Decoding Authorship: Is There Really no Place for an Algorithmic Author Under Copyright Law?

Abstract

The third technological revolution is dragging the world, incrementally but surely, into an era where Artificial Intelligent (AI) is taking over cultural production from human beings. Nevertheless, authorship remains largely defined in humanistic terms across Western cultures as a projection of the image of the “romantic author” emanating from the 18th century. Although it is not spelt out in the law, judges do seem to demonstrate remarked reluctance to grant authorship status to AI. The developments in three important areas may enable a reconstruction of the human-centric authorship ideology: a structural pro-corporate prejudice, a low threshold for originality, and the judicial avoidance of aesthetic assessment. Distilled to its essence, AI can be as equally creative as human beings as they follow the same laws of cultural production. In dealing with the resulting ownership issue of AI authorship, an altered “work made for hire” doctrine is proposed as a promising solution design template.

1 Introduction

The modern copyright regime is deeply entangled with the concept of authorship stemming from the 18th century, which depicts the creative process as an extraordinary intellectual endeavour by individuals with exceptional genius in moments of inspiration springing out of nowhere.¹ Although the embedding of this romantic author into the doctrinal structure of copyright law changes both in a significant way,² its humanistic lens remains the shaping force of the popular understanding of authorship. Nevertheless, the third computer-enabled technological revolution is dragging the world, incrementally but surely, into an era of algorithmic creation where the process of cultural production is engaged with the indispensable assistance of algorithms (AI-assisted works) or even completely in reliance on them without any human intervention (AI-created works) in a wide range of copyright spheres. The ambitious The Next Rembrandt project, for instance, creates new paintings in the style of the Dutch painter Rembrandt by using deep learning algorithms to analyse the statistical properties of existing Rembrandt portraits;³ the story generator programme BRUTUS is capable of producing short stories with all of the dimensional elements that a typical literary work should cover, including plots, characters, settings and writing styles.⁴

These AI-enabled cultural creations redefine the respective role of human beings and machines in the creative process and challenge the traditional paradigm of authorship under copyright law. A series of questions arises therein: How does AI fit into the landscape of copyright law? Does the positive copyright law recognize an artificially intelligent author? Whether and to what extent copyright law should accommodate such algorithmic twist?

Regrettably, policymakers and judges do not seem to have the incentive to go out of their way to engage these discussions. The simple reason is that they do not have to, especially in light of the lower-than-predicted progress in AI research and development. Since until now there has not been any completely autonomous and independent AI, and the current copyright legal framework is admirably capable of dealing with the challenges brought by existing AI technologies (by attributing the AI creation to a particular human actor as the author), the “computational creativity” of the autonomous algorithmic authors is thus “always within view but never actually pressing”⁵

Such satisfaction with the status quo is, nevertheless, not going to last long as it will only be a matter of time for AI to dominate cultural production. The distance, perhaps quite a long one, between now and the surely-to-arrive era of human-minimal algorithmic creation by no means implies that the discussion of AI authorship is of less practical significance. If anything, it provides a “reprieve” from the “reactive, crisis-driven” model of law-making that is typical of copyright law in face of challenges posed by the emerging and disruptive technologies.⁶

This article is not concerned with AI-assisted works as this ground is well covered by a significant strand of scholarship.⁷ Rather, it focuses on the works automatically created by algorithms independently of human beings (i.e. AI-created works). Section 2 examines the current legal construction of authorship as a limit of the scope of the subject matter of copyright law. It starts with a cursory investigation into the complex international copyright treaty system as a general study and then puts copyright in two more concrete contexts by tracing the development of two different lines of case law established in the EU and the US which represent the moral rights model and the utilitarian model of the copyright regime respectively.⁸ Section 3 looks beyond the details of these doctrinal analyses and gazes into a broader picture re-examining the inter-relationship between authorship and copyright law. In interrogating the legal and normative assumptions with regard to the nature of authorship, it seeks to re-measure the limits of the doctrinal elasticity of authorship and shed new light on the possible entry points where AI may be accommodated into this revisited, dehumanized authorial regime. Section 4 takes up the thorny issue of ownership as a second theme developed from AI authorship. A difficult choice is made between a “public domain approach” which directs all AI creations into the public sphere free for uses by the general public and a “proprietary copyright approach” locking AI-created works within the realm of private controls. By taking a bird’s-eye view incorporating the whole production chain of the AI project from the creation of the AI programme per se to the works made by AI, it suggests that an altered work-made-for-hire doctrine which assigns the ownership of AI-created works to the programmer may well be a promising solution to enhancing innovation and promoting cultural progress within the wider society, which is exactly what copyright law aims for through the incentive and competition mechanisms.

