A BRIEF LOOK AT WHETHER AI-GENERATED WORKS CAN BE PROTECTED BY COPYRIGHT
1.11.2020
Yong De Wei, a member of alt+Law, briefly analyzes and summarises his takeaways from a webinar on the "Frontier Issues in Intellectual Property" during Techlawfest 2020.
Intro
TechLaw.Fest 2020 happened recently, and “Frontier Issues in Intellectual Property” was a one of the webinars hosted, which focused on the challenges posed by artificial intelligence with regards to copyright law. In this article, I summarise some of the points raised by the guest speakers, particularly those by Prof Pamela Samuelson from the Berkeley Law School and Prof Lee Pey Woan from the Singapore Management University.
Copyright is a form of protection for intellectual property. It is most apt for literary, dramatic, musical and artistic works such as “a painting, sculpture, drawing, engraving or photograph” per s 7(a) of the Copyright Act (Cap 63) 2006 Rev Ed. This includes computer programs, which are considered “literary works” per s 7A of the Copyright Act. Very cursorily, copyright mostly protects the form in which an idea is expressed and not that idea in itself. “[C]opyright could not subsist” in “bare facts”, as held by the Singapore Court of Appeal in Global Yellow Pages Ltd v Promedia Directories Pte Ltd and another matter [2017] 2 SLR 185 at [67]. Copyright automatically subsists in the work when it is created per s 27(2) of the Copyright Act.
The need to identify a human author
Most pertinent for this article is the fact that copyright requires the work to be traced to an author who is a natural person. In Asia Pacific Publishing Pte Ltd v Pioneers & Leaders (Publishers) Pte Ltd [2011] 4 SLR 381 (“Asia Pacific”), the Singapore Court of Appeal notes at [73] that “[t]he identification of an author is still pertinent as s 27 of the Act provides that in order for copyright to subsist in a work, the work must be original”. Originality refers to “the contribution which is made by the author to the form in which the work is expressed”, the court held at [40] citing Staniforth Ricketson and Christopher Creswell, The Law of Intellectual Property: Copyright, Designs & Confidential Information (Lawbook Co, 2nd Ed (Revised), 2010 release) (“Ricketson”) at para 7.45.
In Asia Pacific, the court found that tables that included a collection of horse-racing data compiled by algorithms and separate people could not be attributed to any individual author or group of individuals as joint authors. At [80], the court recognised that disparate groups of people had “contributed to the end product”, but “data aggregation or input is not creativity.” Copyright and AI With the development of artificial intelligence (“AI”) that is capable of generating artistic works with minimal traditional human ‘input’, this raises the question of whether such works can be protected by copyright. First, however, there is a need to understand briefly the process by which an AI algorithm generates such work. At the risk of oversimplifying what AI is, it is essentially a computer program, designed by a person, that can modify itself usually via machine learning. Raw data must be collected and then refined, often by human hands, onto a structured form the infant AI understands. The AI is then fed this data as training to modify itself, and human trainers may be involved to guide the AI along the process. After being sufficiently trained, the AI can then perform tasks independently and generate its ‘own’ work.
As can be glimpsed from the general process laid out above, there can be a vast many number of people involved in different stages of the long process before the AI generates its work. Each person or groups of persons have contributed many small steps that ultimately led to the AI-generated work in ways that are fundamentally different from an artist who paints a picture for example. People involved with the AI have no idea exactly what virtual brush strokes the AI will imprint on its virtual canvass with its virtual ‘hands’. It is very clear that the current copyright laws in Singapore cannot protect AI-generated works as it is not attributable to any human author. This is also the case in several common law jurisdictions.
The US Copyright Office also upholds this requirement of having a human author for work to be copyrightable. This could be seen recently in Naruto v. Slater (unreported, 2016), where the District Court held that a selfie taken by a monkey could not be copyrighted since it had no human author.
However, this is not the case for the UK. UK Copyright, Designs and Patents Act 1988 (c 48) (UK) (“CDPA”) s 9(3) deems the author of a computer-generated work as “the person by whom the arrangements necessary for the creation of the work are undertaken”. Such computer-generated works are simply those where no human authors as traditionally understood per s 178 of the CDPA. Thus, the UK creates more flexibility in giving courts discretion to ascertain who an author of a specific work is that is highly fact-intensive. For example in Nova Productions Ltd v Mazooma Games Ltd [2006] EWHC 24 (Ch), the UKHC found that visual elements in a video game, specifically the composite frame, that was generated by a program was attributed to the programmer at [105]. This was “because he devised the appearance of the various elements of the game and the rules and logic by which each frame is generated”. Yet, the UK approach does not circumvent the fundamental question of originality that underpins authorship - what kinds of contribution can be considered “original” and how much of it is required to confer authorship onto a specific person or group of persons.
Challenges
This is not exactly a new legal issue and has been considered since the 1980s. However, AI technology had little commercial significance back then due to, inter alia, computing limitations which are less of an obstacle now. Here are some challenges and issues that have been raised.
First, there is the question of whether AI-generated works can ever be protected by copyright laws while retaining their substantial forms. Prof Lee suggests that a fundamental reconstruction of “authorship” might be needed to reform existing copyright laws to accommodate AI-generated works. In essence, this might mean not relying on human creativity and originality to determine authorship to remove the need for a human author. She suggests commercial value as an alternative for consideration.
If such principles are too integral to the copyright regime, a sui generis regime can be considered specifically for AI-generated works. There is some force behind this argument, given that AI-generated works are intrinsically different from traditional human-created works as discussed above. Prof Lee recommends that this regime offer a limited scope of protection with a shorter period of protection of 3 years as compared to, for example, the ordinary 70 years after the death of the author per s 28 of the Copyright Act. This is to reflect the higher speed and volume at which AIs can generate work once trained in contrast to the traditional human content creator.
Second, AIs often relies on input data for its training and is integral to the content generated by the AI, which potentially creates issues if the data is itself copyrighted work. Prof Samuelson notes that such data may fall under fair use under US law. That often depends on the form in which the copyrighted work has been transformed, if at all, before being copied as input data for the AI. For example in AV v iParadigms 562 F.3d 630 (4th Cir. 2009), storing student papers in a database for an AI to detect plagiarism was held to be fair use. It might be surprising to some students that the software operated by iParadigms is none other than “Turnitin” – highlighting how pervasive AI-generated works already are.
Note: The information contained in this site is provided for informational purposes only and should not be construed as legal advice on any subject matter. You should not act or refrain from acting on the basis of any content included in this site without seeking legal or other professional advice.