-
PDF
- Split View
-
Views
-
Cite
Cite
Barry Scannell, When Irish AIs are smiling: could Ireland’s legislative approach be a model for resolving AI authorship for EU member states?, Journal of Intellectual Property Law & Practice, Volume 17, Issue 9, September 2022, Pages 727–740, https://doi.org/10.1093/jiplp/jpac068
Close - Share Icon Share
Barry Scannell is a practising solicitor and consultant in Ireland’s top-ranked Technology and Intellectual Property Group in William Fry LLP. Barry is a legal subject matter expert and thought leader on AI and frontier technologies and is a PhD candidate and scholarship recipient in University College Cork, researching artificial intelligence and copyright in the College of Business & Law.
This article considers the issue of AI authorship in the EU, identifies how this issue is problematic, and sets out to ascertain how much human involvement is required under EU law for a work to qualify for copyright protection. In discussing this issue, an analysis has been carried out on the literature on what constitutes an ‘AI-generated work’, due to the fact that there is a broad spectrum of what could be considered as such, with varying levels of human involvement.
An examination is undertaken of how certain EU Member States have approached authorship in their copyright legislation and carries out a doctrinal review of EU legislation and CJEU case law for the EU approach to authorship, identifying human authorship requirements.
A solution to the AI authorship question is proposed by suggesting that Irish legislation, which confers authorship of computer-generated works on the person who made the ‘necessary arrangements’ for the generation of those works, could be the basis of a model which EU Member States might follow, despite commentary suggesting that the Irish legislation is not compatible with the EU copyright acquis.
Introduction
The matter of how the EU addresses the issue of AI authorship requires an analysis of EU Law and of the requirement for an author to be human. If the human author requirement persists in EU law, works created utilizing AI may fall into the public domain.
An example of what current AI is capable of is the Dall-E system from OpenAI. In their paper on the development of the technology, Ramesh et al note how ‘[w]hen given the caption “a tapir made of accordion…” … The model appears to draw a tapir with an accordion for a body, or an accordion whose keyboard or bass are in the shape of a tapir’s trunk or legs.’ The authors go on to surmise that this ‘suggests that it has developed a rudimentary ability to compose unusual concepts at high levels of abstraction.’1 The ‘AI-Authorship issue’ (discussed below) as applied to this example would be: who is the author (or indeed, is there an author) for the resulting image of a tapir with an accordion for a body? Copyright law often requires a human author for a work to attract copyright protection. In this instance, a team of scientists trained an AI2 to produce images based on text prompts of users. Leaving aside the extraordinary creativity and ingenuity of the scientists behind the creation of the AI, one of the primary considerations must be whether a human asking for a picture of ‘a tapir made of accordion’ is sufficiently creative for the resulting image to have an author and, therefore, copyright protection.
The concept of ‘AI-generated works’ is discussed in this piece on the basis that (at the time of writing), we do not yet have the technology whereby an AI can fully autonomously generate artistic or musical works with zero human involvement. At present, some type of human involvement is required in the AI creative process. A view is taken that, as such, any works generated by an AI are, in fact, ‘AI assisted works’. References in this article to AI-generated works should be construed accordingly.
At the lower end of the scale in computer-generated works is where AI is very much used as an assistive tool. A good example of this would be the almost ubiquitous use of ‘auto-tune’ AI in modern musical works, which corrects off-key singing, amongst other things. McCutcheon posits that generally speaking, ‘computer-assisted works’ will usually be authored works, and uncontentious, since software is used merely as a tool to produce the work. As such, the software carries out the user’s instructions and the user is ultimately responsible for the form of the work at the output stage, such as using word processing software used to compose literary works.3 Ginsburg and Budiardjo echo this sentiment, stating in such circumstances, ‘that creator-user is necessarily the only author of the work produced through the aid of that machine.’4 Autotune would be an appropriate analogue of this example in the music industry, as the software simply (to borrow McCutcheon’s expression) ‘does the user’s bidding and the user is largely responsible for the form of the work’.
McCutcheon notes that there is a continuum at which the opposite extreme of AI-generated works is inhabited by ‘autonomously-generated output’. Here, McCutcheon identifies autonomously generated works as ones where rules and parameters in which the software operates are set by the programmer, and the actual form of the output is ‘unpredictable’.5 The question is whether a work of art or piece of music falling within McCutcheon’s definition of an autonomously generated work would be considered to still be a ‘computer-assisted work’, given the level of human involvement which is required in training the AI and programming it. Taking the accordion-shaped tapir example above, using McCutcheon’s rationale, the resulting image would be considered ‘autonomously generated’, despite the level of creativity which was required to come up with the abstract concept of a tapir shaped like an accordion.
The unpredictability of output is not the defining feature of an ‘autonomous’ work; rather, it is the level (or lack thereof) of human involvement. In McCutcheon’s defence, this was her view held in 2013, before the serious advent of computer-generated works. Jackson Pollock’s artworks are very much protected by copyright, but he only had a general idea of how a splatter of paint on canvas might turn out—the result was unpredictable. It is arguable that unpredictability is a central feature of art and creativity. As such, the unpredictable nature of the finished product of a work should not be a threshold to which we subscribe.
‘Computer-assisted’ is a spectrum, which can run from the computer equivalent of using a pen, such as word processing software, to a vocoder, which is a category of voice codec that analyses and synthesizes the human voice signal for audio data compression, multiplexing, voice encryption or voice transformation, to autotune (which is virtually ubiquitous in modern music), to providing strings accompaniments and harmonies to melodies. Ramalho, positing that in addition to intelligence, AI also presupposes autonomy, notes that ‘[a]utonomy can be defined as a matter of scale, according to the level of human involvement, and not necessarily in the binary state autonomous-non autonomous.’6
The JURI draft report on the proposed EU AI Act suggested an amendment which proposed a definition of ‘autonomous’ in the context of AI systems, as meaning ‘an AI-system that operates by interpreting certain input and by using a set of pre-determined objectives, without being limited to specific instructions, despite the system’s behaviour being constrained by, and targeted at, fulfilling the goal it was given and other relevant design choices made by its provider’.7 This raises the need to point out an important distinction. While there might be limited ‘technical’ autonomy of this type from a product safety or even a computer science perspective, the type of autonomy for free and independent creative choices required for copyright law is a different kind of autonomy.
As such, there is a danger that in the conversations about legislating for AI and copyright, the concept of technical autonomy in an AI system is confused and conflated with the concept of autonomy inherent to creativity and copyright. This means that it is important to ensure that academic discourse, recognizes that we are potentially referring to different paradigms when discussing AI autonomy. There may well be a good deal of AI systems that could fall under the proposed definition of ‘autonomous’, but from a creativity and copyright perspective ‘fulfilling the goal it was given and other relevant design choices made by its provider’8 would potentially fulfil copyright requirements whereby the AI provider could be considered the author. For an AI creation to be truly ‘authorless’, the design choices involved in the output must be the result of the AI independently making those choices, without the AI provider/creator making those choices. It would need to create independently and without parameters.
We are left to consider the central question regarding authorship in copyright law: how much human involvement was there and how much creativity was involved? As Guadamuz notes, we are unlikely to see any move away from personhood as a requirement for authorship, and we will not see any sort of allocation of rights towards machines and animals.9 This said, he goes on to question whether the automatous data analysis of numerous works of art of portraits are sufficient to warrant protection. For such a work to be protected under the EU copyright acquis, there must be originality in the composition of the program. The problem becomes apparent if most of the creative decisions are being performed by the AI system.10 Hugenholtz and Quintais appear to agree with the idea that the most important question is what level of human input is required, saying what is central to the analysis is not the intelligent machine, but the role of human beings in the AI-assisted creative process. They question whether this human role is sufficient to elevate the result of this process, the AI-assisted output, to the status of a copyright-protected work.11
To consider the issue, firstly an analysis of existing authorship rules in certain EU Member States is carried out. Then, an analysis of the human authorship requirements in the EU, at both a legislative and judicial level is undertaken. Finally, the proposition is put forward that the Irish legislation on the authorship of ‘computer-generated works’ could be a possible model upon which EU Member States could base their own legislation as they address the AI-authorship issue.