2 The Legal Construction of Authorship

2.1 International Copyright Law

Before delving into the concrete national legal framework for authorship, it is useful to look at how it is fashioned by international copyright law which forms the paradigm for national copyright regimes. The international copyright system is primarily composed of three key pillars which include almost all countries in the world: the Berne Convention, the Agreement on Trade-Related Aspects of Intellectual Property Rights (TRIPS) and the WIPO Copyright Treaty (WTC). Pursuant to Art. 2(1) of the Berne Convention, the subject matter of copyright protection – that is “literary and artistic works” – includes all production in the “literary, scientific and artistic domain, whatever may be the mode or form of its expression”. Article 10 of TRIPS and Art. 4 of the WCT further clarify that computer programmes fall within the meaning of Art. 2 of the Berne Convention. It thus seems straightforward that both AI programmes per se and the generative contents produced by them could be protected as “literary and artistic works”, provided that they are expressive in nature.

Implicit in the concept of “literary and artistic works” is an additional requirement of originality.⁹ A work would have to be an intellectual creation by an identifiable author. Whether inadvertently or not, the Berne Convention remains silent regarding the crucial issue as to who can be qualified as an author within the meaning of the Convention and leaves it to national legislation. It is argued by some that one should not read too much into this suspicious self-restraint and ambiguity. The reason why the Convention left the concept of authorship undefined was simple – it was unnecessary do to so as the humanistic grounding of the term was generally accepted by the contracting states as a given. Article 4 proscribes an inclusive list of eligible subject matters, and it would be hard to believe that there would have been agreement on these items if a general understanding of authorship had been absent.¹⁰ Nevertheless, Art. 4(a) of the Convention, which extends protection to the makers of cinematographic works, signals a clear non-human author essence, even though it is often construed as an exception to the general human-centric conception.¹¹ The tolerance towards the non-human author is evidenced further in national legislation on copyright, in particular in the US where the status of employers as the author-in-law is firmly upheld. At the very least, the presence of Art. 4(a) indicates the possibility of creating a new Convention norm, or sanctioning a departure from such a norm even if humanness were to be read into the construct of authorship under the Convention on the basis of state practice.¹²

Indeed, with all these “conflicting signals” coming from all directions – whether it is a study of the origins and development of the Convention, a doctrinal analysis of the text of the Convention per se, or the implementation of the Convention in the territory of member states – “the search for a clear Convention philosophy” often ends up being an “elusive” and “frustrating” one.¹³ Such widespread inconsistencies, although having the side effect of acting against a coherent global understanding of copyright protection, actually reflect and result from widely divergent national approaches, which are permitted implicitly by the Convention, in the form of a necessary “price to pay” for reaching a wider international consensus.¹⁴ In other words, the Convention itself does not prohibit a more liberal and less human-centric approach to authorship at a national level. It is this considerable “liberty space” between the doctrines of the Convention that makes possible these seemingly competing understandings and practices.

Another important argument in favour of the view that the Convention only envisages human authorship is largely founded on Art. 6^bis of the Convention which expressly grants moral rights to the “author”. Since AI does not have a legal personhood, it would seem odd to visualize AI as an “author” within the context of the Convention. As Jane C. Ginsburg argues, to allow the recognition of purely computer-generated works as works of authorship under the Convention would entail the expulsion of the human author and “purge copyright of its soul”.¹⁵

A pause for thought should be needed here. Even if it is accepted that this human-centric narrative has some merits, the Convention does not always bind the contracting parties to its effects as a matter of fact. For instance, the US successfully managed to get away with the issue of moral rights by enacting the Berne Convention Implementation Act of 1988, which stated that the Convention was not “self-executing in that existing law satisfied the United States’ obligations in adhering to the Convention”.¹⁶ As a result, the Visual Artists Rights Act of 1990, which only granted limited moral rights to a specific type of artists with a substantial number of carve-out clauses in place, was formulated to “appease the domestic lobby”¹⁷ and soften external criticism for failing to provide a higher protection standard.¹⁸ Indeed, moral rights are simply not “on the ballot for harmonization”¹⁹ on an international stage, and it is so even within the EU itself, which is arguably the very birthplace of the moral rights rationale.²⁰

On the other hand, the humanistic “moral rights” reading of authorship is further undermined by the deliberate omission of the moral rights clause (i.e. Art. 6^bis of the Berne Convention) from TRIPS which incorporates almost all substantive Berne provisions by reference as a result of an outright rejection by the US.²¹ This particularly strong opposition standing of the US, as opposed to its attitude towards the Berne Convention, could be explained by the institutional difference between these two international treaty systems. Unlike the WIPO, which administers the Berne Convention, the governing organization of TRIPS, the WTO, has a compulsory enforcement mechanism set in place which member states could use against those failing to protect the moral rights of authors if the moral rights clause were incorporated into TRIPS. In other words, to include moral rights into TRIPS would give the provision “teeth” and this, apparently, does not bode well with “the broader apparatus of trade relations”²² from the perspective of the US.

