Evolution of Copyright in the Era of Artificial Intelligence: Analysis of Conflicts of Law and Judicial Precedents

Introduction

In the rapidly evolving landscape of intellectual property, the integration of Artificial Intelligence (AI) into the creative process has become an undeniable force, reshaping the very nature of original works and introducing a host of complex legal considerations (Greenstein, 2022). Legal systems are struggling to keep up with the transformative effects of AI on intellectual property¹. At the center of these effects lies AI’s unique ontology and more specifically its autonomy, which makes AI capable of producing creative and original work without any or at least any critical human intervention. Although this is a very recent legal issue, a number of court decisions from various legal systems worldwide have begun to be produced². This article attempts to explore the intersection between AI-generated works and intellectual property norms with reference to relevant recent court decisions, and with a view to examining how these may affect the direction of European Union (EU) law. The point of reference consists in the role of anthropocentrism as a prerequisite for IP protection.

The implications stemming from AI’s burgeoning role in the creation of copyrighted, patented, and owned works are potentially cataclysmic, because of the expanding intelligence of AI and its capacity to emulate human intelligence characteristics which at least resemble with creativity and originality. On such grounds it is reasonable to wonder about who should owe AI-generated inventions in general and in particular under EU law. The next part refers briefly to some elements of AI ontology which are critical in order to comprehend why anthropocentrism is challenged by AI in relation to IP law.

1. AI Ontology and the role of autonomy

AI is defined in the EU AI Act as “a system that is designed to operate with a certain level of autonomy and that, based on machine and/or human-provided data and inputs, infers how to achieve a given set of human-defined objectives using machine learning and/or logic- and knowledge based approaches, and produces system-generated outputs such as content (generative AI systems), predictions, recommendations or decisions, influencing the environments with which the AI system interacts”³. This definition is used in the present article in order to avoid extensive parathesis of AI definitions as well as because of its comprehensive meaning. The definition in “AI act” refers to a wide range of AI output, from decision-making to content. It is on the basis of this ontology of AI that the question about the regulation of AI-generated work emerges⁴.

The concept of AI, fundamentally is built around the quest to create machines which can emulate human intelligence or aspects of it; in other words, AI evolution refers to the quest for a new type of intelligent beings (Gerdes, 2018). The fundamental element of AI is its expanding, intellectual autonomy that provides it with the capacity to adapt to novel environments (Omohundro, 2008; Russell & Norvig, 2010). This is what makes AI so unique: the shift from automation, to autonomy. Autonomy means that AI does not constitute the mere outcome of predetermined, software programming but imitates and reproduces human learning procedure and aspects of human intelligence, through machine learning (McCarthy, 2008; Lake et al., 2016). Alan Turing’s approach was that computers could imitate children’s minds, methodology and evolution (Turing, 1950).

Machine learning means that AI is “taught” how to deliver certain outputs (Bostrom & Ćirkovic, 2008). The goal of machine learning is to achieve in terms of the intelligence of AI, natural-like, evolutionary patterns and therefore to come up with solutions to a wide range of not predetermined, problems, without necessarily having humans in the loop (Bostrom, 2014). Therefore, what AI is doing is to learn and decide, on the basis of the action with the highest expected utility, in light of the basic preferences and goals (Bostrom, 2014).

“Machine learning” takes place on the basis of bigdata harvesting and use so that algorithms can be trained. AI is trained on the basis of our collective, socially produced data and in this sense, AI autonomy is fundamentally – at least partially – a social outcome. It is this procedure that makes AI capable of developing and demonstrating characteristics such as logic⁵ – as a tool of analysis⁶ – creativity, problem solving, pattern recognition, classification, learning, induction, deduction, analogies building, optimization, surviving in an environment and language processing (Hutter, 2010; Hallevy, 2018), cognitive autonomy, intuition and strategic thinking (Yanisky-Ravid & Liu, 2018; Hallevy, 2018; Suchman & Weber, 2016)⁷. AI does not yet understand all these characteristics as a human does since it does not possess self-reflective intelligence but it already produces in many areas and cases, outputs that in humans prerequisite such intellectual capacities (Laton, 2016; Russell & Norvig, 2010)⁸.

