11.1 Introduction
This chapter reflects on the interaction between AI and Intellectual Property (IP) law. IP rights are exclusive rights vested in intangible assets that grant their owner a temporary monopoly as to the use thereof in a given territory. IP rights may be divided into industrial property and literary and artistic property. Industrial property rights protect creations that play a largely economic role and primarily include patents, trademarks, and design rights. The concept of literary and artistic property rights refers to copyright and related rights. Copyright offers the author(s) protection for literary and artistic works, while the three main related rights are granted to performing artists, producers, and broadcasting organizations.
The interface of AI and IP law has been the subject of much research already.Footnote 1 This chapter analyzes some of the relevant legal issues from a primarily civil law perspective, with a focus on the European Union (EU), and with the caveat that its limited length leaves little leeway for the nuance that this intricate, multifaceted topic demands. Section 11.2 treats the avenues open to innovators who seek to protect AI technology. Section 11.3 examines whether AI systems qualify as an author or inventor and who “owns” AI-powered content. Section 11.4 briefly notes the issues surrounding IP infringement by AI systems, the potential impact of AI on certain key concepts of IP law and the growing use of AI in IP practice.
11.2 Protection of AI Technology
Companies may protect innovation relating to AI technology through patent law and/or copyright law. Both avenues are treated in turn below.
11.2.1 Protection under Patent Law
Patent law seeks to reward investment into research and development in order to spur future innovation. It does so by providing patentees with a temporary right to exclude others from using a certain “invention,” a technological improvement that takes the form of a product or a process (or both). This monopoly right is limited to 20 years following the patent application, subject to payment of the applicable annual fees.Footnote 2 It is also limited in scope: while patentees can bring both direct and indirect infringements of their patent(s) to an end, they must accept certain exceptions as a defense to their claims, including use for experimental purposes and noncommercial use.Footnote 3 In order to be eligible for a patent, the invention must satisfy a number of conditions.
First, certain exclusions apply. The list of excluded subject matter under the European Patent Convention (EPC)Footnote 4 includes ideas that are deemed too abstract, such as computer programs as such, methods for performing mental acts and mathematical methods.Footnote 5 Pure abstract algorithms, which are essential to AI systems, qualify as a mathematical method, and are thus ineligible for patent protection as such.Footnote 6 However, this does not exclude patent protection for computer-implemented inventions such as technology related to AI algorithms, especially given the lenient interpretation of the “as such” proviso in practice. If the invention has a technical effect beyond its implementation on a computer – a connection to a material object in the “real” world – patentability may yet arise.Footnote 7 This will for example be the case for a neural network used “in a heart monitoring apparatus for the purpose of identifying irregular heartbeats,” as well as – in certain circumstances – methods for training AI systems.Footnote 8
Further, a patentable invention must satisfy a number of substantive conditions: it must be novel and inventive as well as industrially applicable.Footnote 9 The novelty requirement implies that the invention may not have been made available to the public at the date of filing of the patent application, indicated as the “state of the art.”Footnote 10 The condition of inventive step requires the invention to not have been obvious to a theoretical person skilled in the art (PSA) on the basis of this state of the art.Footnote 11 Finally, the invention must be susceptible to use in an industrial context.Footnote 12 Both the novelty and industrial applicability requirements do not appear to pose any challenges specific to AI-related innovation.Footnote 13 However, the inventiveness analysis only takes account of the patent claim features that contribute to the “technical character” of the invention, to the solution of a technical problem. Conversely, nontechnical features (such as the abstract algorithm) are removed from the equation.Footnote 14
The “patent bargain” between patentee and issuing government may lead to another obstacle. This implies that a prospective patentee must disclose their invention in a way that is sufficiently clear and complete for it to be carried out by a PSA, in return for patent protection.Footnote 15 This requirement of disclosure may be at odds with the apparent “black box” nature of many forms of AI technology, particularly in a deep learning context. This refers to a situation where we know which data were provided to the system (input A) and which result is reached (output B), but where it is unclear what exactly makes the AI system go from A to B.Footnote 16 Arguably, certain AI-related inventions cannot be explained in a sufficiently clear and complete manner, excluding the procurement of a patent therefor. However, experts will generally be able to disclose the AI system’s structure, the applicable parameters and the basic principles to which it adheres.Footnote 17 It is plausible that patent offices will deem this to be sufficient. The risk of being excluded from patent protection constitutes an additional incentive to invest in so-called “explainable” and transparent AI.Footnote 18 The transparency requirements established by the EU AI Act also play a role in this context.Footnote 19 Simultaneously, an overly strict assessment of the requirement of disclosure may push innovators toward trade secrets as an alternative way to protect AI-related innovation.Footnote 20
It is often difficult to predict the outcome of the patenting process of AI-related innovation. This uncertainty does not seem to deter prospective patentees, as evidenced by the rising number of AI-related patent applications.Footnote 21 Since the 1950s, over 300,000 AI-related patent applications have been filed worldwide, with a sharp increase in the past decade: in 2019, it was already noted that more than half of these applications had been published since 2013.Footnote 22 It is to be expected that more recent numbers will confirm this evolving trend.