As such, international copyright law does not provide too much useful guidance on whether AI could be granted the status of author.²³ If anything, the ambiguity or inconsistencies within the complex international treaty regimes as a result of the balancing of competing national interests, does nothing more than to leave a way open for the algorithmic author, although it may well not have been within the contemplation of the EU or the US at the outset.

2.2 EU Copyright Law

The concept of authorship is directly linked to the originality requirement in the EU. The EU copyright directives have harmonized the originality standard for specific subject matters: computer programmes, databases and photographs. In Infopaq, The Court of Justice of the European Union (CJEU) extrapolated from this partial harmonization a generalized concept of originality for all “literary and artistic” works as “the author’s own intellectual creation”.²⁴ This has been reaffirmed by several subsequent decisions. For instance, in Levola Hengelo the CJEU iterated that the subject matter of copyright must be original in the sense of the author’s own intellectual creation.²⁵ In further elaborating on the concept, the CJEU emphasized the need for the work in dispute to demonstrate the exercise of “free and creative choices”²⁶ and reflect the author’s “personal touch”.²⁷ Although the CJEU has never spelt out explicitly that only human authors could trigger copyright protection under the EU aquis, its repeated references to the “personal touch” or “personality”,²⁸ which is a trait unique to human beings, seem to suggest that the originality requirement involves certain degree of humanity, if not entirely, and confines the concept of authorship to natural persons. This anthropocentric view is expressed in even stronger terms by Advocate General Trstenjak in her opinion in Painer where she concluded, from the wording of Art. 6 of the Term Directive, that “only human creations are therefore protected, which can also include those for which the person employs a technical aid, such as a camera”.²⁹

Indeed, the EU copyright aquis is profoundly influenced by its long-standing tradition of author’s rights (droit d’auteur). It attributes authorship, for the purpose of conferring copyright protection, to the person who held the pen and did the actual writing through the making of “free and creative choices”. In Painer, the CJEU detailed three key phases of the creative process: preparation (creating the design or plan of the work), execution (converting the work on paper into the work in reality), and finalization (processing the draft version of the work into final cultural products).³⁰ Although AI can be instrumented to perform many sophisticated tasks and is increasingly taking over the execution phrase from the hands of human beings, authorship places more weight on “mind” over “muscles”³¹ – preparation and finalization are the places where humans are more likely to and more capable of making creative choices, and thus more relevant for the originality assessment.

Nevertheless, the key inquiry remains to be a finding of human input, even if minimal, in the creative process. Where there is no one holding this proverbial pen, although human creation is necessary for the production of the work (i.e. creative input by the AI programmer to write down the codes or algorithm), the human creative actor is not the author of the work in the traditional sense. In other words, the relationship between the work and the human actor is so mediated and attenuated that AI could not be regraded simply as a pen or a tool.³² There is a blind spot in cases where such circumstances occur. On the one hand, the CJEU may feel an impulse to decline to grant AI authorship status in light of the droit d’auteur tradition of the EU aquis. However, it is far from clear how this tradition would give effect to the CJEU’s decision, as a strict adherence to the human input test would actually dictate a finding to the contrary.

2.3 US Copyright Law

The humanness of the concept of authorship is much less evident in the US, which adopts a more pragmatism-oriented, incentive-based approach to copyright law. Under the US model, copyright protection is extended to an original work of authorship³³ and according to the established case law, the test for originality would only require a minimal level of creativity. In Alfred Bell & Co v. Catalda Fine Arts, Inc, the Second Circuit made a distinction between patents and copyrights based on the level of originality required. It was ruled that although patent law requires a high threshold of originality in that a “startling, novel or unusual, a marked departure from the past” must be shown, “original” in the context of copyright means no more than a designation of source, that is, “the particular work ‘owes it origin’ to the ‘author’”, as opposed to a metric of creativity.³⁴

