[This article is authored by Jatin Yadav, a 2nd year B.A. LL.B. (Hons.) student at Hidayatullah National Law University. It discusses the prospect of artists’ legal standing to sue AIs that are trained on their productions for copyright infringement in respect of substantial similarity test and fair use doctrine. This is the second article in the series titled ‘AI Art,’ the first part can be found here.]
The year 2022 gave rise to some unprecedented legal suits embroiled in AI artistic creation; many artists are going against various platforms for employing their productions in AI generative algorithms without proper cognizance of the artists’ contribution. This class action will prove to be pivotal in shaping the future prospects of such disputes.
AIs in such incidents belong to the ‘family of learning algorithms’, which by processing a huge number of man-made paintings/sounds/writings emulate human artists. Such a process is called learning. These learning algorithms help in training AIs to create art on their own or assist humans in the process of doing so. Such a category of AIs is called generative AIs. However, from the very inception of its newfound popularity, the tremendous commercial use of generative AIs casts serious doubts concerning the violation of copyrights of the artists whose works have been employed for training the AIs as theirs. The products of such generative AIs will definitely compete in the same market. Even though it’s clear that art pieces independently drawn up by AIs cannot be registered as copyrights in absence of substantial human involvement, a picture is quite opaque regarding other aspects of it: will creators/ artists have locus standi to bring an action against AIs trained by processing their productions for violation of copyright? There is a lack of judicial precedents concerning this scenario, and plausible inferences can be and are being drawn up to find an equilibrium between creative values in such AIs and copyright law.
Usually in cases short of direct exact copying; two fundamental elements of the copyright infringement test are pertinent to the current discourse: (a) If the AI product closely resembles its human-created antecedent (b) even if it does; whether it can still be considered as fair use given that other elements along with originality are presumed to be met as it is easy to be met and the legal implications that can be drawn from these talks are unaffected by the use of AI.
The most instinctive question to arise in a plausible AI infringement case is if the copy in question is similar enough to the original work so that artistic expressions such as strokes, languages, framework, or arrangements in contrast to styles, ideas and theme becomes confoundingly similar to the eyes of a rational observer. The origin of the substantial similarity test can be traced back to American copyright laws, but it has also been relied upon by the Indian Supreme Court in the case of RG Anand v. Deluxe Films and Delhi High Court in India TV Independent News Service v. Yashraj Films. This is a naturally ambiguous and fact-intensive component, but for AI it’s trickier because at least a portion of the algorithm relies on probability modelling, which, to put it simply, pulls the most frequent features/arrangements from the database it is learning from. The product generated this way; when looked at as a whole, exhibits a similar style or combination of familiar elements used by several artists. This situation gives rise to a dual conundrum for potential suing artists. First, the lack of a prima facie case as recognition of a common style or approach to a common theme in an AI piece is not a substantial ground for constituting the same. As the extent of protection accorded by copyright is limited to expression and excludes ideas/ procedures from its ambit. Second, even claims of an exact emulation of arrangements or rendition face dilution due to the presence of other components the AI learned from other works provided that they are not dispositive elements; most jurisdictions will take them into account while building a prima facie case. Moreover, there exists flexibility in the court’s comparison; varying from jurisdiction to jurisdiction as some adopt to look at the production as a whole while others indulge in artistic details. The outcome of the case concerning the appearance of artistic elements of the suing artist will considerably depend upon the court’s approach. Nevertheless, if an AI model was only trained on the works of a sole author, its output most probably will come under the derivative rights of the author, which extends to the productions that have been construed upon the author’s own work.
Another impediment in the way of suing artists is fair use. This component, however, necessitates a separate discussion of AI generative algorithms and employing them for financial gains. Section 52 of the Indian Copyright Act stipulates specific acts that cannot be construed as copyright infringement relating to artistic, dramatic, literary work excluding computer programs. This provision includes 1) private or personal use including research, 2) criticism and review of the work, and 3) reporting on current events and affairs; among other exceptions. Whether an act would be fair use would depend on a variety of factors such as “1) the purpose and character of the use, 2) the nature of the copyrighted work, 3) the amount and substantiality of the portion used in relation to the copyrighted work as a whole, and 4) effect of the use upon the potential market for or value of the copyrighted work.” laid down in Folsom v Marsh. The rationale of section 52 was laid down by the court in the case of Wiley Eastern Ltd. V. IIM to safeguard freedom of speech and expression imbibed in Art 19(1)(a) of the Indian Constitution. This implies that there is a high probability that the use of artistic works by an academic researcher to train an AI will be deemed to be fair use by a court because it comes under the umbrella of freedom of speech and expression such a principle doesn’t appear to be relevant to the commercial application of AI for mass production or automation of art. As it is often with the motive of profit and strips away many artists of their work’s value. However, there remain chances that such use of AI would constitute fair use. In V Ramaiah v K Lakshmaiah along with above mentioned four factors; the court also considered the transformative character of the production in question i.e., whether it crafts a substantially different and novel objective premised upon the original work. A case considering such transformative use of original work is currently pending before the apex court of the U.S.; in which the court must adjudge if repetitions of one image under several aesthetic reprinting by Andy Warhol is enough to constitute transformative usage of the picture. If the Supreme court of the US answers in the affirmative that the stylistic reprinting and rearrangement of them in an unexpected way is sufficient to be transformative usage then there is a possibility that AI works whose distinguishing characteristic is precisely the kind of re-renderings that frequently stretch human imagination, will likewise be transformative fair use. The argument is speculative in nature, but the probability of such a happening is certainly open. The potential application of the principle of fair use also provides the court an early opportunity to test out strategies for balancing the advancement of AI and the value of human labour, a dilemma the court will inevitably have to address.
Although it is hard to draw a firm legal judgement on whether an artist whose works have been employed to train generative AIs can bring a suit against their owners; the stakes in this arena are rising swiftly. Stakeholders should keep an eye out on the cases that are being filed with special attention to the substantial similarity test and the fair use doctrine as they can have different consequences when applied as compared to their more general application.