China-Singapore Jingwei, December 2 Topic: The copyright of AI-generated content may incentivize large-scale model manufacturers.
Author: Zhang Xin, Director of the Digital Economy and Legal Innovation Research Center of the University of International Economics.
Recently, the Beijing Internet Court made a first-instance judgment in the case of copyright infringement dispute over artificial intelligence generation ** (AI painting**), and the defendant must apologize and compensate the plaintiff for economic losses. This has sparked a lively discussion about the relationship between AI-generated content and copyright. This case recognizes for the first time that AI painting model users have the right to generate, will this benefit upstream model manufacturers?What impact will it have on the large model industry?
Large model manufacturers may be largely incentivized. If the plaintiff in this case enjoys the copyright of the AI large model generated in the case, the content generated by the AI will be exclusive and exclusive to a certain extent, and those original human painters who contribute the training data of the AI large model should also be protected by the corresponding rights and interests.
For many art creators, if the content generated by large-scale model tools is not protected by copyright law, it will deprive them of the opportunity to subsequently authorize and defend their rights, and thus weaken their motivation to use these tools. From the perspective of the industry chain, once the user's willingness to use is reduced, it will lead them to no longer purchase related large-scale model services, which in turn will reduce the enthusiasm of large-scale model manufacturers to invest in research and development.
In addition, the large model discussed in this case, Stable Diffusion, adopts the open source model, but the number of enterprises that can actively open source the large model in China is still relatively limited. For large models that are not open sourced, although users are given the copyright to use the content generated by the model, this may not be entirely beneficial to large model manufacturers. In particular, the service agreement of some enterprises stipulates that the intellectual property rights of the generated content belong to the enterprise and have nothing to do with the user. Such provisions may lead to conflicts and disputes between the rights and interests of users and manufacturers.
In the context of the open source model, because the data collection is publicly available, vendors usually do not claim intellectual property rights for AI creations. However, in the non-open source model, large model vendors will invest huge sums of money to purchase underlying databases, which cover various types of data such as corpus and sound. If the copyright of the content generated by these databases is vested in the customer, this will adversely affect the interests of the manufacturer.
In the future, we should face the industrial chain and pay attention to the rights and interests of multiple subjects. Because in the industrial chain of AIGC, there are primitive human painters who provide training data samples, designers who use prompts to generate AI paintings using AI large models, and deployers who provide AIGC services to users. In the entire industrial chain, what is needed is a complete set of systematic rights and interests protection system.
We should also note that the "black box" of AI models does bring difficulties to determine copyright ownership. Training data plays a crucial role in the generation of AI paintings, but its specific contribution cannot be accurately quantified and traced. This makes determining the ownership of copyright extremely complicated. First, from a technical point of view, due to the complexity of the model, we can't directly understand the specific contribution of each training data point to the resulting work. Secondly, from the perspective of the industry chain, the generation of AI paintings involves multiple subjects, including data providers, model developers, users, etc. Each subject has a role to play in the process, so we need to take into account the contributions and interests of all parties in the attribution of copyright. In order to solve this problem, we need to start from the characteristics and technical mechanism of the entire industrial chain, conduct in-depth research and explore a set of rights and interests protection rules that are compatible with and adapt to multiple subjects in the industrial chain. This may require a concerted effort by the legal, technological, and industrial communities to ensure that all parties get their due rewards in the process of generating and utilizing AI paintings. (Zhongxin Jingwei app).
This article is selected and edited by the Sino-Singapore Jingwei Research Institute, and the works produced by the selection are all rights reserved, and no unit or individual may use it in any other way without written authorization. The views involved in the selected content only represent those of the original author and do not necessarily represent the views of Sino-Singapore Jingwei.
Editor in charge: Sun Qingyang.
For more exciting content, please pay attention to the official WeChat of JWVIEW***