This broad reading went too far. In Baltimore Orioles, Inc. v. Major League Baseball Players Association, the Seventh Circuit was dissatisfied with the unitary approach taken in Bell which conflated the two constitutional requirements for copyright protection – i.e. originality and creativity – into a single criterion and tried to keep them separate from each other. On the Court’s interpretation, a work is original in the sense that “it is the independent creation of its author”; whereas “a work is creative if it embodies some modest amount of intellectual labor”.³⁵ In making such distinction, the Court in fact required a higher threshold for copyright protection for authorial works by contending that “the requirements of independent creation and intellectual labor both flow from the constitutional prerequisite of authorship and the statutory reference to original works of authorship”.³⁶ In support of this restrictive reading, the Supreme Court ruled in Feist Publications, Inc v. Rural Telephone Service Co that copyright only extends to works that possess more than a de minimis quantum of creativity. In this particular case, the Supreme Court was invited to decide whether the white pages of a telephone directory were an “original work of authorship”. The Court did set a relatively loose standard by contending that originality did not require facts to be “presented in an innovative or surprising way”, and that their selection and arrangement “cannot be so mechanical or routine as to require no creativity whatsoever”;³⁷ nevertheless, what it failed to do was to provide a more clear guidance in a more positive tone on the spectrum between the “innovative” and the “mechanical” in order to “locate the break between eligible and ineligible subject matter”.³⁸

The consequence of such ambiguity was that it is unclear whether works created by machines automatically fall out of the copyright pale, especially in the absence of any further elaboration on meaning or scope of “mechanical”. A literary reading of the word per se may imply the humanness of the creative actor as a prerequisite for protection;³⁹ however, it seems that in using that word, the Court was more concerned with the level of creativity required rather than the nature or the origin of creativity. This origin-neutral interpretation is in fact not too strange to the US, which readily recognizes non-human authors. Section 101 of the US Copyright Act makes a special case for legal persons who can be granted the status of authorship upon satisfaction of certain conditions. Under this so-called “work made for hire doctrine”, the employer or the commissioning party is deemed, in the eyes of copyright law, as the author of the work made or commissioned by a third party (the actual author) and “owns all of the rights comprised in the copyright” except for moral rights under Sec. 106A.⁴⁰ However, again, the law is always vague concerning the legitimacy of algorithmic authorship. Although divorcing from the traditional human authorship narrative, the work made for hire doctrine is still based on a human author-in-fact, and the possibility of AI as the author-in-fact rarely has the chance to appear on the agenda of legislation or the courts.

3 Authorship Revisited: How Far Is Humanness Required by Copyright Law?

Technological developments are often considered as an external force upon which the law acts. Notwithstanding, to the extent that a particular attribute of a particular technological advancement (in our case AI-created works) disrupts the law, it does so only because the law is structured in a way that makes that disruptive attribute relevant.⁴¹ Just as Annemarie Bridy puts it, the determination of whether AI could be regarded as an author is in essence a process of evaluation of how far humanness is required by copyright law.⁴² If copyright law constructs authorship as explicitly requiring human brilliance or human intellectual labour, then there may be little room for the algorithmic author. However, if the law moves its focus from the creator to the needs of the audience to which the work is addressed, it may be easier for AI to be incorporated into that authorial map. So the important question here is whether the structure of copyright law could allow for sufficient elasticity, both doctrinally and normatively, to construct itself in a more algorithmic-friendly manner.

Given what has been discussed so far, it is clear that, at least from a purely doctrinal perspective, neither international copyright treaties nor national laws in major jurisdictions have completely ruled out the possibility of algorithmic authorship, although they have demonstrated reluctance, to a greater or lesser extent, to do so. In fact, even in the 18th century, shining with the brilliant sparks of renaissance humanism,⁴³ the romantic authorship ideology was never meant to empower the creation of individual humans, but for greedy corporate cartels who hid in the shadow of such an illusory human-centric narrative and used it as a pretext in disguise of their despoilment of the rightful claims of the actual authors of works.⁴⁴ Although the rise of the concept of authorship did outlaw the then prevalent “best exploiter regime”⁴⁵ in the pre-authorial era, which assigned copyrights to the economic actor who was best positioned to exploit the underlying work in a formalistic sense (i.e. a publisher or a printer rather than the actual author), it was somehow gradually reduced to a one-sided mediated mechanism for two conflicting demands – recognized initial ownership by the author-in-fact on the one hand, and on the other hand, allowed market transactions to transfer the work along with the bundle of rights attached to those who were most capable of extracting commercial value out of them. Unsurprisingly, once again, the non-human powerful operators triumphed over the actual human authors especially in light of the then fast-changing economic landscapes and practices (e.g. mass production, a rising share of collaborative creation, and the formation of national markets).⁴⁶

The authorship construct as it is today has hardly travelled very far from what it was like two or three hundred years ago. If anything, modern copyright law’s prejudice against individual human creative actors is wrapped in a more subtle presentation as part of the efforts to adjust traditional legal doctrines to the new market environment of “corporate liberalism”.⁴⁷ As W. M. Geldart puts it, the new default assumption in almost all contemporary legal fields is that “corporate bodies are really like individuals, the bearers of legal rights and duties”.⁴⁸ In addition to this corporatism lens, the development of two trends which are gaining momentum in both the US and the EU may further contribute to the dehumanization of the accounts of the story of authorship.