Is AI authorship an issue?
AI authorship is an issue because there is no agreement on whether AI-generated works can be protected by copyright. While technology is not at a stage in which AIs can generate works autonomously, it is at a stage where human involvement can be minimized. It is not clear as to how much we should consider AI as merely a tool used by humans, or whether the AI tail is wagging the dog of human creativity. A human author is a common requirement for copyright protection. Yet, it is not clear if EU Member States’ copyright law would consider a work to be generated by a human when certain AI technologies are used, particularly if approaches such as that of McCutcheon to ‘autonomously-generated output’ as described above prevail.
It is necessary to consider why it is important that AI-generated works should be protected by copyright. In a humorous anecdote, Bridy explains that, in 1956, two mathematicians named Martin Klein and Douglas Bolitho programmed a computer to compose Tin Pan Alley songs. Bridy quotes Dr Klein writing in the June 1957 issue of Radio Electronics Magazine: ‘Last Spring, we set out to prove that if human beings could write popular music of poor quality at the rate of a song an hour, we could write it just as bad with a computing machine but faster.’12 It turned out that the computer could churn out four thousand songs in an hour—in 1957.
Given the requirement for a human author for copyright protection in the EU copyright acquis, it will be difficult for machine-generated works to be protected by copyright. As a result, machine-generated works will fall into the public domain, without any economic protections. This will potentially adversely affect human creators, but it could also stifle EU innovation, if the fruits of investment into EU technologies cannot be protected under EU copyright law.
In terms of how human creators may be adversely affected by certain AI-generated works, AI can vastly out-compete humans when it comes to generating works. Music industry bodies such as the International Federation of the Phonographic Industry (IFPI) have noted that a central pillar of any policy aimed at stimulating developments in AI should be to support the creative sector, and that a reduction in the protection of works that reduces incentives for the creation of new works would ultimately harm innovation and investment in AI processes.13 Similarly, in relation to AI and copyright, the International Confederation of Societies of Authors and Composers (CISAC) has noted that the raison d’être of copyright is to incentivize creation, and that by encouraging investment in the development of intellectual resources and the creative activity of authors, copyright creates value that drives the economy. CISAC believes that copyright protection should be provided each time there is creation, ‘whatever its form, however it is achieved.’14
As we live in a free-market capitalist system, driven by supply and demand—over-supply of anything will drive prices down. Even if the quality of AI-generated works is not as good as human-generated works, by sheer volume alone, AI-generated works could damage the precarious economics of the music industry.
What’s more, if AI-generated works are not protected by copyright, they will be public domain, and will not require music royalties to be paid for their use. Therefore, not only would we have a situation where the market is being saturated by AI-generated works, but those AI-generated works would be free, competing against human-generated works for which music royalties must be paid. High-quality public domain AI-generated works flooding the market would create a situation of uneven competition in the EU market, where human creators would need to compete against entities churning out AI-generated works.
In terms of how innovation can be damaged by not addressing the AI authorship issue, if there is no economic protection for the results of the investment of time, money and resources required by AI, there is no incentive to make those investments into developing AI. While not in the realm of copyright, a germane example is how Google’s parent company, Alphabet, created the commercial venture Isomorphic Labs, which utilizes AI technology in order to solve important medical challenges such as protein folding, and which computationally simulates biological systems in a matter conducive to drug discovery.15 It must be considered whether this important technology would ever have been developed if there was no opportunity for profit. Similarly, we must ask, how many works of art and of music might civilization lose out on if works created with the assistance of AI are not protected?
As noted by Guadamuz, because the main aim of intellectual property is the protection of the fruits of the human mind, these fruits of the mind are afforded a set of limited ownership rights allocated to natural and legal persons. This type of protection is personal in nature and, because of this, the argument is that there is no such thing as non-human intellectual property rights.16 Accordingly, in determining whether AI-generated works can attract copyright protection, it is important to embark on an analysis of the requirements for copyright protection and how those requirements have been interpreted by legislators and the courts in the EU. There are different elements required for copyright protection—originality, creativity, the fixation of the work, and, it would appear, a human author.
Our laws should clarify that AI-generated works are protected by copyright. Where AI-generated works are protected by copyright, the creators who utilize the technology could avail of royalty income, the current copyright regime could regulate the use of AI-generated works, and the market could be maintained at a more competitive level than the alternative.
The ‘AI authorship issue’ is that copyright law in EU Member States is not clear on whether AI-generated works could be protected by copyright. Such works are, in fact, capable of being protected by copyright, but it is suggested that this needs to be clarified in Member State legislation.
EU law on authorship
In the EU, the concept of authorship does not have a general definition.17 Usually, a work’s authorship is recognized, and the copyright’s ownership is vested in the natural person who created the work.18 EU legislation does not explicitly require that only original creations by humans can avail of copyright protection. However as noted by Vehar and Gils, several copyright-related directives require a copyright work to be the authors “own intellectual creation”.19
Hugenholtz and Quintais point out that even though EU copyright law does not expressly state that a human author is required for copyright protection, the EU’s ‘“anthropocentric”’ focus on copyright protection is apparent in several aspects of the EU copyright acquis.20 This is echoed by Verhar and Gills who suggest that copyright law in Germany is also anthropocentric.21
EU member state approaches to authorship
A brief glance at authorship considerations of EU Member States (in this section we consider the copyright laws of Spain, Germany, France and Italy—the Irish approach is discussed further below), reveal certain similarities. The requirement for an author to be a natural person is a common prerequisite for copyright protection, and that only ‘works of the mind’ can be protected. It also appears to be quite common that Member States consider the author of a work to be the person identified as the author, unless proven otherwise.
In Spain, the copyright in a literary, artistic or scientific work shall belong to the author thereof ‘by virtue of the sole fact of its creation’.22 ‘Author’ is defined as the ‘natural person who creates any literary, artistic or scientific work’.23 Under Spanish law, a person is the author of a work if they are identified as such, unless proved to the contrary.24
In Germany, the person who creates the work is deemed the author.25 Unless proven otherwise, the person identified as the author of a work shall be deemed to be the author of the work.26 German copyright law stipulates that only ‘personal intellectual creations’ can be protected.27 Verhar and Gils note that the prevailing opinion is that only humans can be authors of personal intellectual creations and that purely mechanical creations are not protected.28
The French law on copyright is particularly interesting with regards to AI-generated works, given that ‘[a] work shall be deemed to have been created … by the mere fact of realisation of the author’s concept, even if incomplete.’29 This would suggest that the author of a work is the person who merely comes up with a concept (such as an accordion shaped tapir) which is then realized in some way. French copyright protection seems similarly anthropocentric, granting protection to the ‘author of a work of the mind’30 (l’auteur d’une oeuvre de l’esprit). As in Spain and Germany, the author of a work shall be deemed to be the person identified as author, unless the contrary is proven.31
In Italy, like France, ‘works of the mind’ are protected, although Italian law adds the requirement that it applies to such works ‘having a creative character’.32 Italian copyright law protects works which are created constituting ‘the particular expression of an intellectual effort’33 (language which we see repeated in EU legislation). Once again, a person who is identified as the author shall be the author unless proven otherwise.34
It must be considered what the similarities in EU Member State approaches to copyright mean for the AI authorship issue. Given that there are no AIs making independent and autonomous creative choices in the creation of works, coupled with legislative references to ‘natural persons’ or ‘works of the mind’, machines themselves cannot be authors. Creators require certainty in the copyright of their works, and legislative provisions which deem whoever is merely identified as the author of a work as that workss author lacks that requisite certainty because this is only the case “until the contrary is proven”.