Machine learning also explains why as AI evolves its ontology becomes probabilistic, nonlinear, complicated, opaque and therefore unpredictable, raising fundamental uncertainties that have been described as the “black-box effect”; we cannot be certain what the outcome of machine learning and AI actions will be⁹. We know the input and we see the output of machine learning but we are not certain of the in-between of the two (Karppi & Crawford, 2016; Van Asselt & Renn, 2011). The unique advantage of AI – its autonomy and therefore adaptability – comes with a significant risk as well – this is the “black box” (Castelvecchi, 2016).

Of course, there are serious disagreements about what and mainly when AI can achieve breakthroughs leading it to a level of general intelligence¹⁰. Nevertheless, even present, narrow AI autonomy produces transformative results – among other areas – in relation to original intellectual work, already significantly limiting or diminishing the human presence in the loop (Martinez, 2019).

The debate about whether AI can be creative and original or these are uniquely human characteristics is interdisciplinary and largely unanswered yet (Hashiguchi, 2017b; Hattenbach & Snyder, 2018)¹¹. For a certain part, AI – at least the one that we currently have – is solely guessing patterns and therefore it can never be creative and original. From another perspective this is a highly unfair approach, overlooking that even at a limited extent, AI is emulating human mind characteristics, including aspects of creativity. What partially bypasses this ontological debate but also answers it in the area of law is that regardless of whether AI can be considered as ontologically creative or not, the fact is that it produces work which if produced by human authors would be considered creative and therefore protected under copyright norms.

Therefore, algorithms produce work which when produced by humans is protected under intellectual property norms. How must law treat such work? Should it be protected by IP law for the benefit of natural or legal persons or should it be freely accessible? (Xu et al., 2018)¹². The next section briefly examines the foundations of IP law in general and EU law in particular. Then, the relevant judicial precedents are examined in order to assess the impact of AI on IP law in general and under EU law in particular.

2. The fundamentals of intellectual property norms and AI impact. The EU law perspective

In order to address the issue of AI and IP law, first the foundations of intellectual property law must be briefly examined. To begin with, intellectual property “…very broadly, means the legal rights which result from intellectual activity in the industrial, scientific, literary and artistic fields”¹³. Intellectual property law is very simple at its core: it transforms knowledge and its practical applications into economic value (Manderieux, 2010). It is supposed to be able to achieve a balance between competitive interests – private and public – and to regulate access to benefits (Pila & Torremans, 2019).

In the framework of intellectual property, scientific works belong to the copyright branch and inventions to industrial property¹⁴. As inventions are defined the new solutions to technical problems, whereas scientific discoveries consist of “the recognition of phenomena, properties or laws of the material universe not hitherto recognized and capable of verification”¹⁵. The fundamental element of intellectual property is the intersection of the creative, nonobvious, original idea or invention, with the practical application of industrial utility. The determination of the fulfillment of each specific criterion constitutes a legal challenge.

The theoretical foundations of intellectual property norms are the labor/desert and the utilitarian/incentive theory (Khoury, 2017). The former emphasizes on the reward of the work of the creator, whereas the second on the motivation to creators to further work on new ideas and new inventories (Fisher, 2001)¹⁶. Both of them are built on two fundamental assumptions: the first one is that there is a human author behind the protected work and the second one is that this human must be rewarded for her/his work.

The concept of intellectual property protection has been criticized on the basis of the lack of social utility of intellectual property norms as promoters of monopolies and therefore as obstacles to innovativeness (Hemel & Ouellette, 2013; Rai, 1999). Further, intellectual property rights do not emerge as other property rights out of scarcity but create scarcity eventually leading to wealth reducing, at least at a general social level (Krauss, 1989). A strong public domain is the “engine”, opening the public to new ideas and inventions, whereas intellectual property protection excludes or restricts access to the protected work, therefore limiting the free flow of ideas and applications (Cohen, 2006; Salzberger, 2006). The “ocean” is the public domain and intellectual property are the “islands”, which eventually “collapse” into the former (Khoury, 2017). Therefore, intellectual property rights under all legal systems must always balance with the wider public interest so that they are not abused as rights: public access must not be unfairly restricted in favor of a natural or legal person. This becomes even more critical when positioned in the framework of AI and the gradual expulsion of human from the loop.