The first important element is associated with the relatively low threshold of originality. Under the influence of “newly powerful economical actors” (e.g. commercial publishers and advertisers) whose for-profit activities often require no more than a modicum of creativity in order to secure their competitiveness by seeking copyright protection on their products as widely as possible, creation is gradually being stripped of its romantic sugar coating and judged by its extrinsic market values rather than intrinsic characteristics.⁴⁹ As a reflexion of this wider change in the market, the courts rarely look into whether the author is creative enough but put more of an emphasis on the creativity of her works instead. It is not something new to judges in the US, as they have been quite used to looking into the social utility rather than the inherent humanness or personhood of copyrightable works for justifications and guidance in making their judgment. It is perhaps astonishing to see that even in the EU where the moral rights thinking still has a quite strong currency, the threshold for originality is also set actually very low. In Infopaq, it was ruled that an 11-word text fragment could qualify;⁵⁰ and in Painer, the originality of the photographic work was treated basically as a given by suggesting that even a fairly obvious choice in the planning, execution or reduction of an AI-assisted output could suffice.⁵¹ Such a subject-matter-focused judicial practice actually opens the door for non-human authorship, with the de-centralization of the identity of the author from the creative process and the rise of “formalist” and “hand-off” judges.⁵²

Furthermore, the increasing judicial reluctance to assess the creative values of works plays an important role in re-modelling the conception of authorship. In the EU, the CJEU has made it clear that the artistic value or aesthetic values of works is not part of the court’s analysis.⁵³ EU copyright law protects fine arts as much as it protects more mundane intellectual outputs.⁵⁴ US judges refrain from making aesthetic judgements⁵⁵ primarily on two grounds: on the one hand, it can prevent inappropriate judicial censorship on the basis of the judge’s personal likes or dislikes;⁵⁶ on the other hand, as non-experts, it is often unrealistic to expect judges to accurately evaluate the creative values of artistic works, which can be a highly technical and specialized matter.⁵⁷ This inhibition paves the way towards a non-human copyright protection mechanism. If judges are prohibited from making aesthetic judgements of a work, as a natural consequence the creativity of the author would also be excluded from the examination by the court. In other words, the humanity of an author would be irrelevant when determining whether an artistic work qualifies for copyright protection.

Even if we set aside these technical doctrinal observations, distilled to its essence, AI could be equally capable, if not any better, of being as creative as human beings. It is widely contended that it is the rule-bound and deterministic nature of AI that forecloses the possibility of algorithmic authorship. However, this argument can hardly stand true because computers can now be programmed to produce unpredictable results by introducing elements of randomness into their processing of data. And it is nearly impossible to imagine a creation which is completely rule-defying or breaks with pre-existing codes and canons. In fact, all works are created with the same formula – by following a particular pattern or set of rules. Take poetry as an illustration. All poets would need to process existing works, extrapolate rules from examples (e.g. grammar, syntax and diction) and then apply those rules in their compositional activity. Instead of being “something intuitive, immediate, authentic, and all-embracing that springs up who knows how” as how most aesthetic theorists would describe it, the actual process of writing is much less romantic. Underneath the “finding-the-muse” sugar coating is “a constant series of attempts to make one word stay put after another” by following particular rules.⁵⁸

In a sense, the human brain is like a computer or algorithmic processor of its own, or as the AI pioneer Marvin Minsky puts it, a “meat machine”.⁵⁹ It can demonstrate a significant computational or algorithmic characteristic. In 1960 the famous avant-gardists Raymond Queneau and Francois Le Lionnais founded a niche writing group called “Oulipo” in search of new writing styles with self-imposed external constraints. Oulipians were obliged to follow an array of rigid and eccentric rules, which vary in their complexity during the process of creation, from the “S+7 method” where each noun in a given text is replaced by the noun found seven places away in a chosen dictionary⁶⁰ to the “boule de neige constraint” for poems, requiring that each line start with a single word with each successive word one letter longer than the one before.⁶¹ A representative work under the influence of this so-called “Oulipian method” is George Perec’s La disparation, which is written entirely without the letter “e”.⁶² Quite contrary to the widely held perception of rules and constraints as something antithetical to creativity, the “radical formalism” embraced by the Oulipian school, which endeavours to bring to the fore the dialectical relationship between “determinism and choice”⁶³ and “the mechanical and the human”⁶⁴ in the context of cultural production, demonstrates that all creation is inherently “derivative” and “algorithmic”.⁶⁵ Indeed, it is those mechanical constraints by which creativity is defined and resides in.⁶⁶