If a company provides AI software to musicians, for example, to assist them in composing songs and providing backing tracks or creating stems of certain tracks,35 users of that product will want the certainty that if they use AI in composing music, their work will be protected by copyright. Indeed, the company providing the AI software, which has invested in and innovated the technology, will want to ensure that it has a product which will allow its users create works protected by copyright, because if the resulting work cannot be exploited economically, then the value of the product shall decrease substantially.
There is a gap in EU Member State copyright laws: while it would appear to be possible that works resulting from a human using the AI and making creative choices could be protected by copyright, there is no certainty that this would be the case, especially if the Sword of Damocles of ‘unless proven to the contrary’ hangs over those works.
Without needing a drastic change to Member State copyright laws, the certainty craved by creators, innovators and investors can be provided. This is discussed further below under the Irish approach to computer-generated works.
EU approach to authorship
It is important to consider the EU approach to authorship because, in the absence of clarity in EU Member State copyright laws, we are likely at some point to see the Court of Justice of the European Union (CJEU) must consider the AI authorship issue.
Prior to the development of CJEU case law on the topic of authorship, there seems to have been a concerted effort in the EU to consolidate and harmonize copyright across the EU. As part of that harmonization project, it seems as though the divergent strands of what constitutes originality (and by extension, authorship) were not considered to represent a risk to that project. The advent of AI technology may have added disharmony to the EU copyright Acquis, as we could potentially see a situation where machine-generated works are protected in some Member States, but not others, causing problematic market a-symmetry.
As Hugenholtz and Quintais suggest, the current EU copyright framework is mostly silent on questions of copyright subject matter and authorship. They note that the 2020 report on AI and IP Framework prepared for the European Commission (however reflecting the views only of the authors) (‘the AI & IP Framework Report’) suggests that Article 1 of the Term Directive36 comes close to a general definition by referring to copyright subject matter as ‘a literary or artistic work within the meaning of the Berne Convention’.37
Despite the anthropocentric approach of the EU copyright acquis noted above, Rosati notes that, in the EU, copyright directives do not address the issue of a human authorship requirement (excepting cinematographic and audio-visual works, computer programs and databases).38 In addition to cinematographic and audio-visual works as noted by Rosati, the Term Directive deals with photographs which are ‘original in the sense that they are the author’s own intellectual creation’.39 Interestingly, the Term Directive says that regarding photographs which are original in the sense of the author’s own intellectual creation, ‘[n]o other criteria shall be applied to determine their eligibility for protection’.40 Importantly, Recital 16 of the Term Directive says that with regards to photographs, ‘[a] photographic work within the meaning of the Berne Convention is to be considered original if it is the author’s own intellectual creation reflecting his personality [emphasis added]’. While the other directives dealt with the question on an author’s ‘own intellectual creation’ basis, the CJEU has paid special attention to the personality aspect of authorship, as we will see in the analysis of CJEU decisions on the issue below.
Rosati succinctly distils the intricacies of EU copyright legislation as it pertains to authorship, pointing out that nothing is preventing Member States legislating for the protection of AI-generated works, and that Member States can protect works created by non-human authors in some instances.41
As the AI & IP Framework Report identifies, in its jurisprudence, the CJEU regularly seeks guidance from the Berne Convention’s concept of work, and the Report’s authors opine that the EU concept of a ‘work’ is therefore intrinsically linked to that of Article 2(1) of the Berne Convention.42 The AI & IP Framework Report identifies that the EU acquis expressly harmonizes three or four specific categories of copyright-protected subject matter: computer programmes, databases, photographs and—possibly—works of visual art, which are protected as long as they are original (i.e. the author’s own intellectual creation).43
Infopaq
In Infopaq, the nature of the activity the subject the subject of the dispute was akin to a rudimentary form of text and data mining. However, it was not the text and data mining activity itself that was at issue, in fact, neither party felt that authorization was required for this activity.44 Rather, it was the act of reproduction as contemplated in article 5(1) of the InfoSoc Directive which was inherent to text and data mining which was at issue.45
Rosati says that until Infopaq, apart from the types of works outlined above where the originality standard had been harmonized, it was believed that ‘… EU Member States remained free to define the conditions of copyright protection under their own legal regimes.’46
Vehar and Gils point out that ‘[t]he CJEU’s Infopaq decision extended the requirement of “the author’s own intellectual creation” to copyright within the meaning of Article 2(a) of the Copyright Directive and ultimately all types of copyright, emphasizing a harmonized approach to EU copyright law.’47
In the Infopaq decision, the court stated that ‘[i]t is … apparent from the general scheme of the Berne Convention, in particular Article 2(5) and (8), that the protection of certain subject-matters as artistic or literary works presupposes that they are intellectual creations’.48 The Court went on to outline how in the Software Directive,49 Database Directive50 and Term Directive,51 computer programs, databases and photographs respectively ‘are protected by copyright only if they are original in the sense that they are their author’s own intellectual creation.’52
Importantly, the CJEU decided that the InfoSoc Directive53 was based on this principle also, saying this was ‘evidenced by recitals 4, 9 to 11 and 20 in the preamble thereto’,54 going on to say that ‘[i]n those circumstances, copyright within the meaning of Article 2(a) of Directive 2001/29 is liable to apply only in relation to a subject-matter which is original in the sense that it is its author’s own intellectual creation [emphasis added].’55 Therefore, the Court held, the various parts of a work are protected by the reproduction right in the InfoSoc Directive ‘provided that they contain elements which are the expression of the intellectual creation of the author of the work. [emphasis added]’56
It is interesting that the CJEU held that only ‘elements’ which are expressed intellectual creation are protectable. An AI may have ‘created’ a work, but if a human curated the training data or arranged the AI’s output with lyrics or instrumentation. While these are only constituent ‘elements’ of the work taken as a whole, it would appear, pursuant to Infopaq, that these elements ought to be sufficient to attract copyright protection.
Ultimately, in Infopaq, the CJEU held that the first two acts of reproduction at issue could be considered transient if they were automatically deleted and excepted from the reproduction right in the InfoSoc Directive. The third act of reproduction, namely the creation of the text file, was not ruled upon by the CJEU and was referred back to the Danish Courts. The final act of reproduction, whereby the results of the data capture process were printed out, was held to be a non-transient reproduction, as it was not automatically deleted.57
The result of this landmark decision, as identified by the AI & IP Framework Report, was that the CJEU extrapolated from the previously piecemeal harmonization of EU copyright law a general, autonomous concept of EU law of the work as ‘the author’s own intellectual creation’.58
Painer
The Painer59 decision followed from Infopaq, cementing the latter as a seminal CJEU decision on copyright law and elaborating on the Court’s rationale in Infopaq.
In Painer, the CJEU made clear it felt that the question on copyright protection was answered in Infopaq,60 in that in order to attract copyright protection, a work must be ‘original in the sense that it is its author’s own intellectual creation’. The CJEU in Painer went on to explain that ‘[a]s stated in recital 17 in the preamble to Directive 93/98, an intellectual creation is an author’s own if it reflects the author’s personality’,61 adding that this ‘is the case if the author was able to express his creative abilities in the production of the work by making free and creative choices’.62 The CJEU explained that a portrait photographer could make free and creative choices in taking portraits, saying that by ‘making those various choices, the author of a portrait photograph can stamp the work created with his ‘personal touch’.63
Therefore, the CJEU in Painer effectively formulated a three-step-test in deciding whether a work attracted copyright protection, by asking was the work (i) a work of the author’s own intellectual creation; that (ii) reflected the author’s personality; by the author (iii) making free and creative choices?
