Commons:Deletion requests/File:A Shiba Inu dog wearing a beret and black turtleneck DALLE2.jpg

From Wikimedia Commons, the free media repository
Jump to navigation Jump to search
This deletion discussion is now closed. Please do not make any edits to this archive. You can read the deletion policy or ask a question at the Village pump. If the circumstances surrounding this file have changed in a notable manner, you may re-nominate this file or ask for it to be undeleted.

https://labs.openai.com/policies/content-policy explicitly forbids commercial use of DALL-E 2 images. This is incompatible with the licensing policy. TilmannR (talk) 20:01, 24 April 2022 (UTC)[reply]

 Delete It is assuredly in breach of the licencing you state, but I think instead of the "Use for non-commercial purposes only." that you state, I believe it misses because it fails to "Disclose the role of AI" plus "Respect the rights of others - Do not upload images to which you do not hold appropriate usage rights". At the very least COM:VRT should be used to regularise the right to upload the picture, so COM:PCP applies.
Additionally, it comes from https://twitter.com/gdb/status/1511718170804908037 where the DALL-E source is stated clearly, but it could be considered to be a copyvio. 🇺🇦 Timtrent 🇺🇦 talk to me 🇺🇦 20:57, 24 April 2022 (UTC)[reply]
I think the role of AI was properly disclosed: DALL·E 2 is mentioned both in the description and as the author. (Although technically existing law does not permit a non-human to be an author. The actual author(s) would be the OpenAI employee(s) who created the input prompt and selected this particular image from the AI's outputs.) TilmannR (talk) 22:21, 24 April 2022 (UTC)[reply]
 Keep The author is an artificial intelligence, which means that the work cannot be copyrighted. The content policy is not legally binding on copyright grounds because the work is in the public domain. From a legal perspective, the content policy's stance on commercial use is more of a request than a license requirement, since the writers of the content policy are not the copyright owners. Di (they-them) (talk) 23:24, 24 April 2022 (UTC)[reply]
@Di (they-them): I believe your assessment is based on a misunderstanding. The U.S. Copyright Office and the U.S. Patent and Trademark Office did not say that all outputs of AI are automatically in the public domain. They said that a work without any creative contribution from a human actor cannot be copyrighted. But since DALL-E's image generation depends on natural language prompts, we can assume that a human did contribute.
Consider the following analogy: A camera is not considered the author of the images it produces. But when a person does the creative work of pressing the shutter button, they typically own the copyright to the resulting image.
Generative AI is a tool much like a camera. The AI itself is not considered an author, because it is not a person. But I can think of no reason why the writer of the input prompt would have fewer rights to the generated image than if it had been taken as a photograph. TilmannR (talk) 00:41, 25 April 2022 (UTC)[reply]
Just to play advocate's devil, there is at least one case (this one) where the prompt was generated by GPT-3... What would be the ruling, then ? --Dfeldmann (talk) 20:48, 13 May 2022 (UTC)[reply]
@Dfeldmann: The bot that generated the prompt apparently generates an infinite stream of random tweets. I don't believe any human claims authorship of that text. And the bot itself is not a person and therefore also not an author. But the person, who selected that specific tweet, felt like it was suitable for feeding into DALL-E 2 and decided that the resulting image was worth sharing, might have done something sufficiently creative to be considered the author of the image.
Extending the camera analogy: If you photograph a mountain, you own the copyright to the image, even if you didn't create the mountain.
I'm neither a judge nor an admin, so none of this is an actual "ruling", but I hope this answers your question. TilmannR (talk) 21:48, 13 May 2022 (UTC)[reply]
@TilmannR: It cannot be said that a human created this artwork. At the very most, a human gave an AI an idea, but ideas and concepts are not protected by copyright. A human had no input on this work of art besides the text prompt, whereas with a camera a human actually has to make creative decisions and take the photo. I think a more apt comparison would be commissioning an artist. The commissioner does not hold the copyright, the creator of the art does. In this case the creator of the art is an AI, meaning that there is no copyright protection. Di (they-them) (talk) 11:08, 24 May 2022 (UTC)[reply]
@Di (they-them):
You mentioned that the idea for this image is not protected by copyright. Note that the idea to take a photo of something is also not copyrightable, but that has no effect on the copyright of the resulting photo.
The artist analogy is weaker than the camera analogy, because DALL-E 2 is a tool and lacks personhood. E.g. "In this case the creator of the art is an AI" is either false or irrelevant, depending on the definition of "creator".
  • I assume that entering a prompt and selecting one of DALL-E's various output images is approximately as creative as pointing a camera at something and pushing the shutter button.
  • I assume that pointing a camera at something and pushing the shutter button are all the "creative decisions" needed for an original photo (e.g. of a clothed dog) to be copyrightable.
Is either of these assumptions wrong? TilmannR (talk) 03:01, 25 May 2022 (UTC)[reply]
The image was not created by a person, thus there is no copyright. You cannot copyright an idea, only an expression of an idea, and unlike in the case of a camera, the creative expression in this case is entirely executed by software, not by a person. As an analogy, if I was a painting instructor and gave everyone in my class the prompt "A Shiba Inu dog wearing a beret and black turtleneck", would you expect me to own the copyrights for all the paintings created by my students? Of course not. Nosferattus (talk) 22:40, 4 June 2022 (UTC)[reply]
@Nosferattus: But why wouldn't you own the copyrights to your students' paintings? Is it because the paintings were created by brushes, which aren't human and therefore incapable of claiming authorship? No. It's because your students are people with strong authorship claims that make your contribution de minimis. But if your classroom contained any recording devices, then you own the performance rights to the recordings of you saying "A Shiba Inu dog wearing a beret and black turtleneck" despite you doing exactly the same amount of expressive work in both cases. TilmannR (talk) 12:56, 5 June 2022 (UTC)[reply]
@TilmannR: Because I didn't create the paintings, just as the authors of DALL·E 2 didn't create the Shibu Inu image. Let's take your argument to the logical extreme. If I wrote a program to automatically generate all possible musical melodies (which someone already did), would I then be entitled to sue the authors of every new musical work for copyright infringement? Nosferattus (talk) 14:46, 5 June 2022 (UTC)[reply]
  • I'd like to emphasize your use of the word "automatically". Image generation via DALL-E is not an automatic process, but is conditioned on the input of a prompt.
  • Suing the authors of musical works for using a melody contained in that 1202 GB file is like suing authors of literary works for using words that occur in a dictionary or suing artists for using colors that also occur in other paintings.
  • We're not even dealing with a case where someone generated their own image of a Shiba Inu dog wearing a beret and black turtleneck and was sued by OpenAI. This is an exact copy of OpenAI's image. TilmannR (talk) 15:41, 5 June 2022 (UTC)[reply]
 Keep - No human author, thus no copyright protection. (Creating a prompt is not authoring a work.) Nosferattus (talk) 22:32, 4 June 2022 (UTC)[reply]
Please note that the people at OpenAI did not merely create the prompt, but they went through all the steps necessary to convert the prompt into a concrete image, which is what makes them the authors of this work. (Copy-paste from Commons:Deletion requests/File:DALL-E sample.png.) TilmannR (talk) 12:57, 5 June 2022 (UTC)[reply]
See my reply at Commons:Deletion requests/File:DALL-E sample.png. This is why it's not helpful to create multiple deletion discussions about the same issue at the same time. Nosferattus (talk) 14:48, 5 June 2022 (UTC)[reply]

Kept: no valid reason for deletion. Public domain as not created by a human. The teacher / students analogy is convincing. --Yann (talk) 19:47, 5 June 2022 (UTC)[reply]