4 Divorcing Ownership from Authorship: Who Owns AI-Authored Works?

One key problem flowing from the recognition of algorithmic authorship is the allocation of ownership. Under the traditional construction of copyright law, authorship is the legal basis of allocation of ownership. The assumption goes as follows: since the work originates from the author, she is the owner of the copyrights in the work in the absence of evidence to the contrary. However, this assumption would be problematic in the context of the algorithmic author as it would make no sense to assign the ownership of any proprietary rights to non-human actors.

The work made for hire doctrine in the US, which assigns the ownership of copyrighted works away from the author-in-fact, provides an ideal template to deal with this potential conflict. Under this doctrine, the employer or other person for which the work in dispute is made is considered as the author in the eyes of copyright law. Similarly with regards to AI-authored works, treating the programmer as the employer, that is, the owner-in-law, would avoid vesting rights in machines. This is a pure legal fiction, but it also has a dimension which reflects the truth. It is both intuitively and conceptually convenient to jump straight to the conclusion that the programmer is the author of the authorial works created by AI because apparently she is the author’s author. But this is simply not correct in a doctrinal sense as the programmer does not make any meaningful input in making the creative choices necessary for the particular creation. AI has its own mind which is completely independent of its programmer as soon as the codes for creating AI are finished. Rather it is AI who is the de facto author.

In fact, some common law countries have already taken a similar approach to protecting works created by an algorithmic author. For instance in the UK and New Zealand there is a dedicated provision for computer-generated authorial works which vests ownership in “the person by whom the arrangements necessary for the creation of the work are undertaken”.⁶⁷

Indeed, even if by acknowledging an inconsistency between the author-in-law and the owner-in-fact, this proposed approach erodes the traditional authorial ownership rules in a strict sense as the legal recipient of the copyright in the work is not the actual author of the work. But this should not be something that calls for concern as such authorship-ownership linkage has long been dissolved since the middle of 19th century when cultural production was more likely to be conducted in a collaborative manner and the default rule of ownership allocation gradually shifted to favour corporate creators who could best utilize the works (i.e. exploit their commercial value) to the detriment of the actual authors.⁶⁸

Perhaps, a more pressing concern arising from this altered employer ownership approach is the issue of overprotection. Copyright protection does not always guarantee more prospering creativity, although in many cases or at least in many theories it does so. As evidenced in the fashion or cooking industries, some creative areas are actually better off without the intervention of copyright law.⁶⁹ In the context of AI, there may already be sufficient incentives in the upstream of the value chain of the creative process, which enhance both the quantity and quality of AI creations.⁷⁰ If a systematic approach or a bird’s-eye view was taken to analyse the whole creative process of a well-performed AI system from scratch, one might find that there is copyright protection for the algorithmic programme per se, e.g. sui generis protection⁷¹ or trade secret protection⁷² for data that the programme relies on to function,⁷³ as well as potential technological and contractual protection measures. With these multiple protective filters, would it then be reasonable to add one more safety net for AI-authored works, which may well be proven unnecessary or even counterproductive?⁷⁴

An oft-provoked response to this train of thought would be that all works created by an algorithmic author should be channelled into the public domain. This article does not intend to challenge the significant role that the public domain plays in safeguarding democratic social values and basic human rights (e.g. the right of free expression and the right of access to information), in addition to incentivizing innovation and cultural production.⁷⁵ Rather, it is on this very basis that more deliberate and thorough consideration is called for before dismissing algorithmic authorship as irrelevant or unjustified.

