GALAPAGOS
GALAPAGOS
Generative AI and the Law of Art: Proving Authenticity, Genuineness, Origin and Substantial transformation
This project has received funding from the Sector Council for research and valorisation in Human Sciences (CDR-SH) of the University of Liege
Ambitions of the CDR-SH Research Project GALAPAGOS
The advent of generative artificial intelligence (GenAI) with content-generating capacity (e.g., image, music and text, including computer code), has encouraged us to inquire if the difference between AI-generated and ‘traditional’ (human-made) art is still tenable. Its intellectual appeal aside, addressing this issue is of practical importance for the art market. AI-generated artwork causes the cost of artwork to lower and, therefore, contributes to art commodification and devaluation of creative activities through mass consumption. This begs the question of whether art audience—continue to—value human-made art more than AI-generated art, and, consequently, whether they should have a right to know the artwork’s origin.
Legally speaking, Article 50(2) EU AI Act requires providers to ensure that the outputs of the AI system are marked in a machine-readable format and detectable as artificially generated. As an exception, the AI Act mentions that, if a GenAI system had not substantially altered the input data provided by the deployer, it is not necessary to label it as AI-generated. The concept of ‘substantial alteration’ is left undefined, generating doubt on how it ought to be assessed and applied in practice.
In a context where new technologies undeniably affect the art production, without the AI Act providing clear guidelines on how to ensure optimal artist and art audience protection, there are two questions that scholarship has not yet answered.
First, does the audience value more human-made artworks than AI-generated one?
Second, bearing in mind the requirements (and their exceptions) in Article 50 AIA, how can human or non-human artworks be distinguished for the purpose of art audience protection?
The key ambition of the GALAPAGOS project is to fill an important gap in scholarship by answering these questions. The main vantage point of GALAPAGOS is, therefore, to investigate regulations authenticating creative productions as human-made.
Addressing the first gap: empirical investigation of the effects of GenAI on creative industries
The main premise is as follows: GenAI lowers the production costs of creative goods and, therefore, reshapes creative industries by enabling the mass production of artistic content. This begs the question whether human-made art can still have access to viable markets due to competition from “cheap(er)” AI-generated content. In a similar vein, it is sometimes argued that GenAI might lead to the replacement of human creativity, marking the death of (human) authorship. In light of this, a burgeoning current is investigating the transformative impact of GenAI on the arts: GenAI challenges traditional creativity—often seen as an individual or collective human endeavour, where originality is highly prized—by introducing AI-generated outputs that blur the line between human and artificial creativity. Building on Theodore Adorno’s seminal argument, GenAI seems able to transform art into a commodity, consumed in the same manner as any good intended for mass consumption. The obvious counterargument here is that there will always be artists to create, and audience to cherish human-made artworks. However, some evidence from the music industry—such as the case of some fully AI-generated song which went viral—suggests otherwise. Despite this, the existing research remains largely anecdotal and not based on a systemic research.
The GALAPAGOS project will fill this first gap by empirically investigating the effects of GenAI on creative industries. To do so, two global public consultations will be launched in parallel to (1) assess whether human-made artworks are valued more than AI-generated ones, and (2) to map the human artist’s use of GenAI and its impact on their creative productions.
Addressing the second gap: exploring the concept of substantial transformation
People must be able to distinguish artworks according to their origin (human or GenAI) to properly value them. Yet, while artists know whether their artwork was AI-generated, the audience does not: human-made and AI-generated artworks are indistinguishable. Economists describe such information asymmetries as a lemons problem: when buyers are unable to distinguish good- from poor-quality (‘lemons’) products, they assess quality on average and internalise the risk of buying a poor-quality product at the price of a good-quality one (reducing their willingness-to-pay). Good-quality products, therefore, are never purchased at a fair price.
Consequently, honest sellers (of good-quality products) withdraw from the market. The art market constitutes a fertile ground for the lemons problem: not only are GenAI artworks indistinguishable from human-made ones, but GenAI users also have an incentive to cheat because, more often than not, AI-generated artworks will be uncopyrightable. As copyright is an exclusive property right that grants its owner a temporary monopoly, GenAI users are incentivised not to disclose their work’s true origin. The upshot is consumer deception, which, as stressed by Recital 133 AI Act, is the rationale behind the EU authentication rule for AI-generated content. Interestingly, both California and China have similarly chosen to adopt a rule of origin. Yet, no guidance whatsoever is given to users on when to disclose the origin of the work and when not to, because it has (not) been substantially transformed.
To fill this second gap, the GALAPAGOS project will explore the elusive concept of substantial transformation. To do so, it will engage in a discussion with artists to identify what they consider is, or not, a substantial transformation of their work. This research project will then launch a final global public consultation, in which participants will first assess whether, and to what extent, an AI-generated artwork (output) based on a human-made one (input) constitutes a substantial transformation of the latter. Second, they will answer the opposite question: to what extent the modifications made by a human artist to an AI-generated work allow the final result to be considered not (anymore) as AI-generated, but as the creation of a human author.
Clarifying the concept of substantial transformation is not necessary only for the purpose of the AI Act’s enforcement. It is also needed in copyright law.
From the input perspective, US fair use doctrine allows the use of copyrighted materials—without rightsholder’s permission—as training data if, inter alia, GenAI output is transformative and does not compete with or harms the market for the original work. The GALAPAGOS project answers these two conditions.
From the output perspective, significant human intervention is required for AI-generated output to be copyrighted. Although in line with traditional copyright doctrine, this solution leaves room for inconsistencies, as no threshold can be drawn in this regard. The elusiveness of substantial transformation is, therefore, source of deep legal uncertainty for creators who rely on GenAI, who may increasingly rely on either licensing terms that govern the use and distribution of AI-generated content.
Members of the CDR-SH Research Project GALAPAGOS
Principal investigator: Jerome De Cooman
Previous publications related to the CDR-SH Research Project GALAPAGOS
CDR-SH Research Project GALAPAGOS Achievements