Interestingly, in her opinion, Advocate General Trstenjak pointed out that, according to the Term Directive, ‘only human creations are therefore protected’.64 As Hugenholtz and Quintais point out, this might be the clearest formulation of the principle that an author must be a human, and that Trstenjak’s conclusion was, after all, approved by the CJEU.65
As such, in circumstances where an author must be a human, current EU law would appear to be that autonomously AI-generated works cannot be protected by copyright. However, as outlined above, given that we are not at that level of technological sophistication, creative works created with AI will still require a human to be involved. The issue, therefore, is to ensure that the human responsible for the AI-generated work is considered the author of that work.
Funke Medien
The Infopaq requirements were elaborated in subsequent case law and, most recently, consolidated in in the Funke Medien decision.66 There, the CJEU stated that in deciding whether the documents the subject of the case were capable of being regarded as ‘works’ within the meaning of Article 2(a) and of Article 3(1) of the InfoSoc Directive, the national court should ascertain whether ‘the author was able to make free and creative choices capable of conveying to the reader the originality of the subject matter at issue, the originality of which arises from the choice, sequence and combination of the words by which the author expressed his or her creativity in an original manner and achieved a result which is an intellectual creation.’67 Importantly however, the Court went on to say that for the purposes of applying the principles which were established in Infopaq for determining whether or not something was a work, ‘the mere intellectual effort and skill of creating [something] are not relevant in that regard’.68
The Irish legislative approach: potentially a model for AI authorship legislation?
In the EU, the concept of authorship does not have a general definition,69 and usually, a work’s authorship is recognized, and the copyright’s ownership is vested in the natural person who created the work.70 As a 2022 study of the European Commission on Copyright and New Technologies (‘the 2022 Commission Study’) pointed out, this principle knows certain exceptions both in civil and common law countries. Such exceptions are more common in the common law tradition, due to the economic logic supporting copyright protection, whereas civil law countries animated by a personhood tradition are more reluctant to admit exceptions.71
The Irish and UK approach
Despite the near universal requirement for a human originator of a work, there are jurisdictions such as Ireland and the UK,72 which have specifically legislated for computer-generated works and how authorship of these works can be defined, with the Irish legislation stating that ‘in the case of a work which is computer-generated, the person by whom the arrangements necessary for the creation of the work are undertaken.’73
Perhaps ironically for a piece of copyright legislation, the Irish Copyright and Related Rights Act, 2000, borrowed very heavily, with many similar (or identical) provisions, from the UK’s Copyright, Designs and Patents Act, 1988. One such example of heavy borrowing and similarity (but not identicality) is the Irish provision on granting authorship for computer generated works. Despite this, however, the Irish legislation is concentrated on here, due to Ireland still being a Member State of the EU.
The Irish legislation is proposed as a model because it attempts to deal with computer generated works, but as a model, it is not perfect, and this common law approach could very much be assisted by the civil law considerations on the issue.
The AI & IP Framework Report points out that British and Irish legislation have been criticized as begin incompatible with the EU acquis.74 The report notes that ‘a national rule that accords copyright protection to subject matter that does not meet the standard of “the author’s own intellectual creation” is hard to reconcile with the CJEU’s case law that implies that the notion of a “work” is fully harmonized and therefore does not allow national laws to accord copyright protection under more lenient conditions.’75
The reason Ireland’s legislation is not a perfect model on which to deal with the AI authorship issue, is that section 2(1) of the Irish Act defines ‘Computer-generated’ as ‘in relation to a work, means that the work is generated by computer in circumstances where the author of the work is not an individual’ (emphasis added). This is to be contrasted with s.178 of the UK’s Copyright, Designs and Patents Act 1988, which states ‘computer‐generated’, in relation to a work, means that the work is generated by computer in circumstances such that there is no human author of the work.”
As technology does not yet allow for situations where the author of a work ‘is not an individual’, there can be no ‘computer-generated works’ as envisaged in the Irish legislation. The ‘computer-generated’ definition sought to deal with the thorny issue of who is the author when a work is created by/with the assistance of a machine, because in such circumstances, the author of the work is not immediately apparent. With AI-generated works, it is not always clear whose ‘work of the mind’ it is or whose creative choices were made which led to the creation of the work. This is due to the profoundly complex nature of the tool used to create AI-generated works. A painter using a paintbrush is clearly the one making the creative choices, because the person who made the paintbrush has no direct impact on how the painter’s finished work will turn out.
Vehar and Gils note that under the German law on the issue of how much human involvement is required, copyright protection depends on whether the creative activity of humans or the work of AI is centre stage. Does AI only serve the work by a human or does it take over the creation?76 The authors point out that generally, the literature suggests a person will be protected by copyright in a work if they ‘created the essential basic pattern of the work and makes a selection from several versions generated by the computer.’77
Creativity in training
With AI, the people who train and create the AI have an enormous impact in how the AI generates work. Those people choose datasets and training material, and fine-tune and modify the AI so that it creates the desired results.78 In AI technology which produces artistic works, selecting the right music or the right pictures or paintings is an art and a creative pursuit in-and-of-itself. This said, the user of a ‘trained’ AI which can create pictures based on word prompts, who suggests the phrase ‘a tapir made of accordion’, is also displaying a ‘work of the mind’ and is making free and creative choices. With AI platforms such as AIVA, where a user can create music tracks using their pre-set algorithms to compose music in pre-defined styles and can modify items such as which musical instruments are used, the tempo, etc, users are still making creative choices, and the resulting pieces of music are still works of the mind, albeit with AI assistance.
Due to the level of intricacy, creativity and innovation required to design the tools which create AI-generated works, and the ease with which users of AI can create works simply with an idea which the AI then fixes into a tangible medium, it is not always readily apparent who the author could or should be in an AI-generated work.
Ginsburg notes in her article,79 that Ricketson addressed the then-recent 1991 WIPO Proposal for Protection of Computer-Produced Works, which considered whether the original ownership of copyright should be vested in the physical or legal entity who undertook the arrangements necessary for the creation of the work (as was the case in the UK’s 1988 legislation). While ultimately WIPO did not end up adopting the Proposal for Protection of Computer-Produced Works, several other jurisdictions in the Commonwealth (and Ireland) did follow the lead of the UK.
Germane to the question of whether or not humans are involved in machine-generated works, Ricketson observed that the WIPO proposal assumed that such machine created works actually did have human authors, but the ripple was that it was impossible to identify their individual contributions.80 Due to the ‘black box’ nature of machine learning, where the creators of advanced algorithms do not know the calculations an algorithm is performing at a given time in response to data, and because of which attribution to specific data sets is extremely difficult, Ricketson’s observation was extraordinarily prescient. The very selection of training data for a generative AI system is a creative act, reflecting the trainer’s personality. Francois Pachet, one of the trailblazers in AI-generated musical works and the man behind the famous ‘Daddy’s Car’ song which is an AI-generated work emulating a Beatles’ song,81 said of the song, Daddy’s Car, ‘The song … was generated with Flow Composer, using a corpus of 45 Beatles song (from the latest period, considered usually as the richest and most singular in the recording history of the Beatles)’.82 Creating the training set for the AI for this track wasn’t a simple matter of randomly selecting a bunch of Beatles tracks, but rather involved an artistic choice. Even deciding to use Beatle’s songs could be considered a creative choice. There are 213 Beatles songs in the world, deciding which of those 213 songs to use required deep considerations into how the AI creators wanted the resulting outputs to sound.