Another common element among all legal systems is that they exclude mental activities from intellectual property protection. A variety of legal precedents have clarified this issue by introducing a crucial distinction between mental activities per se and mental activities with an industrial application. The former, mental activities without industrial application are not patent-eligible. The latter, mental activities with industrial applications may be patent-eligible on the basis of the assessment of the relationship between the mental activities and their industrial application.¹⁷

According to the European Patent Office, there are four basic requirements for intellectual property protection: “there must be an «invention», «susceptible of industrial application», which is «new» involving an «inventive step¹⁸. Under the EU legal system, it is the personality of the author that is protected by copyright laws¹⁹ (Kur et al., 2013). The author’s personality and the creativity that the latter demonstrates constitute a synthesis leading to the originality of work (Hugenholtz & Quintais, 2021). The level of creativity is assessed ad hoc and on the basis of general guidelines (van Gompel, 2014). What is undoubted is that without author’s personality there is no further assessment that needs to be done²⁰.

The fact that the protection of the personality of the author constitutes the foundation of EU copyright law is apparent in a variety of rules: a creation cannot be modified or distorted without the permission of the author, regardless of any potential transfer of the copyright²¹; it must be associated with the author’s name; the disclosure of the creation is prohibited until the author adheres to it the author retains the right to retract the creation²². The above elements essentially comprise the EU natural law theoretical approach (Holst, 2006; Adler, 2009).

Under EU law and in particular the influence of the European Court of Justice (ECJ) originality is linked with the “author’s own intellectual creation” and therefore implicitly with human authorship (van Eechoud, 2012). It was during a short period of time, between 2009 and 2012, that the ECJ elaborated further on the linkage of originality with authorship through five decisions²³. In these decisions it was clarified that originality is established in “author’s own intellectual creation”, which presupposes free and creative choices. This type of choices is absent when the relevant techniques, functions or rules make it imperative for the author to express in only one specific way, therefore leaving no space for free choice. Creativity and free choice are essentially qualitative and not quantitative characteristics. It is not the effort of work that is put in each creation that matters but the level of creativity which goes with freedom of choice²⁴. This is the case with both traditional works and the ones based on new technologies²⁵.

Until the emergence of AI, authorship was obviously human. This dimension was inferred by the fact that creativity, as well as freedom of choice, all presuppose the intellectual capacity, which was self-evidently the realm of human²⁶. Since intellectual property protects author’s personality, what personality is to be protected if there is no human author?²⁷ It is in this relationship that AI steps in, which explains why all these self-evident until recently facts are tested under the influence of AI.

The advent of technology, as already mentioned prompted fundamental reconsiderations of the traditional concept of authorship even prior to AI. From the 1988 Green Paper, to the Directive 91/250/EEC of 14 May 1991 on the legal protection of computer programs, the issue of human authorship gradually emerges as a matter of debate (Walter & von Lewinski, 2010). Until AI there was little doubt about the anthropocentrism of authorship. Computers are “automata”, not autonomous. AI however, through its variety of applications can produce the type of work that until the AI era could be only the output of human intelligence (Hugenholtz & Quintais, 2021). Even existing, narrow AI (ANI) can by now produce creations in practically all the areas of human creativity and intellectual activity (Senftleben & Buijtelaar, 2020; Gervais, 2019; Senftleben & Buijtelaar, 2020; Butler, 1982).

Given that human presence in the loop – or on the loop – decreases, the linear causality between the (distant) human “mind” behind an AI algorithm and the final invention or work sublimates at least up to a significant extent or even – by now – completely. How are we going to assess it so that we can reach a conclusion about whether copyright protection should be provided or not (Hashiguchi, 2017a)? When is it not fair anymore to reward a human for AI’s work (Spector, 2006; Jaszi, 1992; Grinmelmann, 2016)²⁸?

In this context, a number of court rulings internationally help us further elaborate our approach to the effect of AI on the requirement of an author who must be protected under IP law.

3. Judicial precedents assessing the impact of AI on copyright law

The relevant judicial precedents emerge both from EU member states’ and internationally. All of them are useful in order to clarify the intersection of AI and IP protection in general and under EU law.

