You Mingyang, CEO and co-founder of Rayark Games, said at the AI Cantors Annual Meeting (the 0th AIGC Annual Meeting) held on December 22, 2022 that the application of AI tools is bound to cause chaos, and all past morals Concepts, legal norms, and intellectual property concepts will all be reinterpreted.
Looking back at the copyright disputes about AI drawing tools in 2022:
|February||━━||━┥||The U.S. Copyright Office rejected the copyright registration of the AI-generated painting “A Recent Entrance to Paradise” because AI cannot be the copyright owner, and humans are too little involved in the creation and are not creators.|
|September||━━||━┥||The world’s largest gallery platform, Getty Images, has removed AI-generated images from the shelves due to “the copyright of AI-generated content is subject to discussion”, prohibiting the upload and sale of AI-generated images; Shutterstock and other galleries have also followed up.|
|September||━━||━┥||The U.S. Copyright Office granted copyright to Midjourney’s collaborative comic book Zarya of the Dawn.|
|November||━━||━┥||The U.S. Copyright Office has reaffirmed the authorship of Zarya of the Dawn, requesting details of the creative process to demonstrate the high degree of human involvement in its creation.|
|December||━━||━┥||The homepage of the digital art platform ArtStation was washed with pictures of “NO TO AI GENERATED IMAGES”. Artists protested against AI using unauthorized images for machine learning.|
By this year (2023), Getty Images sued Stable Diffusion, an AI image generation tool, in January. On the other hand, Shutterstock sent a letter to users in January, through cooperation with Dall E2, allowing users to experience its AI image generation tool first.
Wu Zhixun, a researcher at the Center for Artificial Intelligence and Digital Education at National Chengchi University, said that AI is a tool, but how to make good use of this new technology? The issues of AI and ethics, AI and infringement will need to be faced sooner or later.
Infringement Controversy Over AI-Generated Images
Since December, artists have uploaded a large number of “NO TO AI GENERATED IMAGES” pictures on the digital art platform ArtStation, protesting AI’s use of unauthorized images for machine learning; On the “Voice of the Painting World”, there are also many articles submitted to express dissatisfaction with the behavior of imitation through AI drawing tools.
According to the author’s observation, in the novice channel of Midjourney, it is common to see users use the names of “Ghibli”, “Hayao Miyazaki”, “Makoto Xinkai” and other names as prompts for image style calculations. What is even more amazing is that in Nijijourney, The user can generate an image very similar to the movie poster of Makoto Shinkai’s “Your Name” without typing “Makoto Shinkai” if the user prompts “a couple, starry sky, meteor, movie poster”.
In the drawing circle, some painters often say that they have become victims of AI. AI users upload the painter’s paintings as “reference pictures” to the AI drawing system for imitation and extended creation without permission. In addition, some fans also said that the use of AI drawing tools is not strict enough, which caused the painters to remove their works from the Internet platform in order to maintain the value of creation.
Chen Bingxun, an associate professor at the Institute of Technology Management and Intellectual Property at National Chengchi University, explained that in judging whether a work is infringing in law, it mainly checks “substantial similarity” and “contact”. If it looks similar, it meets the requirements of “substantial similarity”. What constitutes “contact” needs to be determined on a case-by-case basis, and will be the key to discussing whether AI works infringe in the future.
Chen Bingxun believes that in the current judicial practice, as long as the AI user does not directly look at the original description and generate images, the requirement of “contact” is not easy to be established; and what can continue to be discussed in the legal judgment in the future is that, If the AI tool itself has direct contact with the original work and the AI user has indirect contact, is there any infringement?
And if the person who provided the original work information to AI also uses AI to generate derivative works of the original work, it constitutes infringement. Chen Bingxun said: “Hey, the data of these original works have been in contact with themselves, and then use AI as a tool to create This is a typical state of 100% copyright infringement.”
Feng Zhenyu, a professor at the Institute of Technology Management and Intellectual Property at National Chengchi University, said that if the AI calculation model grabs data from the Internet without authorization, violates anti-theft measures and digital rights management, there will be suspicion of infringement.
Feng Zhenyu added that most large-scale AI calculation models such as Diffusion use open source data or data authorized by large databases for machine learning; According to the reporter’s year-end meeting (the 0th AIGC annual meeting), a recently completed study that has not yet been peer-reviewed (Peer Review) claims that the Diffusion model actually has pirated images.
