Feilin v. Baidu: Beijing Internet Court tackles protection of AI/software-generated work and holds that copyright only vests in works by human authors
Whether an AI-generated work can be protected by copyright is currently one of the most heatedly discussed issues almost everywhere in the world. Although the discussion in academia and practice is prevalent, to the best knowledge of this InternKat, there are still very few cases decided by courts on this matter as of today.
Just a few months ago, the Beijing Internet Court in China issued its decision (here in Chinese) in Feilin v Baidu, ruling on whether a (partly) AI-generated work could be protected by copyright.
This post only discusses the part of the judgment devoted to the issue of copyright protection, though there were also other issues addressed therein.
The work at issue
 |
| So, how did the court come to this conclusion? |
The work in relation to which copyright subsistence was uncertain was a report titled ‘Analytical Report on the Judicial Big Data in the Film and Entertainment industry: Film Industry in Beijing’ (here, in Chinese), which had been created/generated with/by software named Wolters Kluwer China Law & Reference (here). As explained on the relevant website, this software is a legal information query tool. It covers a wide range of contents, including bilingual legislation, judicial cases, commonly used legal documents and commentaries.
As described by the plaintiff, the report consisted of drawings and words. The drawings were first generated by the software and then modified and coloured manually by the plaintiff's staff. As for the written part, it was created by the plaintiff's team. The creation process was as follows. First, the teams used the software to search the judgments following the criteria set by the plaintiff. Then, the team examined the search results and removed irrelevant court decisions. After this, the team created the report by analysing the statistics.
The defendant pointed out that the software generated the whole report. The data in the report was not investigated, sought or collected by the plaintiff. The drawings in the report were not created by the plaintiff.
To understand the creation/generation of the disputed report better, at the hearing, the court conducted an investigation. The plaintiff's attorney did first login in the software and then perform the following instructions set out by the defendant: (1) setting the search condition: key words 'films'; trial courts: 'Beijing', Date of Judgement: '1.1.1995 to 31.12.2017' ; (2) click 'visualisation' button after the search results being made available. After this process, a data report was generated (the ‘software-generated report’).
By comparing this report to the disputed report, some drawings and words in the disputed report were not the same as or similar to those in the software-generated report. For instance, the pictures and words concerning the analysis of the trend of the number of cases were not identical or similar with regard to the type of drawings and the content of word analysis.
The defendant argued that the continuously updated data in the software caused these differences. The structure, the types of drawings, the content on the distribution of the cause of actions and the trial court, as well as the pattern to present were similar or identical in both reports.
The Judgment
The Beijing Internet Court held that the disputed report had not been generated by the 'visualisation' function of the software because of the difference between the software-generated and the disputed content. Besides, the disputed report was created by the plaintiff's team and sufficiently original. Thus, the disputed report was found to be a work protected by copyright under Chinese law.
In addition to the above, the court also expressed its view as to whether the report automatically generated by the 'visualisation' function of the software could be also protected.
The originality of the report
The court first confirmed that the content of the report was original. The content of the report concerns the judicial analysis in the film and entertainment industry. The words in the report meet the formal requirement for them to constitute a 'literal work'. Additionally, the content of the report reflected the choice, judgement and analysis of the relevant data.
A production created by a natural person is a necessary condition for a 'work' in copyright law
The court then held that the production created by a natural person should be the necessary condition for this production to qualify as a work under copyright. The court noted that the originality of the content itself would not be a sufficient condition for that content to qualify as a work in a copyright sense. Under the existing law, a work shall be created by a natural person. Although the content generated with intelligent software is increasingly similar to work produced by a natural person in terms of content, form, and expression thereof, it is not possible (or desirable) to break the fundamental tenets of copyright law, including the notion of authorship.
A software developer or a software user is not an author of AI-generated work
The court further analysed how natural persons participated in the generation of the report. In the process of production of the report, two stages involved the participation of natural persons: the software development phase and the use of the software. A software developer does not use keywords to conduct searches. Also, the report did not pass on any original expression of the developer. Therefore, the software developer does not create the report. Although a software user chooses the keywords for searching, the report is generated automatically by the 'visualisation' function and does not pass on the original expression of idea or feeling of the software user. Thus, the court held that neither a software developer nor a user could be regarded as the author of the report.
Software shall not be an author of a work
The court acknowledged that the report was to some extent 'created' / 'generated' by the software in the way of a combination of inputted keywords and algorithm, rules and templates. In this case, the report might be originalbut would not be a copyright 'work'.
The issue of indicating name in the report
Neither the software developer nor the software user can impress their names on the work because they did not create the work. Instead, the report should include a sign of automatic generation explicitly stating that the report was generated automatically by the software to protect the public's right to know, maintaining social honesty and trust, and favouring cultural dissemination.
No copyright protection but….
The court held that the report could not be protected as 'work'. However, this would not mean that the report would be in the public domain and the public could freely use it, considering both the input of the software developer and the user.
As for the developer, his/her investment could be realised by charging for the use of the software. The payment for the software rewards his/her investment in developing software. Besides, the developer would have no motivation to distribute the work because the report is generated to the need and with the search of the users, not the developer.
The software user invests in the report by paying for the use of the software. He/she has the motivation to use and distribute the report. Without certain protection, the number of software users could decrease, and the users would not be willing to further distribute the report, which in turn could hinder cultural distribution. Based on this analysis, although the software user could not be named as the author of the report, they could share the interests from the report in a reasonable way in order to protect their legal interests and to protect the public's right to know.
In conclusion, the judgment spelled out the following criteria:
1. Under existing Chinese copyright law, an AI-generated production could not be regarded as a 'work' in a copyright sense, irrespective of whether it is original; only a production by a natural person could be regarded as a ‘work’.
2. AI software could not be considered an author; a notation should be added indicating that the production at issue was created/generated by AI software.
3. A software developer or a software user could not be the author because they do not create or generate production.
4. Although the AI-generated work cannot be protected by copyright, the investment in the generation of the production is still deserving of some sort of protection.
5. The investment of a software developer is already rewarded with payment from the use of the software. Therefore, there is no need to protect their interests.
6 The software user shall be given some protection so that they are motivated to distribute and generate the production, and they are willing to pay for the use of software.
Comment
Several Chinese media reported on this case calling it the first case concerning an AI-generated work in China. As seen, however, the disputed report was actually created by the plaintiff with the use of AI software. Despite this, this was indeed the first time that a Chinese court explicitly expressed its view regarding copyright protection of AI-generated content.
Even if the court denied the copyright protection could be available to purely AI-generated works, it also accepted that some sort of protection should be nonetheless available, on consideration that: (1) a work of this kind represents the investment of the software developer or the software user; (2) the content has the value of dissemination; (3) if there is no protection for the investment, the distribution of the content would be negatively affected. Regrettably, the court did not indicate what kind of protection should be granted. A neighbouring rights regime? A sui generis scheme? The judgment is silent in this regard.