A US Court, the District Court of Columbia held that an entirely AI-generated artwork, in the production of which there is not any human involvement, is ineligible for copyright protection because of the lack of human authorship. The case was built on the request of the owner of a computer system called “Creativity Machine” to register for copyright protection the visual art which was produced by the AI and then transfer to him the copyright because he was the owner of the system. The US copyright office declined the copyright protection because of the lack of human authorship. The plaintiff invoked the common law “work-for-hire” doctrine in his favor. The Court held that these arguments concern “…to whom a valid copyright should have been registered, and in so doing put the cart before the horse”. it further held that “Copyright is designed to adapt with the times. Underlying that adaptability, however, has been a consistent understanding that human creativity is the sine qua non at the core of copyrightability, even as that human creativity is channeled through new tools or into new media”²⁹.

In another case, the US Copyright Office concluded that AI-generated content must be disclaimed in the registration application to provide the Office with the information relevant to the preparation or identification of the work or to the existence, ownership or duration of the copyright, eventually refusing to register under copyright norms the relevant AI-generated work. In particular the Board found that “… the Work contains more than a de minimis amount of content generated by artificial intelligence (“AI”), and this content must therefore be disclaimed in an application for registration”³⁰. The work under assessment was upgraded by the human who claimed copyright protection but the Board found that since he refused to disclaim the material produced by AI and this material exceeded a de minimis amount of AI-generated content, copyright protection could not be provided. The Board in its decision reiterated the findings of the Thaler v. Perlmutter case in which it was held that “human authorship is a bedrock requirement of copyright”³¹. The Board also mentioned another famous case, “Urantia Found. v. Kristen Maaherra”, the judge of which had found that “some element of human creativity must have occurred in order for the Book to be copyrightable” and that “it is not creations of divine beings that the copyright laws were intended to protect”³².

The US copyright office has been consistent in that the fundamental necessity for copyright protection is human authorship. In this framework it issued public guidance according to which the fundamental question consists in ““whether the ‘work’ is basically one of human authorship, with the computer [or other device] merely being an assisting instrument, or whether the traditional elements of authorship in the work (literary, artistic, or musical expression or elements of selection, arrangement, etc.) were actually conceived and executed not by man but by a machine”³³.

The United States Patent and Trademark Office (USPTO) Inventorship Guidance for AI-Assisted Inventions, which came into effect by February 13, 2024, establish joint inventorship rules between human and AI in AI-assisted inventions. In cases where a human and generative AI are each instrumental in creating an invention they are determined as co-inventors. Given however the fact that AI cannot be protected as inventor the crucial issue becomes whether, for each patent application, “at least one natural person has made a “significant contribution” that satisfies the joint inventorship Pannu factors required of being an inventor. If not, then the invention cannot be patented because there is no inventor to list”³⁴.

In order therefore to have intellectual property protection in a joint human-AI scheme, the human co-inventor must be able to prove that simultaneously and crucially participated both in the conception and in the industrial application of the invention; not only in one of the two. The USPTO guidelines are particularly helpful in cases such as for example when the human simply owns the AI system, or provides to the latter the problems that it must solve. These are not cases of IP protection. On the contrary, the design, building and training of AI in order to solve a particular problem, as well as a critical participation of the human in the problem-solving procedure of AI may lead to recognition of inventorship³⁵. The afore-mentioned guidelines coincided with the judicial precedent of the Thaler v Vidal case, where the recognition of AI as inventor was declined³⁶.

Both US courts and authorities have contributed significantly in the relationship between AI and intellectual property: AI-oriented inventions are in principle excluded from IP protection. It is not the actual capacity of AI to invent things that is denied but its potential protection under IP law. In this sense, IP protection is safeguarded only after careful, case-by-case, quantitative and qualitative examination, assessing the role of the human factor and AI work³⁷.

The Internet Court of Beijing partially distanced itself from the afore-mentioned reasoning in a similar case³⁸. It also made a quantitative assessment between human work and AI but found that the work was eligible for copyright protection because it was original. Its originality stemmed “from the numerous positive and negative prompts inserted and the adjustments and amendments made by the human user to select the final image that matched his expectations”³⁹. The Court held that in spite of AI involvement, the AI-generated image “… reflected the plaintiff’s individual creativity and aesthetic choices made during the creation process”. The plaintiff’s creative intellectual input included designing, selecting prompts, setting parameters during the image’s creation, amending and adjusting the output image several times, until he reached a final image that matched his expectations. In light of the above, the Court found that “AI models lack legal personality and humans remain the creators of works generated using this technology”⁴⁰.