It is fairly obvious that conferring no protection on AI-authored works would increase the influx to the public domain; however, what is less obvious is that this protection-free practice in the downstream of the algorithmic creative process may backfire on the willingness of upstream creators to contribute to the public domain. In the digital world, the “free software movement” often accounts for the extremely vigorous vitality of innovation and creation occurring in this virtual playing field.⁷⁶ The key to keeping such practice alive lies in the fact that all creators who put their software or programme into the public reservoir of knowledge could free-ride on other creators’ improvements or inputs in their programme.⁷⁷ However, suppose that the AI programmer forgoes their claims of copyright in the programme creating the AI, if works made by AI were not copyrightable. This would mean that their prospects or ability of exploiting the commercial value of these AI-authored works would be significantly restrained. The programmers may be discouraged to stick to the traditional pro public domain practice as the gains from placing the AI programme into the public domain and free-riding on others’ inputs may still be outweighed by the potentially huge market value of claiming copyrights over the massive amount of works created by the AI programme. Out of the fear of losing such more-direct and more-tangible commercial profits, the programmer may simply not wish to share their software with others free of charge.⁷⁸ In other words, the overall utility drops under a cost-benefit calculation and the “Unus pro omnibus, omnes pro uno” (“One for all, all for one”) circle collapses.

Indeed, we seem to be trapped in a no-win situation. If we extended copyright protection to AI-authored works, the public domain may shrink by losing these possible creative sources in the downstream; however, if we refused to do so, the public domain may also shrink as the creators upstream may no longer share their AI software or programmes with others. This paradox does exist, but bear in mind that the value of the public domain is not measured in a quantitative sense and the second scenario actually presents more of a hazard to the flow of creation in the public domain since the core creative value resides in the creation of the AI programmes per se rather than the products they make. Especially for non-European jurisdictions, the multiple layers of upstream intellectual property protection may still be ineffective, as the legal protection for data or databases may often be found inadequate or even absent.⁷⁹

Furthermore, there is a pro-competitive dimension of this approach which is often overshadowed by the fear that the mass-production and viral proliferation of AI-created works would soon deplete the usable public domain and make it impossible for creators to navigate within the safe area between these copyright anti-commons.⁸⁰ To worsen the situation, the aesthetic quality and conceptual coherence of these AI creations, at least judging from the current phase, are often disappointing. Talk to Transformer’s stories were often found unintelligible⁸¹ and the Portrait of Edmond Bellamy even had no discernible nose.⁸² However, mundane creation has long been no stranger to us without the hypothetical “contribution” by AI, which, by the way, is unlikely to confine itself to creating things of little value considering how far technology has gone in evolving itself and revolutionizing our way of life in the past. In fact, our world is pervaded by worthless human-made cultural products which are nevertheless protected by copyright law in light of the extremely low threshold of creativity worldwide, from teasing TV commercials to dollar store magazines.

Even if it is accepted that AI-authored creations are more disruptive in a quantitative sense, these algorithmic works are most likely to be returned to the public domain eventually as copyright would only makes sense when the work for which copyright is sought is of some market value. Thus, if these works were indeed low in creative value and thus undeserving of commercial exploitation, the wise strategy for the rightholder is to abandon the entitlement and leave it in the public domain in order to benefit more or sustain less loss in maintaining and enforcing copyrights. In the unlikely scenario where they remain in the hands of the copyright holder, algorithmic works, by further unleashing competition in their market, may well act as a screening mechanism or “bad money”⁸³ in the economic meaning, i.e. to drive out the “bad money” or “worse money” made by human beings or other inferior AI systems, and induce the formation and development of the market for more valuable creative works. In the anticipated “post-scarcity” society or “artificial intelligence society”, an “accumulationist” model of competition will just stop making any sense.⁸⁴ To survive on the playing field, market players would have to provide more cultural products of higher quality and of more value, and thus contribute to the social progress of sciences and arts which is exactly what copyright law is aiming for.

Nor should we be blind to a more subtle vertical competitive relationship, although not in a strict competition law sense, between the programmer of AI and the investor, who is the supporting legal entity. The complexity of research and development of advanced (autonomous) AI projects often requires a substantial investment in, inter alia, infrastructure building and data collection.⁸⁵ This would mean that the AI programmer, who actually created the work, is not the author in the eyes of the positive law in most countries. Rather, it is the company who employed the programmer that is entitled to claim copyright in the AI programme. By locating copyright in the works created by an employer-owned AI system in the hands of the programmer, the altered “work made for hire” doctrine helps to remedy the distributional inequality which has long existed between the employer and the employee creator.