Regarding the possible infringement of machine learning, Feng Zhenyu raised hidden concerns. If the artist wants to claim that the AI system has misappropriated the work without authorization, he will face the burden of proof. As Midjourney player Li Yizhi said, “The richer the content and style of the prompt, the more the AI output will deviate from the original work.” Since the AI system is trained with massive data, the final generated image may not be consistent with the original work. Similarly, it is difficult for artists to confirm and claim that their works published on the Internet have been exploited by AI.
Who owns the copyright to AI-generated images?
In addition to whether AI-generated content involves infringement, who owns the copyright of AI-generated images is also a focus of discussion.
Feng Zhenyu explained that AI is divided into two types in the world, one is “weak AI (Weak AI)”, human beings can control the creation of AI, and the works it generates are called “AI generated”; the other is “strong AI”. AI (Strong AI), humans do not participate in the process of AI generating works, and the works it generates are called “AI originated”.
The degree of human participation in creation will determine the ownership of copyright, because the purpose of copyright is to protect human creativity. The works of “strong AI” have no copyright, because the degree of human investment in their own creativity is too low to be regarded as the creator, and AI is not the object of rights protection under the current legal concept.
In February this year, the United States Copyright Office rejected AI researcher Stephen. Copyright application for Stephen Thaler’s AI-generated painting “A Recent Entrance to Paradise”. Taylor argued that the work was calculated by the “Creativity Machine” with minimal human intervention. Taylor asked the Copyright Office to grant copyright protection to AI and AI works, but the Copyright Office still insisted that the purpose of copyright is to “protect human thinking and creative expression”.
In September this year, American AI artist Chris. Kris Kashtanova’s comic work “Zarya of the Dawn” has obtained the copyright in the United States. The images of this comic work are generated by Midjourney and the text is written by Kashtanova. However, in November, the U.S. Copyright Office wanted to reaffirm the copyright and asked Kashtanova to provide details of the creative process, explaining that the creative process involved a lot of human participation.
Chen Bingxun believes that the works completed by using AI tools such as Midjourney cannot be called “creation”, because AI does not know how to create, and AI only calculates according to the model. Moreover, since human beings cannot fully grasp the whole picture of the work by inputting prompt words, users of AI tools cannot claim that this work is his creation, “This model is like asking a painter to help me draw the picture. I’m just giving a statement.”
Chen Bingxun gave an example, if you only tell AI to “draw a rabbit”, then this is not creation, because the user does not control the appearance of the final product, but it is produced by AI; the user must describe it in detail and in detail For the configuration of the rabbit’s facial features, ear shape, etc., AI is only used as a drawing tool to draw according to the instructions, so that it can be regarded as the user’s creation.
In the actual use experience of Midjourney, the influence of each prompt word is not the same, even if the results obtained by using the same set of prompt words are not the same, the AI art creator Bird Nest once described: “The creation of Midjourney is sometimes a matter of luck, Luck, that’s why we call ourselves “summoners”, and the results of the summons are somewhat random.”
However, Feng Zhenyu believes that the current AI drawing tools are still “weak AI”. The process of human inputting prompt words is enough to prove that human beings follow their own thinking and ideas to participate in the creation. Therefore, the works generated by using AI drawing tools should be protected by copyright. The copyright holder is the creator of the work.
Feng Zhenyu took Jason. Take the creation process of Jason Allen’s midjourney painting “Théâtre D’opéra Spatial” as an example. Allen input prompts one by one to produce more than 900 images, and then manually selected three images, and then Using other software to enhance the image quality is the final product. “You can’t say that there is no human intervention in this process. This is self-deception. AI is still just a creative tool, which is no different from using computer software as a tool.”
Bird’s Nest also said: “Our world is not yet ruled by AI. It is humans who make judgments and decisions, and the value of creation is still on humans.” Bird’s Nest, designer/AI image communicator Lin Zixiang all use “director theory” Describe the process of using AI tools to express human creativity. The director communicates with actors and professionals according to his own ideas, and then these people complete the director’s expected screen; AI tool users issue prompt word instructions to AI, and AI performs calculations and produces results through big data. Directors can share final results with actors and professionals, and humans can share collaborative works with AI.
In addition to national laws defining the ownership of copyright, the terms of service of AI drawing tools will also regulate and restrict copyright and intellectual property rights. Hou Yixiu mentioned that the images generated by Stable Diffusion are owned by the public and do not belong to the creators. The images generated by Dalle-E are owned by its company OpenAI. The terms of Midjourney are more complicated, but in principle users own the copyright of the images.
Hou Yixiu further pointed out the current unfinished business: “When the works of human beings and the works of human and AI collaboration are subject to different copyright regulations, how to judge whether it is a human creation or a creation involving AI? This will be a problem in the future. A big problem that has to be solved.”