The Court in particular has reached a rather problematic conclusion in this regard. On the one hand, it maintains the assessment of the degree of impact of human intelligence on the produced work on a case-by-case basis, but on the other hand it lowered the threshold of necessary human participation considerably. Therefore, while it does not deny the prerequisite of human authorship in principle, it practically relativizes its significance. Further, the fact that AI does not have legal personality should not imply that copyright passes to its human owner automatically. Why the lowering of the threshold of necessary human involvement is problematic becomes profound on the basis of the founding on intellectual property protection on human personality, as among various cases, the recent one “Lithoss Nv V Vimar S.P.A. And Vecolux BV” reiterated⁴¹. The reasoning of the Beijing Court seems to falsely confuse automation with autonomy. It also seems to be partially “manipulating” the true ontology of AI in furtherance of not jeopardizing the continuation of intellectual property protection and profit.

Chinese courts nevertheless have also come to decisions which are better aligned with the goals of intellectual property norms and their emergence from human authorship. The Beijing Internet Court held in the “Beijing Film Law Firm v Beijing Baidu Netcom Science & Technology Co Ltd (Film)” case, that creation by natural persons was a prerequisite for protection under the Chinese copyright law. In that case the work produced by AI was not to be protected by copyright law, regardless of its originality (Yong Wan & Hongxuyang Lu, 2021).

Not all states’ copyright laws and judicial precedents follow the same logic with the US and the EU or at least not always: the United Kingdom, South Africa, New Zealand, Ireland, and India, all have recognized copyright protection for computer-generated work, even absent any human intervention. According to the High Court of England and Wales (High Court of Justice), composite screen frames generated by a computer program of a coin operated video game are computer-generated works because the software built up composite images by overlaying the digital image of a pool table with images of the balls and cue.⁴² “In 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”⁴³. While UK courts decrease up to some extent the level of copyright protection in the afore-mentioned cases and the subsequent profitability from the relevant copyright protection, still in that instance had reached a legally and ethically unjustifiable norm of securing profits and ownership for humans who have shown no real authorship. Such approaches constitute an abuse both of the intellectual property norms and rights, as well as of the employer-employee relationship, given as previously mentioned that machines are not employees but owned technology, part of capital (Hristov, 2017).

Nevertheless, in the “Comptroller-General of Patents, Designs and Trade Marks v Emotional Perception AI Ltd” case, the English Court of Appeal overturned the previous argumentation and found that an Artificial Neural Network (ANN) was not patentable. The Court dealt with three questions: “Is an ANN a “computer”? If it is a computer, is an ANN a program for a computer “as such”, within the meaning of section 1(2) of the Patents Act 1977 (meaning that it would be excluded from patentability unless the next bullet applies). If the ANN is a program for a computer, did the ANN nevertheless fall outside the computer program exclusion (and so be patentable) because it had a “technical contribution” outside itself?”⁴⁴ The Court reached the conclusion that ANN is a computer, functioning as a computer program and in this sense that its recommendations are not technical but a matter of aesthetics. In this sense it “corrected” the precedent of the High Court and ascertained that there can be no IP protection for AI generated work.

The Federal Court of Australia, as well, in the Thaler case of 2022 had concluded that AI cannot be recognized as an inventor⁴⁵. The Court had found that only natural persons can be labelled as “inventors” and that it is necessary to have a legal relationship between the human inventor”… and the person first entitled to the grant, which is a legal impossibility in… case where the purported inventor has no legal identity and therefore cannot give effect to an assignment”⁴⁶.

A crucial decision albeit with a problematic syllogism was issued by the German Federal Court of Justice (Bundesgerichtshof – BGH), in the “DABUS” case, according to which only natural persons can be named as inventors under IP protection law. The German Court reached three landmark points in its judgment: first that AI can not be the inventor; second that behind every AI-generated invention there is a certain level of human contribution, which even if not inventive or substantial can lead to the designation of a human as the inventor provided that (s)he is the one with the decisive influence; and third that patent applications must not include contradictory statements. It must be declared that either the human or AI conceived the invention⁴⁷. The decision is crucial but at the same time characteristic of the perplexity of certain part of the judiciary. In fact, it is crucial especially because of this last reason.