Proposal for a legislative model
As a starting point, it is proposed that in order to utilize the Irish approach to authorship of computer-generated works, that the Irish definition of ‘computer-generated’ should be defined as ‘in relation to a work, means that the work is generated by computer based on inputs which are creative efforts of the mind’. Such a definition for ‘computer-generated’ is sufficiently broad so that either the creator of the AI or the user of the AI (or both) could be considered the author of the resulting work. It is common course that in all creative pursuits, authorship and shares of authorship are agreed between authors and potential authors, often contractually, and such a definition would on the one hand, grant the certainty that is missing from Member States’ authorship provisions, but on the other hand, be broad enough so that the longstanding tradition of authorship allocation being dealt with privately remains intact and is not dictated by legislation.
Another nuanced difference between Irish and English legislation is that, in the UK, moral rights are limited for computer-generated works, but this is not the case in Ireland,83 and it is not clear why there would be a need to limit moral rights in AI-generated works.
As the AI & IP Framework Report points out, ‘if the existence of a “work” is conditional upon human authorship, the existing statutory language seems to suggest that the Irish and British regimes allocate authorship to productions that would not qualify as “works” according to EU copyright law standards.’84 Making this change to the definition of ‘computer-generated’ would make the Irish model more compatible with the existing EU copyright law acquis by requiring a human intellectual effort.
Therefore, the Irish model on which EU Member States could legislatively resolve the AI authorship issue would be a definition of authorship whereby, in relation to computer-generated works, the author would be the person by whom the arrangements necessary for the creation of the work are undertaken.
This would allow the user of an AI platform to generate a work with relatively little creative input, but still be deemed the author of the resulting work, because the creation of the AI itself required ‘creative efforts of the mind’. It also adds the certainty which is currently missing from EU Member State laws on the AI authorship issue, in that, under the current regime, it might be debatable whether a user displayed the requisite element of creativity or intellectual effort to warrant copyright protection, but there can be no debating that creating an AI capable of generating works of art required creative and intellectual effort.
Such a model effectively allows for the deferral of the intellectual effort, if necessary, from the creator of the AI to the user. This deferral is required because AI technology has made it so easy to fix a work, that it is often the case that relatively little intellectual effort may be required to create that work. The proposed model moves copyright law into the era of AI, by recognising that with the help of machines, realising human thoughts and dreams and ideas is becoming easier and easier. However, just because the fixation of ideas is becoming easier, that does not mean the resulting art is no less important.
The destabilization effect of copyright law’s authorship approach has been noted by Boyden, by obscuring the connection between the creative process and the work. Boyden’s concern is that once this happens, it will not be readily apparent that all of the creative elements of a work derived from the mind of a human author.85 This is the central problem when trying to ascertain human authorship of machine-generated works. How certain will courts need to be in order to establish the nexus of authorship between the work generated, and the spark of human creativity? While the technology may not be (and may never be) at the level where zero human involvement is required to allow a machine create something, the disruptive nature of computer-generated works to copyright law is noted by Kaminski, who suggests that even if humans are not removed from the equation, the technology removes or greatly distances, a human author from the work.86
‘Necessary Arrangements’ and the deferral of the ‘Intellectual Effort Requirement’
There is only one main authority in the UK on the issue of computer-generated works, which predates the UK legislation—Express Newspapers v Liverpool Daily Post,87 where the Liverpool Daily Post published the winning sequences of cards similar to lottery cards which had been distributed by Express Newspapers, with the latter seeking an injunction for copyright infringement. The contention of the defendants was that the sequences could not be protected by copyright because they had been generated by a computer, and as such, there was no copyright to infringe.
Awarding the injunction, the High Court held that the computer here was not the ‘author’, but merely a tool which produced the number sequences on the instructions of the programmer, with Whitford J stating: ‘The computer was no more than the tool … It is as unrealistic as it would be to suggest that, if you write your work with a pen, it is the pen which is the author of the work rather than the person who drives the pen’.88
Adrian suggests that Whitford J’s analogy here could be used to confer copyright ownership to the user of the program, and not to the programmer.89 This said, it could be argued that s.9(3) of the Copyright, Designs and Patents Act 1988 does this already.
In the UK, the Whitford Committee considered copyright reforms which led to the 1988 Act, including the computer-generated works provision, noting in relation to the latter that ‘the author of the output can be none other than the person, or persons, who devised the instructions and originated the data used to control and condition a computer to produce a particular result’.90 If the proposed Irish model was adopted, the ‘author of the output’, as envisaged here by the Whitford Committee, could either be the creator/trainer of the AI or the user of the AI.
Applying this statement from the Whitford Committee to AI; ‘devising instructions’ for a computer program in the 1970s is not the same as devising instructions for an AI system. The former requires strict line-by-line code and rules, whereas a machine learning system merely needs a general set of parameters. In the age of AI, what is ‘controlling and conditioning a computer’ is unclear. It could be the trainer of the AI or it could be the user, and it will depend on the level of choice and involvement the AI system permits the user.
The nature of the inputs into a program in the context of ‘necessary arrangements’, which generates an artistic work was considered briefly in Nova Productions v Mazooma Games, where Jacob LJ opined, in relation to a person playing a video game that ‘[t]he player is not, however, an author of any of the artistic works created in the successive frame images. His input is not artistic in nature, and he has contributed no skill or labour of an artistic kind. Nor has he undertaken any of the arrangements necessary for the creation of the frame images. All he has done is to play the game.’91
Undertaking the necessary arrangements (as per Irish and UK legislation), does not mean that the programmer can be the only author of works produced by that programmer’s software. Indeed, if a company instructed an employee to use a computer program to create works, then under the work for hire doctrine, it would be the company which is the author. If a twelve-year-old uses an AI interface to create a work, then that child could be the author of the work, having undertaken the ‘necessary arrangements’ to create the work.
It is entirely unclear what constitutes the ‘necessary arrangements’. Is clicking a button which says ‘GENERATE’ sufficient? If one was to click a button which generates a computer-generated work, that work only exists because a person clicked that button. However, somebody must have made the necessary arrangements at some point, and if the user of an AI’s actions is not enough to constitute ‘necessary arrangements’, then the creator of the AI itself would be the person who made those arrangements.
The proposed model ‘defers’ the intellectual effort requirements. Ginsburg and Budiardjo deal with the concept of a user and an AI designer by considering there to be two potential creators, an ‘upstream creator and a ‘downstream creator.92 By ‘upstream creator’ they mean a participant in the process who contributes to the work’s creation, but does not cause the final manifestation of the work, referring to how in Burrow-Giles Lithographic Co. v. Sarony while Sarony set the scene and posed Oscar Wilde, he did not operate the camera.93 The authors explain that by ‘downstream creator’ they mean the person responsible for the last steps required to create the work, suggesting that continuing the example, Sarony’s cameraman would have filled that role by choosing when to press the shutter to fix the image. In the context of AI creative works, this would be akin to an AI system being designed by a person who could be deemed author and used by a person who could also be deemed author—clicking that ‘GENERATE’ button.
Ginsburg and Budiardjo note that the downstream actor is less likely to be attributed as author if there are fewer options available,94 thus it may be the case that AI designers limit the options available to users of AI systems so as to ensure that users cannot be considered authors, but the designers could be.
As we have seen in the discussion of the case law above, this threshold might be a low one to meet. Taking a piece of music as an example, it could simply be the case that being able to select the tempo, the genre, the musical instruments to be used, the ‘mood’ of the piece would be enough to impart enough originality and creativity into a computer-generated work so that the ‘necessary arrangements have been made’.