11.2.2 Protection under Copyright Law
AI-related innovation may also enjoy copyright protection. Copyright protection is generated automatically upon the creation of a literary and artistic work that constitutes a concrete and original expression by the author(s).Footnote 23 It offers exclusive exploitation rights as to protected works, such as the right of reproduction and the right of communication to the public (subject to a number of exceptions), as well as certain moral rights.Footnote 24 Copyright protection lasts until a minimum period of 50 years has passed following the death of the longest living author, a period that has been extended to 70 years in, for example, the EU Member States.Footnote 25
The validity conditions for copyright are the requirement of concrete form and the requirement of originality. First, copyright protection is not available to mere abstract ideas and principles; these must be expressed in a concrete way.Footnote 26 Second, the condition of originality implies that the work must be an intellectual creation of the author(s), reflecting their personality and expressing free and creative choices.Footnote 27 Applied to AI-related works in particular, the functional algorithm in its purest sense does not satisfy the first condition and is therefore not susceptible to copyright protection.Footnote 28 However, the object and source code of the computer program expressing this idea are sufficiently concrete, allowing for copyright protection once the condition of originality is fulfilled.Footnote 29 Given the low threshold set for originality in practice, software that implements AI technology is likely to receive automatic protection as a computer program under copyright law upon its creation.Footnote 30
11.3 Protection of AI-Assisted and AI-Generated Output
This section analyzes whether AI systems could – and, if not, should – claim authorship and/or inventorship in their output.Footnote 31 It then focuses on IP ownership as to such output.
11.3.1 AI Authorship
Can AI systems ever claim authorship? To answer this question, we must first ascertain whether “creative” machines already exist. Second, we discuss whether an AI system can be considered an author and, if not, whether it should be.
Certain AI systems available today can be used as a tool to create works that would satisfy the conditions for copyright protection if they had been solely created by humans. Many examples can be found in the music sector.Footnote 32 You may be reading this chapter with AI-generated music playing, such as piano music by Google’s “DeepMind” AI,Footnote 33 an album released by the “Auxuman”Footnote 34 algorithm, a soundscape created by the “Endel”Footnote 35 app or one of the unfinished symphonies of Franz Schubert or Ludwig van Beethoven as completed with the aid of an AI system.Footnote 36 If you would rather create music yourself, Sony’s “Flow Machines” project may offer assistance by augmenting your creativity through its AI algorithm.Footnote 37 If you are bored with this text, which was written (solely) by a human author, you may instead start a conversation with “ChatGPT 4,”Footnote 38 read a novelFootnote 39 drafted by an AI algorithm or translate it using “DeepL.”Footnote 40 AI-generated artwork is also available.Footnote 41 Most famously, Rembrandt van Rijn’s paintings were fed to an AI algorithm that went on to create a 3D-printed painting in Rembrandt’s style in 2016.Footnote 42 Since then, the use of AI in artwork has skyrocketed, with AI-powered image-generating applications such as “DALL-E 3”Footnote 43 and “Midjourney”Footnote 44 gaining exponential popularity.Footnote 45
In most cases, there is still some human intervention, be it by a programmer, a person training the AI system through data input or somebody who modifies and/or selects output deemed “worthy” to disclose.Footnote 46 If such human(s) were to have created the work(s) without the intervention of an AI system, copyright protection would likely be available.