The court confused the recognition of AI under intellectual property law as inventor with the actual capacity to conceive inventions. In this sense it confused law with ontology. The true ability of AI to conceive inventions is almost undoubted. This decision ignores this real condition. Instead, the court’s attempt in this decision appears to be to ignore AI’s ability in question, to banish AI from the true ability of inventing, and then, on the basis of the above false conception, to attribute invention to a human, even if his/her actual participation is entirely secondary or even insignificant. The Court limits itself only to the formality of whether only human is referred to as inventor or not. This is an attempt by the court to cling to the traditional approach to IP law, without taking into account the transformations taking place, resulting in an abuse of rights and unfair restriction of public access to the benefits of AI. The problem with the Court decision is not that it does not recognize AI as inventor under IP law but that it insists to do so with a human, regardless of the latter’s actual role in the invention. The result of this approach is to limit public access to an invention when there is really no human personality that needs to be protected. In this sense, it conflicts with EU law’s natural law-oriented approach to the need to protect the personality of the person who actually invents something original thanks to his creative conception.

Czech Courts produced a landmark decision, consistent with the requirement of human authorship. The case was brought in front of the Municipal Court in Prague and the plaintiff had used DALL-E, an AI program, in order to generate an image according to his request: “create a visual representation of two parties signing a business contract in a formal setting, such as a conference room or a law firm office in Prague. Just show your hands“⁴⁸. The image that was used by the plaintiff on his website, was copied and posted without his authorization by the defendant. The argument of the plaintiff did not doubt that it was AI that created the image but claimed that because of his assignment to the AI he should be copyright protected as the author. The Court rejected his argument by reaching the decision that first AI cannot be recognized as the author and secondly there was no unique creativity in his action. The Prague Court very adamantly held that the AI-generated work did not constitute a «work» because it was not the unique result of the creative conception of a natural person. According to the court: «Copyright is an absolute right belonging to an individual. If the image in question was not created personally by the applicant, but by an artificial intelligence, it cannot, by definition, be a copyrighted work”⁴⁹.

All the above jurisprudence, despite its internal contradictions, is particularly illuminating for the effect of AI-oriented or generated inventions on intellectual property norms internationally and in particular under EU Law. The first and fundamental point of reference is that copyright protection is exclusively safeguarded for human authors. It is a timeless and international common foundation of IP law. Characteristic in these regards – before the AI era – is the precedent in the Burrow-Giles Lithographic Co. v. Sarony case, where the Court held that the crucial factor for copyright protection is the originality of the ideas of the author. The definition of the term “author” is according to the court anthropocentric through the characterization of copyright as “the exclusive right of a man to the production of his own genius or intellect”⁵⁰. This reasoning has been followed both in the era of automation, as well as in the era of autonomy, i.e. of AI.

The central question is if a natural person can be protected as the author of the work that has been generated by AI. Different approaches can be found here. The common thread between case-law from different legal systems is that copyright is not automatically recognized in favor of the natural person who owns the AI system. In some legal systems, mainly that of the USA and EU, there is the requirement of a substantial, significant contribution of the natural person to the invention, both in terms of conception and industrial application, in order to be recognized as a co-inventor. In other legal systems, as we observed, it is easier to deliver intellectual property protection even with a minimal participation of the natural person. In any case, however, the weighting of the degree of human participation, whether a higher or lower level of contribution is required, is also a horizontal among different legal systems point of reference.

In order to assess the contribution of these judicial precedents in the clarification of EU law we must keep in mind that copyright protection under EU law is founded on natural law theories. Intellectual property protection exists not for speculative reasons but as a material reward of the personality of the human author and more specifically of her/his creativity. Absent this element, there is no moral and legal foundation for copyright protection in general and especially under EU law (Sobel, 2017). In principle, there is no sense of fairness in protecting AI-oriented creations under intellectual property norms. The attribution of such rights to a natural person for work that does not include her/ his own creativity constitutes an abuse of intellectual property norms since a human or a legal entity will be profiting by restricting the wider public’s access to work that was not of any human author.

In order to understand even better how unbalanced and disproportionate the attribution of such rights is, the role of machine learning must be also taken into account, which is making AI at a significant extent a social project, given that machine learning is conducted on the basis of collective, big data, produced by us all.

In addition, intellectual property norms by definition are not designed in order to create a framework of “the goose that lays the golden egg”, or to disproportionately restrict public access and interest. Intellectual property is designed to protect each specific human author’s creativity leading to original work with industrial application as an “island” in the “ocean” of ecumenically accessible knowledge and applications.

