Hive and Generative AI
Although in the #Hive blockchain, the debate between legitimacy in posting an AI work is still open, (because it would simply an attempt at #fraud: #copyright appropriation), some simple concepts should be kept in mind:
AI is not a tool like a pencil or a mouse.
Many people make this mistake: equating AI with a tool for creating #art.
So creating art with a painting software is equivalent to creating it with AI.
The mouse, the painting software, the microprocessor, the chisel, the pencil, the paintbrush....they are all "passive" tools.
The creator must command the tool with the ability/skill to generate art or text.
Claiming its own "textual prompt" to attribute authorship of the generated image is not enough.
When posting an AI generated image, a strictly necessary condition would be to always specify that the image is produced by AI, because, as stated in the OpenAI Terms of Use:
“(v) represent that output from the Services was human-generated when it is not or otherwise violate our Usage Policies;”
( https://openai.com/policies/terms-of-use )
In fact, generative AI, following prompts, introduces elements independently (such as perspectives, details, color gradients, visual effects...) as a result based on the prompt's input.
This is why it’s called "generative AI", because it compiles data using database sources
(terabytes of textual classification instructions).
This means that the generated image does not belong to the human creator.
How to legitimize the use without having the rights?
One motivation used by many is “Fair use”, using the AI tool to generate new art.
"Fair use" is the option of not explicitly obtaining permission to use copyrighted material, for example, to create a new informational or derivative work under the protection of the right to expression and criticism.
In the context of text-to-image generators and the resulting creations, the question is mostly about what’s “fair use,” he explained. Under U.S. copyright law, fair use ( https://www.copyright.gov/fair-use/ ) is a “legal doctrine that promotes freedom of expression by permitting the unlicensed use of copyright-protected works in certain circumstances.”
( https://venturebeat.com/ai/who-owns-dall-e-images-legal-ai-experts-weigh-in/ )
But it is equally true that Stable Diffusion and MidJourney have allegedly used millions of images without permission, for "scraping" (collecting data for database entry), and the giant Getty Images has taken legal action to defend the infringed copyright.
( https://newsroom.gettyimages.com/en/getty-images/getty-images-statement )
So, if there is no profit, one can use the images (within certain limits) with "Fair use”.
Generative AI companies are often open source, but there are companies that use AI-generated products for advertising.
Who pays copyright to the creatives and artists? Moreover, "Fair Use" is not provided for in other nations around the world, so how is this handled internationally?
About free generative AI engines we can mention Openjourney, that uses the results of MidJourney.
“The creators of PromptHero, a collection of prompts and images from various AI models such as Midjourney, Stable Diffusion, and DALL-E 2, have released their own Stable Diffusion model on Hugging Face. According to them, this model was trained with more than 60,000 images generated by Midjourney V4.”
( https://the-decoder.com/openjourney-is-stable-diffusion-trained-with-midjourney-images/ )
“The release of Openjourney is also heating up the copyright discussion. Midjourney itself is already facing criticism for using millions of copyrighted photos and artworks to train its models without being asked. A lawsuit by Getty Images against Stability AI, the company behind Stable Diffusion, is currently underway.”
“According to Midjourney’s terms of service, paying users own all rights to the images they create. Also, according to Midjourney’s moderators on Discord, users would have to be asked for permission before their images could be turned into training material.”
( https://the-decoder.com/openjourney-is-stable-diffusion-trained-with-midjourney-images/ )
And:
“On the one hand, Midjourney uses artists’ images without asking to train and market its own model.
On the other hand, it refers to image rights when Midjourney images are used to train other AI models. Not to mention that the rights of the artists could be violated again during this second model training.”
( https://the-decoder.com/openjourney-is-stable-diffusion-trained-with-midjourney-images/ )
International law provides that a copyright can only be attributed to the direct result of humans.
In the future, the law may change adapting to the evolution of AI, but not now.
Although some opt-out services exist for artists to take their works out of the AI training model, this is a small percentage compared to, for example, the huge LAION database used for Stable Diffusion training.
Ok but...
if the AI generated image does not belong to the human creator (the "artist" who wrote the prompt), because the copyright cannot be assigned, it simply means that the image may belong to everyone.
This happens precisely for the reason that AI introduces graphical elements independently,
because no matter how detailed the textual prompt may be, the result will never be directly and strictly comparable with the prompt.
In this example, the prompt says:
"Surreal landscape of a man standing in front of a portal to another dimension."
DALL-E generates this image:
The prompt does not indicate:
- the shape of the portal and its color,
- how much blue glow,
- what kind of sky,
- the shape of the clouds and the colors,
- why the man has his legs slightly apart,
- the length of his shadow,
- the ground texture,
- the slightly tilted framing (camera roll)…
These are all choices not decided by the "artist".
Again...
In this example, the suggestion is “What do you want to generate?”
The prompt says:
"emptiness"
Craiyon (DALL-E mini) generates these images:
It seems pretty clear that there is an "interpretation" not determined by the user.
The choice to include a human figure is totally arbitrary, as are all set designs and light and color choices.
So, no copyright can be attributed to the “artist”. No ownership.
If the "artist" cannot copyright the AI-produced image, it means that image can be used by anyone.
In Hive's blockchain, interaction with users can generate rewards by curation but the “fraudulent” method would lie precisely in the fact that the artificial image is not really produced by the "artist".
This fact goes against Hive's community standards, which require that the external source always be cited.
Specifying that the source is an AI means that the product is not human in nature and therefore, by law, cannot be assigned a copyright.
“As a result of the human authorship standard, “under U.S. current law, an AI-created work is likely either (1) a public domain work immediately upon creation and without a copyright owner capable of asserting rights or (2) a derivative work of the materials the AI tool was exposed to during training,” Esquenet continues. “Who owns the rights in such a derivative would likely be dependent on various issues, including where the dataset for training the AI tool originated, who, if anyone, owns the training dataset (or its individual components), and the level of similarity between any particular work in the training set and the AI work.”
Although users in Hive are free to "rate" and "curate" an AI image, it’s equally true that in the absence of a copyright, anyone can take that same image and publish it in their own post (and it would not technically be a "repost/reblog"), so anyone could get rewards posting the "work" of others, and by “work” is meant the time (and credits/money) spent on "making" that image.
In conclusion,
if by law those who produce art or text with generative AI cannot claim copyright, the only way to do so is to go to court and claim it, showing that there has been a substantial change to the final product.
It means, for example, in the art field, to edit and do postproduction to the image, so you can claim your own "human input".
Or, if it is an AI-generated article, it means showing that you have done editorial work to supplement or revise it.
It’s not enough to change a few shades of color (filter/color correction) or change a few words in a paragraph.
The changes have to be substantial, but there is probably no specific legislation on this, at the moment.
This does not change the underlying concept.
What is created with AI has no international regulation that guarantees authorship of the work.
In Hive, this is against community standards.
Thank you for your time.