Copyright law requires the work at issue to show authorship; the personal stamp of the author. The author is considered to be a physical person, especially in the civil law tradition, where copyright protection is viewed as a natural right, granted to the author to protect emanations of their personality.Footnote 47 Creativity is viewed as a quintessentially human faculty, whereby a sentient being expresses their personality by making free, deliberate choices.Footnote 48 This tenet pervades all aspects of copyright law. First, copyright laws grant initial ownership of copyright in a certain work to its author.Footnote 49 Further, the term of protection is calculated from the author’s death. Also, certain provisions expressly seek to protect the author, such as those included in copyright contract law as well as the resale right applicable to original works of art. Moreover, particular copyright exceptions only apply if the author is acknowledged and/or if an equitable remuneration is paid to the author, such as the exception for private copies. The focus on the human author also explains the importance of the author’s moral rights to disclosure, integrity, and attribution.Footnote 50 Such a system leaves no room for the authorship of a nonhuman entity.Footnote 51 If there is insufficient human input in the form of free and creative choices on the part of an author, if the AI crosses a certain threshold of autonomy, copyright protection is unavailable.Footnote 52 This anthropocentric view is unsurprising, since IP laws were largely drafted at a time when the concept of nonhuman “creators” belonged squarely in the realm of fiction.
However, the core of the issue is whether the abstract idea of originality should be held to include the creating behavior of an AI system. Account must hereby be taken of the broad range of potential AI activity and the ensuing distinction between AI-assisted and truly AI-generated content. At the one end of the spectrum, we may find AI systems that function as a tool to assist and/or enhance human creativity, where the AI itself acts as a mere executer.Footnote 53 We can compare this to the quill used by William Shakespeare.Footnote 54 Further down the line, there are many forms of AI-exhibited creativity that still result from creative choices made by a human, where the output flows directly from previously set parameters.Footnote 55 Such AI activity may still be viewed as pure execution. In such cases, copyright should be reserved to the human actor behind the machine.
At the far end of the spectrum, we could find a hypothetical, more autonomous, “creative” AI, having independently created a work that exhibits the requisite creativity, which experts and nonexperts alike cannot distinguish from a work generated by a human. Even in such a case, it may be argued that there is no real act of “conception” in the AI system, given that every piece of AI-generated output is the result of prior human input.Footnote 56 Arguably, precisely this act, the process of creation, is the essence of creativity. As long as the human thought process cannot be formulated as an algorithm that may be implemented by a computer, this process will remain human, thus excluding AI authorship. However, the “prior input” argument also applies mutatis mutandis to humans, who create literary and artistic works while “standing on the shoulders of giants.”Footnote 57 This could render the “act of conception” argument against AI authorship moot, as could choosing the end result and thus the originality of the output as a (functionalist) focal point instead of the creative process.Footnote 58 Additionally, it is argued that granting AI systems authorship may stimulate further creative efforts on the part of AI systems. This appears to be in line with the economic, utilitarian rationale of copyright.Footnote 59 However, copyright seeks to incentivize human creators, not AI systems.Footnote 60 Moreover, it is difficult to see how AI systems may respond to incentives in the absence of human consciousness.Footnote 61 Without convincing economic evidence, caution is advised against tearing down one of the fundamental principles of copyright law. The mere fact that we can create certain incentives does not in itself imply that we should. Further, if we were to allow AI authorship, we must be prepared for an upsurge in algorithmic creations, as well as the effects on human artistic freedom that this would entail.Footnote 62
The risk of extending authorship to AI systems could be mitigated by instead establishing a related or sui generis right to AI-generated works and provide a limited degree of exclusivity in order to protect investments and incentivize research in this area. Such a right could be modelled in a similar way to the database right established by the EU in 1996.Footnote 63 This requires a substantial investment for protection to be available.Footnote 64
11.3.2 AI Inventorship
We now turn to AI inventorship. By analogy to the previous section, the first question is whether “inventive” machines already exist. Such systems are much scarcer than AI systems engaged in creative endeavors.Footnote 65 However, progress on this front is undeniable.Footnote 66 The AI sector’s primary allegedly inventive champion is “DABUS,”Footnote 67 labelled the “Creativity Machine” by its inventor, physicist Dr Stephen Thaler.Footnote 68 DABUS is a neural network-based system meant to generate “useful information” autonomously, thereby “simulating human creativity.”Footnote 69 In 2018, a number of patent applications were filed for two of DABUS’ inventions.Footnote 70 The prosecution files indicate DABUS as the inventor and clarify that Dr Thaler obtained the right to the inventions as its successor in title.Footnote 71 These patent applications offer a test case for the topic of AI inventorship.