The definition of creativity under EU law also deserves attention in conjunction with the afore-mentioned judicial precedents as well. The CJEU talks about “…creative abilities [of the author] in the production of the work by making free and creative choices”⁵¹. The word “free” implies “autonomy”. From such a perspective, AI already can at least up to some extent make choices which if done by humans would be considered as free choices and therefore be creative. Once this is the case, there is no human creativity to be rewarded (Ginsburg & Budiardjo, 2019). There is only AI’s creativity which is excluded from IP protection however.

An attempt to “come around” the lack of human authorship is the one arguing in favor of re-defining the concept of employer and employee (Hristov, 2017). Under this approach, the relationship between the human owner and the AI is considered as an adjusted relationship between employer and employee. The problem persists however: there is no human personality being the main – even more the sole – author of the copyrighted work and therefore nothing to be protected under IP protection norms. Copyright protection regardless of the lack of human author would simply constitute an abuse of the right. It is very simple: the foundations of copyright protection for AI-generated work are not there since no human author’s personality can be found. Copyright constitutes an anthropocentric concept (Zurth, 2021). In a characteristic case brought in front of a US Court about copyright from a photo clicked by a monkey it was very clearly held that only human authorship can be protected under copyright law: “[W]e conclude that this monkey – and all animals, since they are not human – lacks statutory standing under the Copyright Act”⁵².

Further, “baptizing” machines as employees is simply arbitrary. All types of machines, until at least they reach the level of general or super intelligence and the equivalent, fully autonomous legal personality is considered as part of capital. The ECJ has also answered to the afore-mentioned approach: copyright protection’s prerequisite is the expression of author’s “creative abilities in the production of the work by making free and creative choices”⁵³. It is not the owner who makes free and creative choices but the one conceiving the idea and producing the original work.

After all, under EU law and according to the European Court of Justice precedents, there is no doubt about the linkage between author’s personality and intellectual property; any work is original once it is the “author’s own intellectual creation”⁵⁴. Even skill and work are secondary to the “personal stamp” element. They cannot justify protection by themselves but constitute elements in the wider scheme of intellectual property protection⁵⁵. The potential protection of AI-oriented work under copyright law constitutes a violation of the “alterum non laedere” principle: public access to AI benefits will be restricted without any fair and legitimate basis for this to happen. Such an approach is both disproportionate and unreasonable (Sganga & Scalzini, 2017; Mizaras, 2012). For intellectual property norms to apply they must be relevant with each case, proportional and fair.⁵⁶ This is obviously not the case when one of the fundamental pillars of intellectual property – human author – are absent. All of the above lead us to the same direction: copyright law is anthropocentric (Ginsburg, 2018).

The ineligibility of AI – oriented work for copyright protection however does not completely clarify the issue of the required threshold for human contribution in order to talk about human author or co-inventor. A more difficult question in other words, is whether we can have some qualitative and quantitative criterion about when AI’s intervention is so catalytic that human can no more be considered as the author of the work under potential copyright protection.

The threshold of human contribution is as shown above, debated. As already mentioned, EU law in this area is natural law-oriented. Therefore, the EU legal system must be interpreted as leaning to impose the highest possible threshold of human contribution in order to have copyright protection. Such an approach aligns with EU Courts’ decision – and the ones of the US. The human author’s personality must be directly linked with the invention either in the sense of being the sole inventor or as co-inventor. This means that for example it is not enough to simply present the question to AI but that either AI does some secondary work or solei or at least that the human has re-arranged and combined AI’s outputs in a creative way.

The concept of abuse of copyright law can further clarify the issue at hand. While the issue of the potential abuse of copyright law rights because of AI-oriented inventions has not been fully developed, still some guidelines can be inferred from EU courts’ past decisions in relation to human author and his/her behavior as rightsholder, in cases such as “Deutsche Grammophon”, “Coditel” I “Coditel II” and “Metronome Musik”⁵⁷. The fundamental notion in all of the afore-mentioned cases is that of reasonableness, proportionality and appropriateness in ‘‘the protection of the moral and economic rights” of the author⁵⁸. When there is a human author reasonableness, proportionality and appropriateness take the form of reasonable renumeration from the commercial use of the creation. When there is no human author, the same concepts should take the form of non- attribution of intellectual property norms due to the ellipsis of the primary prerequisite of copyright protection. When there is some human interference, this should be proven to be creative outside the output of AI. According to the CJEU copyright laws require the “…“indispensable” intervention by the operator (without this intervention, the customers would not be able to enjoy the work)” (Xalabarder, 2016). In AI- oriented work the question must be: is the human contribution in the invention indispensable for its materialization?