If a computer-generated work does not allow for a user of the software to impart this element of creativity into the AI-generated work, then presumably, the AI creator would be the potential author. The issue, however, is that with AI, the types of convolutional neural networks employed in generative AI, are not ‘programmed’ in the traditional sense: they are instead trained.
UK IPO consultation
The UK appears to be of the view that in relation to offering copyright protection for computer-generated works, it should only be considered ‘if the benefits to society outweigh the costs’.95 The UK IPO’s Consultation stage impact assessment on Artificial Intelligence and Intellectual Property (‘the UK IPO Consultation’) document96 suggests three choices which are available to the UK in relation to copyright protection for computer-generated works, (i) make no legal change, (ii) remove protection for computer-generated works and (iii) replace the current protection with a new alternative.97
Under option (i), making no change would allow computer- generated works to continue to benefit from existing copyright protections.
Under option (ii), removing copyright protections for computer-generated works would fall into line with the human author requirements of the EU and USA. The UK IPO Consultation points out that this would not affect the protections of what they call ‘AI-assisted works’, where a work is ‘created’ by a human author and the AI is used as just a tool,98 but this presupposes a fully autonomous AI which creates with zero human involvement, which we do not yet have, as discussed above. All technology is just a human tool; the question is whether the human involvement in the tool use could constitute a creative choice.
The UK IPO Consultation considers it a ‘benefit’ that removing these rights would mean that users of computer-generated works would no longer have to seek licences or permission to use those works. It remains to be seen if the potential rightsholders who find themselves no longer protected by copyright would consider this a benefit.
Option (iii) proposed by the UK IPO Consultation is to replace the current regime with a new alternative. This new alternative is suggested as a more limited form of protection, if such protection incentivizes production and investment. The proposal suggests that the work would not need to be ‘original’, and it would only apply to the reproduction right in the work, and not to adaptations of it.99
The results of the UK IPO Consultation will be interesting, as they may provide a window into how we will address the AI authorship issue into the future.
On the one hand, it is tempting to say that the Irish proposal is the solution to the difficult question of AI and authorship, however, on deeper inspection, in order to qualify for copyright protection, the person who ‘made the necessary arrangements’ for the generation of the work, would likely have to display a ‘modicum of creativity’ in doing so. It is difficult to ignore the criticism which has been levelled at the Irish and UK approach to authorship of computer-generated works. As Hugenholtz and Quintais point out, today, to be original and therefore protectable under copyright law, a work must be the result of free and creative choices. This requirement, therefore, makes impossible the existence of work ‘in circumstances such that there is no human author of the work’.100
Conclusion
Central to the concept of AI is autonomy—the ability to think for oneself and to make independent choices. Given that technology is not yet at the point where AI has free will, it is simply the case that AI is unable to make free and independent choices. As we have seen, autonomy is also a central feature of copyright, due to the requirement for an author to make free and creative choices. If a work has been created, it didn’t simply pop into existence from the ether—there was a choice made to make it, and creative choices were required at some point in order to make that happen. Perhaps, in relation to copyright and AI, the approach we should take ought to be like the approach an administrator of an estate would take in an intestacy. If there is no obvious heir to inherit the estate, the administrator will continue to work backwards through the family tree until they find a living person capable of inheriting the estate. Perhaps, in relation to copyright law and AI, if a work is created, but not by a human, we should work back through the creative process until a human was involved, and analyse their actions to ascertain if free and creative choices were made by that person.
As Hugenholtz and Quintais note, what we can take from EU law is that the CJEU does not seem to require that the author’s creativity or personality (‘“personal stamp”’) be objectively discernible in the resulting expression (the output). What appears to be sufficient is that prospective authors exercise their ‘“free and creative choices”’ and thereby express their personality.101 However, they go on to suggest that this in and of itself may not be enough. In Painer, the Court spoke of choices that must be ‘“creative”’, and that ‘[b]y making those various choices, the author of a portrait photograph can stamp the work created with his “personal touch”’.102 However, as Hugenholtz and Quintais point out, this language suggests that exercising creative freedom in a non-creative way, e.g., by making only obvious choices, would not result in a protected work. On the other hand, as we have seen before, ‘the requirement of originality or creativity does not entail a test of artistic merit or aesthetic quality, or that the work be novel (new).’103
Essentially, if something can be fixed in a tangible form, and if it is original, and the product of creativity, whereby a human-made free and creative choices, then it is a work capable of being protected by copyright. In the context of AI-generated works, the question is, did a human make free and creative choices whereby the author’s creativity can be discernible in those works? In real and practical terms, what this means is that under EU law, whether an AI-generated work is capable of copyright protection will need to be decided in a case-by-case basis, with the level and type of human interaction with the process closely analysed. What remains to be seen is how far back in the AI generative process courts will be willing to go. If the courts are willing to go as far back as the selection of the training data point, then there is a very strong case to be made that curating and selecting what music to be used constitutes a creative choice, indicative of the AI trainer’s personality.
The question of which other stages of the AI generative process could constitute creativity depends on the type of machine learning which is used. If it is supervised machine learning, then the choices made in training the AI and guiding the AI could be creative. Even at the output stage, depending on what is output (i.e., whether it’s a whole audible piece of music or a notated lead sheet), may require creativity. The real issue will be whether a user who selects parameters such as genre, instrumentation, mood, tempo, etc. will have elicited enough creative independence to ‘stamp’ their personality into the resultant AI work.
Ginsburg and Budiardjo note that, while a designer might program randomness to vary a machine’s outputs and its processes, the resulting output, even if unique and completely unpredictable, is the direct result of the machine’s process, which, in turn, is inevitably the brainchild of some human developer or user.104 Perhaps copyright law must adapt, has it has done in the past, to advances in technology. It would not require an enormous shift in judicial interpretation to simply go back along the creative chain until a human making creative choices is found. There is nothing in copyright law which suggests that creativity is concurrent to output.
It may be the case that a hybrid model of EU originality and creativity requirements, with the Irish model of granting authorship to the person who made the necessary arrangements for the generation of a work, could be a way of ensuring that AI works are protected by copyright. If the person who made the necessary arrangements for the creation of the work did not make creative choices, and if they do not meet the criteria set out in CJEU caselaw, then the work would not qualify for copyright protection. However, if the work was generated by a machine, and the person who made the necessary arrangements for the generation of the work made some creative choices, then under the proposed model, work generated by a machine could be granted copyright protection.
The central consideration for whether Ireland’s legislation could be interpreted in line with EU authorship requirements, is whether the programming of the machine was creative. For example, on the one hand, the US Copyright Office will not copyright designs created by a weaving machine which is programmed to ‘create’ random designs, but on the other hand, as Ginsburg and Budiardjo’s point out, ‘[e]very action, step, or calculation made by a machine is the product of the precise articulation of commands by a human programmer or machine-operator (including programmed randomness)’.105
Perhaps what is required is not a new set of rights or laws in order to protect computer-generated works with copyright, but rather, a reassessment of what is creative. An engineer will be granted a patent if amongst other things, they show that they have made an ‘inventive step’. What is an ‘inventive step’, if not one of the primary products of human creativity? If the courts appreciate and acknowledge the creativity that goes into modern computer programming and machine learning, perhaps the courts would be more mindful to grant copyright protection to those who made the ‘necessary arrangements’ for the creation of a machine-generated work.
Footnotes
A Ramesh, M Pavlov, G Goh, S Gray, C Voss, A, Radford, M Chen, I Sutskever ‘Zero-Shot Text-to-Image Generation’ (2021) arXiv:2102.12092v2 [cs.CV] 26 Feb 2021.
Ramesh et al (supra at 4) note that their preliminary experiments for models up to 1.2 billion parameters were carried out on Conceptual Captions, a dataset of 3.3 million text-image pairs that was developed as an extension to MS-COCO, which was then scaled up to 12 billion parameters, by collecting 250 million text-image pairs from the internet.