Patent law requires inventors to be human. While relevant legislative provisions do not contain any explicit requirement in this sense, the inventor’s need for physical personhood is implied in the law.Footnote 72 While the focus on the human inventor is much less pronounced than it is on the human author, a number of provisions would make no sense if we were to accept AI inventorship. First, many patent laws stipulate that the “inventor” is the first owner of an invention, except in an employment context, where the employer is deemed to be the first owner under the laws of some countries.Footnote 73 Since AI systems do not have legal personality (as of yet), they cannot have ownership rights, nor can they be an employee as such.Footnote 74 Given that those are the only two available options, AI systems cannot be considered “inventors” as the law currently stands, as confirmed in the DABUS case, not only by the Boards of Appeal of the European Patent Office in the DABUS case, but also by the UK Supreme Court and the German Federal Supreme Court.Footnote 75 Another argument against AI inventorship may be drawn from the inventor’s right of attribution. Every inventor has the right to be mentioned as such and all patent applications must designate the inventor.Footnote 76 This moral right, which is meant to incentivize the inventor to innovate further, may become meaningless upon the extension of the concept of inventorship to AI systems.Footnote 77
The second aspect of the discussion is whether there should be room for AI inventorship. The main argument in favor of this is that it would incentivize research and development in the field of AI.Footnote 78 However, in the absence of compelling empirical evidence, the incentive argument is not convincing, especially since AI systems as such are not susceptible to incentives and the cost of AI invention will likely decrease over time.Footnote 79 Another reason to accept AI inventorship would be to avoid humans incorrectly claiming inventorship. However, the as of yet instrumental nature of AI systems provides a counterargument.Footnote 80 Further, there is no AI-generated output without some form of prior human input. The resulting absence of an act of “conception,” of the process of invention, excludes any extension of the scope of inventorship to nonhuman actors such as AI systems.Footnote 81 Again, however, the “prior input” argument also applies mutatis mutandis to humans. Also as to patent law, therefore, the “act of conception” argument against AI inventorship is susceptible to counterarguments.Footnote 82 A final aspect is that allowing AI inventorship would entail an increased risk of both overlapping sets of patents indicated as “patent thickets,” and the so-called “patent trolls,” which are nonpracticing entities that maintain an aggressive patent enforcement strategy while not exploiting the patent(s) at issue themselves.Footnote 83
11.3.3 Ownership
The next question is how ownership rights in AI-powered creations should be allocated.Footnote 84 As explained earlier, IP law does not allow AI systems to be recognized as either an author or an inventor. This begs the question whether the intervention of a creative and/or inventive AI excludes any kind of human authorship or inventorship (and thus ownership) as to the output at issue. It is submitted that it does not, as long as there is a physical person who commands the AI system and maintains the requisite level of control over its output.Footnote 85 In such a case, IP rights may fulfil their role of protecting the interests of creators as well as provide an indirect incentive for future creation and/or innovation.Footnote 86 However, if there is no sufficient causal relationship between the (in)actions of a human and the eventual end result, the argument in favor of a human author and/or inventor becomes untenable. What exactly constitutes “sufficient” control is tough to establish. A further layer of complexity is added by the black box nature of some AI systems: How can we determine whether a sufficient causal link exists between the human and the output, if it is impossible to find out exactly why this output was reached?Footnote 87 However, both copyright and patent protection may be available to works and/or inventions that result from coincidence or even dumb luck.Footnote 88 If we take a step back, both AI systems and serendipity may be considered as a factor outside the scope of human control. Given that Jackson Pollock may claim protection in his action paintings and given the role that chance plays in Pollock’s creation process, can we really deny such protection to the person(s) behind “the next Rembrandt”?