Although the employee creator may still eventually give away her copyrights in the works created by the AI programme through a transfer or pre-assignment contract with the employer, this legal arrangement, if nothing more, confers to her significant leverage in the negotiation process. From the perspective of the employer, the commercial value of the AI programme would be at its maximum when complemented with the unrestrained exploitation of the works made by the AI as an integral marketing package. Furthermore, to minimize transactional costs, an employer would tend to negotiate with the employee creator over the ownership of these by-products in advance. The reason is twofold: for one, conducting negotiations at the commencement of the AI project is simpler than with deals struck after the completion of the project; secondly, it avoids the potential problem of “holdup”⁸⁶ by the employee creator.⁸⁷ With all this in mind, the employee creator, who is often in a disadvantaged negotiating position, could well take advantage of this favourable situation and bargain for more attractive terms for the employee or ownership transfer contract. In addition, if the AI project were promisingly creative and valuable, the employer would be likely to invest more in the employee creator, which, in turn, would be an additional incentive for her to make more creations of higher value.

5 Conclusion

Policymakers and judges are satisfied with the status quo of copyright law because in the way it stands now it is fully capable of absorbing the latest developments in AI where the creative process of cultural production can still be attributed to a human creative actor. Nevertheless, this article is more concerned with a more remote but destined-to-happen future where AI is enabled to initiate creation independently of any human intervention. A “realist” approach to copyright law is taken, that is, to take copyright law as it is now with little prospect of significant change. This makes sense for two reasons: first of all, it is unlikely that the copyright landscape is going to change, at least in the near future;⁸⁸ on the other hand, the principle of technological neutrality cautions against an urge to rush into action in making significant reforms to positive copyright law in order to clear the way for disruptive technology.⁸⁹ Thus, the inquiry of this article is largely confined to the written texts of positive copyright law. To the extent that this article has a “reformist” agenda, it searches for the liberal space between the doctrines of copyright law, which may allow the re-construction of a dehumanized conception of authorship.

The remaining influence of the 18th century’s romantic authorship ideology is still intertwined with the popular understanding of authorship as a human prerogative. It works in an implicit way – neither the texts of international copyright treaties nor those of national copyright laws expressly deny the eligibility of AI for a membership in the “authors’ club”, although they do demonstrate noticeable hesitance or reluctance to recognize AI authorship. The case law established by the CJEU would require the work to reflect the “personal touch” or “personality” of the author, whereas the US has a more tolerant attitude.⁹⁰ The fact that policymakers or judges do not have the incentive to break the deadlock of silence and dispel such ambiguity around the legitimacy of AI authorship leaves open the possibility of alternative perspectives on this matter.

Three important observations may support a dehumanized authorship construct. First of all, a re-examination of the historical trajectory of the authorship ideology reveals that authorship, at the very outset, was really not about romanticized human creativity but rather for rapacious corporate exploiters (i.e. publishers or similar grantees) who fought for individual authors’ natural rights to claim moral and economic privileges in creations, but then despoiled these claims for their own benefit out of view. On the other hand, copyright law seems to be experiencing a gradual shift of focus from the creator of the work to the audience that the work targets with a very low threshold of creativity and deliberate judicial avoidance of aesthetic assessment. This de-centralization of the role or identity of the author may well leave workable doctrinal space for algorithmic authorship. Indeed, the dividing line between AI and human actors is more like an artifact – AI can be equally unpredictable in terms of creation by introducing elements of randomness, and conversely the creative process of the human brain can be highly algorithmic. They follow the same laws of creation during the process of cultural production. The nature of creation or authorship is never to make something ex nihilo, but rather it relies on an array of pre-established rules.

A crucial secondary theme developed from the premise of AI authorship is who owns these works created by AI. This article argues that an altered “work made for hire” doctrine, which assigns the ownership of AI-authored creations to the programmer of the AI programme, may serve as an ideal template to deal with the ownership issue. The reason for this arrangement is twofold: first of all, it avoids vesting ownership in an AI, which clearly could neither own any property nor be the subject of any rights; secondly, it reflects the basic reality that the author of these works is the AI rather than the AI programmer. Indeed, there is strong opinion against such legal fiction which contends that works made by AI should go straight into the public domain. To the extent that this article agrees that a robust public domain is beneficial in many ways, it casts doubts on the proposition that the removal of copyright protection for AI creations would be a de facto “plus factor” for the public domain.

On the one hand, the loss of possible benefits arising from copyright claims in the AI-authored works in the downstream of the value chain of an AI project may disincentivize the programmer to put the upstream AI system into the public domain. On the other hand, we should not lose sight of the pro-competitive side of the “work made for hire” approach. Horizontally, by conferring copyrights to mass-produced AI creations, it may well intensify market competition for cultural products in a “‘giant warehouse’ of intellectual commodities”;⁹¹ and vertically, it may help curb the distributional imbalance between the employee creator (i.e. the AI programmer) and the employer and put the former in a better negotiating position vis-à-vis the latter.