J McCutcheon, ‘The vanishing author in computer-generated works: a critical analysis of recent Australian Case Law’ (2013) 36 Melbourne University Law Review 915–969, 929.
JC Ginsburg and LA Budiardjo, ‘Authors and machines’ (2019) 34 Berkeley Tech. LJ. 343, 452.
J McCutcheon, ‘The vanishing author in computer-generated works: a critical analysis of recent Australian Case Law’ (2013) 36 Melbourne University Law Review 915–969, 929–930.
A Ramalho, ‘Will robots rule the (artistic) world? A proposed model for the legal status of creations by artificial intelligence systems’ A Proposed Model for the Legal Status of Creations by Artificial Intelligence Systems (13 June 2017) (2017), 13.
Draft Opinion of the Committee on Legal Affairs for the Committee on the Internal Market and Consumer Protection and the Committee on Civil Liberties, Justice and Home Affairs on the proposal for a regulation of the European Parliament and of the Council laying down harmonized rules on artificial intelligence (Artificial Intelligence Act) and amending certain Union Legislative Acts (COM(2021)0206—C9-0146/2021–2021/0106(COD)), 25.
ibid.
A Guadamuz, ‘Do androids dream of electric copyright? Comparative analysis of originality in artificial intelligence generated works’ (2017) 2 I.P.Q. 169–186, 173.
ibid at 173.
PB Hugenholtz and JP Quintais, ‘Copyright and Artificial Creation: Does EU Copyright Law Protect AI-Assisted Output?’ (2021) 52 IIC-International Review of Intellectual Property and Competition Law 1190–1216.
A Bridy, ‘The Evolution of Authorship: Work Made by Code’ (2016) 39 Colum JL & Arts 395, note 1.
IFPI submission to WIPO on draft Issues Paper on Intellectual Property Policy and Artificial Intelligence (2020) Available at https://www.wipo.int/export/sites/www/about-ip/en/artificial_intelligence/call_for_comments/pdf/org_ifpi.pdf (accessed 5 May 2022).
CISAC Comments to WIPO Draft Issues Paper On Intellectual Property and Artificial Intelligence (2020) Available at https://www.wipo.int/export/sites/www/about-ip/en/artificial_intelligence/call_for_comments/pdf/org_cisac.pdf (accessed 5 May 2022).
See e.g., J Jumper, R Evans, A Pritzel, et al., ‘Highly accurate protein structure prediction with AlphaFold’ (2021) 596 Nature 583–589. https://doi.org/10.1038/s41586-021-03819-2.
A Guadamuz, ‘Do androids dream of electric copyright? Comparative analysis of originality in artificial intelligence generated works’ (2017) 2 I.P.Q. 169–186, at 173.
Judgment in Luksan v. van der Let, C-277/10, EU:C:2012:65.
P Goldstein and PB Hugenholtz, International Copyright: Principles, Law, and Practice (4th edn OUP Oxford 2019), 229.
F Vehar and T Gils, ‘I’m sorry AI, I’m afraid you can’t be an author (for now)’ (2020) 15 Journal of Intellectual Property Law & Practice 718–726, 719.
PB Hugenholtz and JP Quintais, ‘Copyright and Artificial Creation: Does EU Copyright Law Protect AI-Assisted Output?’ (2021) 52 IIC-International Review of Intellectual Property and Competition Law 1190–1216, 1195.
F Vehar and T Gils, ‘I’m sorry AI, I’m afraid you can’t be an author (for now)’ (2020) 15 Journal of Intellectual Property Law & Practice 718–726, 720.
Spanish Copyright Law—Title I Article 1, Revised Law on Intellectual Property, regularizing, clarifying and harmonizing the applicable statutory provisions (approved by Royal Legislative Decree 1/1996 of 12 April 1996).
Ibid Article 5(1).
Ibid Article 5(2).
German Copyright Law—Article 7, Urheberrechtsgesetz, UrhG.
Ibid Article 10(1).
Ibid Article 2(2).
F Vehar and T Gils, ‘I’m sorry AI, I’m afraid you can’t be an author (for now)’ (2020) 15 Journal of Intellectual Property Law & Practice 718–726, 720.
French Copyright Law—Article L111-2, Code de la propriété intellectuelle—No 92–597 of 1 July 1992.
Ibid Article L111-1.
Ibid Article L113-1.
Italian Copyright Law—Article 1, Diritto d’Autore—No. 633 of 22 April 1941 (as amended).
Ibid Article 7.
Ibid Article 8.
A stem is an audio file which breaks down a complete track into individual mixes, such as the drums, bass, vocals, etc.
Directive 2006/116/EC of the European Parliament and of the Council of 12 December 2006 on the term of protection of copyright and certain related rights.
European Commission, Directorate-General for Communications Networks, Content and Technology, C Hartmann, J Allan, P Hugenholtz, et al., ‘Trends and developments in artificial intelligence: challenges to the intellectual property rights framework’ (2020): final report, Publications Office, 68 and 69.
E Rosati, ‘The Monkey Selfie case and the concept of authorship: an EU perspective’ (2017) 12 Journal of Intellectual Property Law & Practice 973–977, 975.
Article 6 Directive 2006/116/EC of the European Parliament and of the Council of 12 December 2006 on the term of protection of copyright and certain related rights.
Article 6 Directive 2006/116/EC of the European Parliament and of the Council of 12 December 2006 on the term of protection of copyright and certain related rights.
E Rosati, ‘The Monkey Selfie case and the concept of authorship: an EU perspective’ (2017) 12 Journal of Intellectual Property Law & Practice 973–977, 975. (With the exclusion of cinematographic and audio-visual works, computer programs and databases).
European Commission, Directorate-General for Communications Networks, Content and Technology, C Hartmann, J Allan, P Hugenholtz, et al., ‘Trends and developments in artificial intelligence: challenges to the intellectual property rights framework’ (2020): final report, Publications Office, 69.
ibid.
Judgment in Infopaq International A/S v Danske Dagblades Forening, C-5/08, EU:C:2009:465, paras 23 and 24.
ibid at para 25.
E Rosati, ‘The Monkey Selfie case and the concept of authorship: an EU perspective’ (2017) 12 Journal of Intellectual Property Law & Practice 973–977, 976.
F Vehar and T Gils, ‘I’m sorry AI, I’m afraid you can’t be an author (for now)’ (2020) 15 Journal of Intellectual Property Law & Practice 718–726, 719.
Judgment in Infopaq International A/S v Danske Dagblades Forening, C-5/08, EU:C:2009:465, para 34.
Directive 2009/24/EC of the European Parliament and of the Council of 23 April 2009 on the legal protection of computer programs.
Directive 96/9/EC of the European Parliament and of the Council of 11 March 1996 on the legal protection of databases.
Directive 2006/116/EC of the European Parliament and of the Council of 12 December 2006 on the term of protection of copyright and certain related rights.
ibid at para 35.
Directive 2001/29/EC of the European Parliament and of the Council of 22 May 2001 on the harmonization of certain aspects of copyright and related rights in the information society.
ibid at para 36.
ibid at para 37.
ibid at para 39.
Judgment in Infopaq International A/S v Danske Dagblades Forening, C-5/08, EU:C:2009:465, paras 70 and 71.
European Commission, Directorate-General for Communications Networks, Content and Technology, C Hartmann, J Allan, P Hugenholtz, et al., ‘Trends and developments in artificial intelligence: challenges to the intellectual property rights framework’ (2020): final report, Publications Office, 69.