On such grounds, a normative framework which will provide us with guidance on a step-by-step basis is required. Just as the EU AI Act was adopted, a law about AI and intellectual property can be designed and put in practice as well. Ahead of such a potential legal development we need to envisage what could be the case for such a development.

An initial prerequisite could consist in the obligation to maintain accessible by the relevant patent authorities “log files” showing every step until the final work is produced and each “actor’s” – both human and AI – participation in the final outcome. This is a “sine qua non” so that we can then assess qualitatively and quantitatively each actor’s contribution. The log files should be submitted to the relevant authorities and in order for a human to be eligible for copyright protection these files should be accessible for assessment.

On the basis of this first step, the second one could be the reconstruction of the creative process. Some crucial thresholds of creativity must be determined: the selection of the area of interest, meaning in which scientific and industrial area an effort to become creative will take place; the conception of the original idea and of its later versions; the data on the basis of which the training will take place; the repetitive work until the original idea or its versions have been finalized. It is obvious that not all of these steps share the same qualitative value in the conception and the produce of the final creative work.

Through the log files the relevant authorities will be able to determine who is the subject of the causal link between the conception of the idea and the industrial application. In order for a work to be copyright-eligible, it should be proven that it is the human intelligence the one behind both the conception of the idea and the work that is needed in order for this idea to become industrial application. If the human fails to prove both the conception of the idea and its transformation to industrial application, then there can be no copyright protection. Secondary contributions by the human, such as improvements in the final work or only partial revisions of it will not be enough to justify copyright protection. Obviously, the relevant administrative and judicial authorities will have to make assessments.

There may be cases of course where regardless of the provision of whatever log files it will be impossible to determine whether it is the human or the artificial intelligence the subject of the causal link, due to a constant, back and forth interaction between the two. In such a case, of close and equally creative collaboration between human and artificial intelligence, again it would be unfair for the human to gain profit since (s)he is not the sole creator. Even if the time frame of copyright protection is reduced, for this period, the human will benefit from something that is not only human work. Therefore, the request must be for solely human authorship and for the burden of proof on the human.

Conclusions

EU law has been facing increasing challenges in terms of dealing with the impact of emerging technologies and especially AI on intellectual property norms (Rosati, 2014). In fact, it is not only the EU legal system that faces such difficulties but legal systems all over the world. AI is raising new challenges in front both of lawmakers and courts – which especially under EU law have played significant role in shaping the intellectual property norms (Favale et al., 2016). It cannot be questioned that AI transforms the intellectual property landscape at an unprecedented degree (Cabay & Lambrecht, 2015).

Intellectual property norms are not absolute. They must be balanced with public interest and competitive rights. Several EU member-states’ national legislations contain such provisions, as well as the EU law. While the exact extent of intellectual property protection both at the constitutional level as well as at the level of ordinary laws differs among member-states, the need for balanced approach is not questioned⁵⁹. The CJEU has adopted this position as well. In the “Scarlet Extended” and in “NetLog” cases it held that while intellectual property rights are protected, “there is (…) nothing whatsoever in the wording of that provision or in the Court’s case-law to suggest that that right is inviolable and must for that reason be absolutely protected”⁶⁰. Intellectual property protection must be fair towards the author – rewarding author’s personality – and also proportional in relation to public interest which needs the widest possible access to knowledge, creations and industrial applications.

The proposal of this article is that in AI-oriented work, AI can be ontologically creative in spite of the fact that its inventions cannot be protected under copyright law. AI is not mere automata but a unique, distinct actor with capacities of autonomous creation. The prohibition of public access to its work therefore, is unfair, disproportionate and abusive under intellectual property norms, both under EU law and international copyright law. What is needed is in fact a new set of norms and regulations under EU law, a type of AI autonomy standards and metrics that can guide us in terms of when an AI entity is so autonomous that its outputs must be freely accessible by us all. The principle must be ecumenical access to AI-generated work. This must be the new guiding principle in the emerging era of AI-generated work.