We’re pleased to bring to you a guest post by Bhavik Shukla and Hatim Hussain, discussing authorship in AI-related works under the Indian copyright law, with specific reference to the judgment in Shenzen Tencent v. Shanghai Yinxun Shenzen passed by a Chinese court in December last year. Bhavik is a recent graduate from NLIU, Bhopal and is currently working with the IP team at Khaitan & Co. Bhavik’s previous posts on the blog can be viewed here, here, here, here, here and here. Hatim is a recent graduate from Gujarat National Law University, currently pursuing Masters at Tsinghua University on Schwarzman Fellowship.
Shenzen Tencent v. Shanghai Yinxun: AI Authors, Copyright and Some (Hard) Lessons for India
Bhavik Shukla & Hatim Hussain
In December 2019, the Chinese court in Shenzen through its judgment in the case of Shenzhen Tencent v. Shanghai Yinxun (‘Tencent case’) held that an AI-created work would be a literary work eligible for copyright protection. This development comes at a time when the world is finding it difficult to grapple with the novel conceptions of authorship for copyright eligible material. Not only technology related matters seek to broaden the understanding of authorship (covered here, here, and here) vis-à-vis copyright, but so do intervention of non-humans (covered here). Artistic flair has not been uncommon for AIs, with next-gen robot-journalists such as Toutiao’s ‘Xiaomingbot’, Forbes’s ‘Bertie’, Bloomberg’s ‘Cyborg’ and Tencent’s ‘Dreamwriter’ publishing articles with lightning speed and accuracy. Legally-speaking, decisions premised on authorship in AI-generated works, granting (in most cases) ownership to natural persons in-the-loop, have also been no-brainers. What, then, makes the Tencent case that groundbreaking, one might ask?
For one, ascertaining ownership for output generated by advanced AI systems have become increasingly difficult, often striking at the core of the idea that all AI works require dependency on humans or legal persons. On a more comparative note, the Tencent case also highlights the impending disparities on this issue internationally, despite its increasing relevance in practice. In Australia, US (sections 306 and 313.2) and many European countries, human intervention is fundamental to accord copyright protection, whereas other jurisdictions, namely UK, Ireland, Hong Kong and New Zealand take an altogether different approach in their copyright laws, allowing protection to computer-generated works. As we shall see later, India falls within the latter set of countries, granting limited protection to computer-generated works.
The Tencent Case
The Tencent case builds on yet another interesting case, Feilin v. Baidu, delivered by the Beijing Internet Court in early 2019. In that case, the plaintiff’s copyright in the analytical report, created with the help of a fairly-automated Woltas Kluwer software, was contested by the defendant. Quite interestingly, while the Court recognised that the report met the originality requirement (on account of some human involvement in it), it did not constitute a copyrighted ‘work’. The case shut the door on the possibility of an autonomously-created literary or artistic work to be considered capable of copyright protection. While a bit extreme, Chinese courts have sought to course-correct themselves, with the Tencent judgment. While the Court in this case disregarded the copyrightability of works generated solely by software, it acknowledged that works created by such creative processes could be subject to copyright protection.
In 2019, the Plaintiff was granted a non-exclusive license of the software ‘Dreamwriter’, a software that has generated about 300,000 literary works each year. The impugned article – a financial review of the stock market – was one published using Dreamwriter – with a disclaimer stating ‘This article was automatically written by Tencent Robot Dreamwriter’ at the end. On publication of an identical article by the Defendant, the Plaintiff sued for copyright infringement and unfair competition.
The question before the Court was two-fold, first, whether the involved article was ‘original’ as per Chinese law, and second, whether copyright of the article subsisted in the Plaintiff. On the first issue, the Court held that the article was reflective of the selection, analysis and judgment of the stock market information and data as available at that time, and to that extent, it possessed a certain degree of ‘originality’. The Court highlighted the detailed inputs of the Plaintiff’s creative team, based on which it concluded that if the software was to be the subject of creation, it would disregard the personalised arrangement and selection of the creative team. Accordingly, the Court declared the Plaintiff to be the author, based on an interpretation of Article 11 of the Chinese copyright law which grants authorship to the entity under whose supervision and direction the work is created.
In fact, this notion of the ‘originality’ requirement, as being a solely ‘human’ conception is central to the question of determining authorship in AI-works. In this respect, the Tencent case props up a fitting example of acknowledging human involvement in granting IP protection to AI-created works.
AI Authorship and India: Need for a Fix?