Judgment in Eva-Maria Painer v Standard VerlagsGmbH, Axel Springer AG, Süddeutsche Zeitung GmbH, Spiegel-Verlag Rudolf Augstein GmbH & Co KG, Case C-145/10, EU:C:2011:798.
ibid at para 87.
ibid at para 88.
ibid, at para 89.
ibid at para 92.
Opinion in Eva-Maria Painer v Standard VerlagsGmbH and Others, C-145/10, EU:C:2011:239, para 121.
PB Hugenholtz and JP Quintais, ‘Copyright and Artificial Creation: Does EU Copyright Law Protect AI-Assisted Output?’ (2021) 52 IIC-International Review of Intellectual Property and Competition Law 1190–1216, 1195.
Judgment in Funke Medien NRW GmbH v Bundesrepublik Deutschland, C-469/17, ECLI:EU:C:2019:623.
Ibid, paragraph 23.
ibid, paragraph 23.
Judgment in Luksan v van der Let, C-277/10, EU:C:2012:65. See analysis at paras 6 to 20.
P Goldstein and PB Hugenholtz, International Copyright: Principles, Law, and Practice (4th edn OUP Oxford 2019), 229.
‘Study on copyright and new technologies: Copyright data management and artificial intelligence’ European Commission, February 2022, 172.
Section 9(3) of the Copyright Designs and Patents Act, 1988 (as amended) defines an author ‘[i]n the case of a literary, dramatic, musical or artistic work which is computer-generated, the author shall be taken to be the person by whom the arrangements necessary for the creation of the work are undertaken’. New Zealand too recognizes that copyright works may be computer-generated. Section 5 (2) (a) of New Zealand’s 1994 Copyright Act defines the author of a computer-generated work as the ‘person’ who ‘makes the arrangements necessary for the creation of the work’.
Section 21(f) of the Copyright and Related Rights Act, 2000.
See e.g., BG Otero and JP Quintais, ‘Before the Singularity: Copyright and the Challenges of Artificial Intelligence,’ Kluwer Copyright Blog (blog), 10, 2018.
European Commission, Directorate-General for Communications Networks, Content and Technology, C Hartmann, J Allan, P Hugenholtz, et al., ‘Trends and developments in artificial intelligence: challenges to the intellectual property rights framework’ (2020): final report, Publications Office, 88.
F Vehar and T Gils, ‘I’m sorry AI, I’m afraid you can’t be an author (for now)’ (2020) 15 Journal of Intellectual Property Law & Practice 718–726, at 721.
ibid, at 721.
See e.g. A Ramesh, M Pavlov, G Goh, S Gray, C Voss, A, Radford, M Chen, I Sutskever ‘Zero-Shot Text-to-Image Generation’ (2021) arXiv:2102.12092v2 [cs.CV] 26 February 2021 and J Jumper, R Evans, A Pritzel, et al., ‘Highly accurate protein structure prediction with AlphaFold’ (2021) 596 Nature 583–589. https://doi.org/10.1038/s41586-021-03819-2.
S Ricketson, ‘People or Machines: The Berne Convention and the Changing Concept of Authorship’ (1991) 16 Colum.-VLA J. L. & Arts 1.
JC Ginsburg, ‘People Not Machines: Authorship and What It Means in the Berne Convention’ (2018) 49 IIC 131–135, 132.
‘Daddy’s Car’. Available at https://www.youtube.com/watch?v=LSHZ_b05W7o (accessed 13 July 2022)
F Pachet, P Roy, and B Carré, ‘Assisted music creation with Flow Machines: towards new categories of new’ (2021) Handbook of Artificial Intelligence for Music 485–520. Springer, Cham, para 6.
See sections 78 and 81 of the Copyright, Designs and Patents Act 1988.
European Commission, Directorate-General for Communications Networks, Content and Technology, C Hartmann, J Allan, P Hugenholtz, et al., ‘Trends and developments in artificial intelligence: challenges to the intellectual property rights framework’ (2020): final report, Publications Office, 88.
BE Boyden, ‘Emergent Works’ (2016) 39 Colum. J. L. & Arts 377, 380.
ME Kaminski, ‘Authorship, Disrupted: AI Authors in Copyright and First Amendment Law’ (2017) 51 UCD L Rev 589, 598. However, at page 132, Ricketson considered advancements in computer science, observing, ‘In light of the general propositions put forward concerning the concept of authorship under the Berne Convention, one might query whether there still remains much in this “soul” [of copyright] worth protecting. … [One might argue that we should] declare that copyright is not really concerned with the protection of the fruits of human authorship, but only concerned with the question of commercial value, however that value is embodied or arrived at … such a change … would make it possible to bring all new forms of technological creation within the broad umbrella of the Berne Convention … rather than [adopting] tailor-made … sui generis protection [for these new forms of creation]. … While this suggested approach appears attractive in a pragmatic sense, I do not believe that we should move in this way … there are powerful arguments, both in principle and necessity, in favour of retaining this human-centred notion of authorship and authors’ rights’.
Express Newspapers Plc v Liverpool Daily Post & Echo Plc [1985] 1 W.L.R. 1089 Ch D.
ibid at para 1093.
A. Adrian, ‘Law and Order in Virtual Worlds: Exploring Avatars, Their Ownership and Rights’ (2010) Hershey, PA: Information Science Reference.
‘Report of the Whitford Committee to Consider the Law on Copyright and Designs’ (1977), para.513.
Nova Productions v Mazooma Games [2007] EWCA Civ 219; [2007] Bus. L.R. 1032 at para 106.
JC Ginsburg and LA Budiardjo, ‘Authors and machines’ (2019) 34 Berkeley Tech. LJ. 343, 377.
Burrow-Giles Lithographic Co. v. Sarony 111 U.S. 53 (1884). Burrow-Giles can be compared with a contemporary UK decision in Temple Island Collections Ltd v New English Teas Ltd (No.2) [2012] EWPCC 1; [2012] E.C.D.R. 11 on the same issue, whereby the copyright of a photograph was in question. Birss QC held that in relation to a photograph, the composition is important, namely the angle of shot, the field of view and the bringing together of different elements at the right place and the right time are enough to prove skill and labour, and therefore should have copyright (at [68]—[70]). See also, University of London Press v University Tutorial Press Ltd [1916] 2 Ch. 601 Ch D; and Interlego AG v Tyco Industries Inc [1988] 3 All E.R. 949 PC.
JC Ginsburg and LA Budiardjo, ‘Authors and machines’ (2019) 34 Berkeley Tech. LJ. 343, 378.
‘Consultation stage impact assessment on Artificial Intelligence and Intellectual Property’, UK IPO, 29th of 18 October 2021.
‘Consultation stage impact assessment on Artificial Intelligence and Intellectual Property’, UK IPO, 29th of October 2021.
ibid 18 to 21.
Ibid 40.
ibid 20.
B Hugenholtz, and JP Quintais, ‘Copyright and Artificial Creation: Does EU Copyright Law Protect AI-Assisted Output?’ (2021) 52 IIC-International Review of Intellectual Property and Competition Law 1190–1216, para 4.3.
ibid, 1198.
Judgment in Eva-Maria Painer v Standard VerlagsGmbH, Axel Springer AG, Süddeutsche Zeitung GmbH, Spiegel-Verlag Rudolf Augstein GmbH & Co KG, Case C-145/10, EU:C:2011:798, para. 92.
B Hugenholtz, and JP Quintais, ‘Copyright and Artificial Creation: Does EU Copyright Law Protect AI-Assisted Output?’ (2021) 52 IIC-International Review of Intellectual Property and Competition Law 1190–1216, 1198.
JC Ginsburg and LA Budiardjo, ‘Authors and machines’ (2019) 34 Berkeley Tech. LJ. 343, 402.
ibid, 403 and 404.
