintiffs do not provide anything for the court to analyze concerning copyright law as they do not provide any of their expressions but vague assertions of not wanting to compete with automated output and the elimination of not visually similar competition in the art market is not in the spirit of copyright law.
@@anonnymous7009
1-
> see google vs authors
Court determinated Google didn't infringe fair use because it was only giving snippets. And it didn't affected Author's source of income.
It's a whole different case and it can't be compared to this case. That case is Transformative (Google vs Authors) meanwhile Stability Ai seems more like a derivative one.
2-
>*It provides a software that can be changed by the user* - as seen in the hundreds of unique models that are based on Stability's product. After that, a user has to type in a prompt that may use an artist's name. That output has to be proven to directly compete with a plaintiff's copyright and lead to a substantial impact on their profitability.
This only proves this program might potentially harm IP, people and generate illegal content, Stability should've made limitations before launching SD. This can't be comparable to a camera, since camera ONLY captures what you're "seeing" and envolves the minimum creative process and authorship to link the creation to it's creator. In SD case, the mayority of the work is made by the program Itself, so the images created by SD should be directly linked to SD.
3-
>And the plaintiff must prove that a specific output is hurting a specific product of theirs also based on visual similarity.
Since AI depends entirely of it's database
this might be easy to prove, *Stable Diff gives the facility to replicate or fake the work or an IP of an artist or company by directly feeding it with images that might be copyrighted.* This could lead potentially into Scams, copycats and lose of an income. *Easily proven by all the SD models created based on SD code that Stability AI deliberately released without any restrictions.*
Again, It also violates 60d (2) which is directly linked to Section 23 by:
(2) Information obtained through acts as referred to in subsection (1) may not be
1. used for purposes other than to achieve the interoperability of the independently created program,
2. given to third parties, except when necessary for the interoperability of the independently created program,
*3. used for the development, production or marketing of a computer program which is substantially similar in its expression or for any other acts which infringe copyright.*
*(3) Subsections (1) and (2) are to be interpreted such that their application neither impairs the normal exploitation of the work nor unreasonably impairs the rightholder’s legitimate interests.*