On a more domestic note, the Tencent case affirms the principle enacted under section 2(d)(vi) of the Copyright Act, 1957 (‘Act’), granting authorship of a computer-generated work to a person who ‘causes it to be created’. Indian courts have, in the past, also emphasized on human involvement in creation of copyrightable works (here [para. 10] and here [para. 6]). The Copyright Office considers (Form-XIV) only natural or juristic persons to be authors, requiring them to disclose their name, nationality and address.
In other jurisdictions as well, ‘decisive’ human involvement is seen as a necessary factor for the grant of copyright protection. As noted above, the Plaintiff in the Tencent case had engaged a creative team which guided the AI software’s working. Similarly, calls had been raised during the determination of the iconic Naruto v. Slater that Slater had decisively influenced (pp. 328-329) the macaques into clicking their own photos. This case was subsequently settled, but it clarified the importance of human involvement in creation of copyrightable works. Contrast this example to one where the AI machine ‘DABUS’ was designated as an inventor in the EU. The EPO declined the grant of patent, holding that only a ‘natural person’ could be designated as an inventor. Therefore, it appears that human involvement offers a satisfying solution which enables IP offices to maintain their core understanding of ‘authorship’ and ‘copyright’. But again, the desirable degree of human involvement is highly contested.
The Indian Copyright Office follows the same position as most other jurisdictions, requiring ‘decisive’ human involvement in AI-created works as an essential requisite to grant IP rights. Accordingly, a strict perusal of section 2(d) shows that while a ‘person’ who causes a work to be created shall be granted copyright in India, the position in respect of computer-generated works does not apply (p. 7) to AIs across the spectrum. At best, this may apply to weak AI requiring human intervention, but is certainly irreconcilable with strong AI, wherein no human involvement may be required.
Given the above, it is worthwhile to delve into AI’s three recognizable stages: (1) Artificial Narrow Intelligence, (2) Artificial General Intelligence, and finally (3) Artificial Super Intelligence. As for the first stage, it refers to classic weak AI, commonly found in tech gadgets of the day- speech recognition software, manufacturing robots, even self-driving cars. This type of AI relies heavily on human involvement in programming them to perform certain tasks. In contrast strong AI refers to machines possessing ‘human-like’ intelligence that can be applied to solve problems. The attainment of an Artificial Super Intelligence is an even longer shot! Since the development of AI lies, at best, on the spectrum between weak and an almost strong AI, the decision in the Tencent case seems to be well cut out that such AI programmes can be said to fall squarely within section 2(d)(vi). However, this is subject to change as technology constantly inches closer to the attainment of a purely strong AI (here and here). The question which arises, then, is – will authorship of an exclusively AI-created work be granted to humans involved or will it lead to no authorship at all?
The question posed above finds resolution in the Indian context through reference to section 17 of the Act. However, it is highly unlikely that an AI machine would be covered under sections 17(a), (b) or (c), considering the narrow interpretation of the term ‘author’ in India. Regardless, AI-created works may still not be eligible for authorship, as they are unlikely to fit into the wordings of sections 17(a) or (b) as an AI-machine is not in the course of employment of a company or an individual. Similarly, it cannot be said that an AI-machine can fulfill the requisites of a contract of service (para. 7) under section 17(c). This position has been affirmed by the EPO recently, noting that an employment agreement is limited to natural persons.
In that case, what is the true scope of authorship in AI-created works in India? Well, it is quite straightforward. As the law stands today, with weak AI in the forms of computer programmes, section 2(d)(vi) shall grant authorship to the person who ‘causes the work to be created’. But, where there is no human involvement, no authorship in the resultant work can be granted. Similarly, the concepts enshrined in section 17 shall not offer any respite to AI-created works.
While incorporation of AI-based works into the copyright law system creates crucial policy considerations, an alternate reality has already surfaced where AI-generated works might not fulfil the requirements of originality. Present regimes prioritise protection of human creations over AI-based works, and rightly so, although it may be equally important to consider striking a just balance between human creativity and development of AI technologies.
One way to do this may be to recognize that ‘authorship’ may extend beyond the traditional notions of copyright law. In that respect, it may be beneficial for jurisdictions to reconcile legal positions to incorporate a renewed concept of ‘authorship’, keeping AI in mind. At the same time, it may also be difficult to ignore sight of the fact that ‘AI-created’ work with no human involvement may not result in claims for authorship by themselves. We would suggest first, that there can be no claim of ‘no human involvement’, but merely copyright doctrine suggesting specific claims to authorship. Secondly, and relatedly, there may be policy considerations for not granting authorship to works not situated within doctrinal claims to authorship (Pamela Samuelson deals with these points eloquently on pages 1224-1225). In a climate beset with such opportunities for AI to become the mainstay in our dealings, it is imperative that the Indian Copyright Office initiates a discourse to clarify this authorship dilemma.