In copyright jargon, we could say that for a human to be able to claim copyright in a work created through the intervention of AI, their “personal stamp” must be discernible in the end result. If we continue the above analogy, Pollock’s paintings clearly reflect his personal choices as an artist. In patent law terms, human inventorship may arise in case of a contribution that transcends the purely financial, abstract or administrative and that is aimed at conceiving the claimed invention – be it through input or output selection, algorithm design, or otherwise.Footnote 89 In an AI context, different categories of people may stake a claim in this regard.
First in line are the programmer(s),Footnote 90 designer(s),Footnote 91 and/or producer(s) of the AI system (hereinafter collectively referred to as “AI creators”). By creating the AI system itself, these actors play a substantive role in the production of AI-generated output.Footnote 92 However, the allocation of rights to the creator sits uneasily with the unpredictable nature of AI-generated output.Footnote 93 While the AI creator’s choices define the AI system, they do not define the final form of the output.Footnote 94 This argument gains in strength the more autonomous the AI algorithm becomes.Footnote 95 Then again, a programmer who is somehow dissatisfied with the AI’s initial output may tweak the AI’s algorithm, thus manipulating and shaping further output, as well as curate the AI output based on their personal choices.Footnote 96 However, an economic argument against granting the AI creator rights in AI-generated output is that this may lead to “double-dipping.” This would be the case if the creator also holds rights in patents granted as to the AI system or the copyright therein, or if the AI system is acquired by a third party for a fee and the output at issue postdates this transfer.Footnote 97 In both cases, the creator would obtain two separate sources of income for essentially the same thing. Moreover, enforcing the AI creator’s ownership rights would be problematic if the AI system generates the output at issue after a third party has started using it. Indeed, knowing that ownership rights would be allocated to the creator, the user would have strong incentives not to report back on the (modalities of) creation of output.Footnote 98
A similar claim to the AI system’s creator may be made by the AI’s trainer who feeds input to the AI system.Footnote 99 Alternatively, the user who has contributed substantially to the output at issue may claim ownership.Footnote 100 The list of stakeholders continues with the investor, the owner of the AI system and/or the data used to train the algorithm, the publisher of the work, the general public, and even the government. Moreover, some form of joint ownership may be envisaged.Footnote 101 However, this would entail other issues, such as an unnecessary fragmentation of ownership rights and difficulties in proving (the extent of) ownership claims.Footnote 102 It could even be argued that, in view of the ever-rising number of players involved, no individual entity can rightfully claim to have made a significant contribution “worthy” of IP ownership.Footnote 103
As of yet, no solution to the ownership conundrum appears to be wholly satisfactory. The void left by this lingering uncertainty will likely be filled with contractual solutions.Footnote 104 Consequent to unequal bargaining power, instances of unfair ownership and licensing arrangements are to be expected.Footnote 105 A preferable solution could be to not allocate ownership in AI-generated output to anyone at all and instead allot such output to the public domain. Stakeholders could sufficiently protect their investment in AI-related innovation by relying on patent protection for the AI system itself, first-mover advantage, trade secret law, contractual arrangements, and technological protection measures, as well as general civil liability and the law of unfair competition.Footnote 106 However, there is a very pragmatic reason not to ban AI-generated output to the public domain, namely that it is increasingly difficult to distinguish output in the creation of which AI played a certain role from creations that were made solely by a human author.Footnote 107 This could be remedied by requiring aspiring IP owners to disclose the intervention of an AI-powered system in the creation and/or innovation process. However, the practical application of such a requirement remains problematic at present. The prospect of having a work be banished to the public domain would provide stakeholders seeking a return on investment with strong incentives to keep quiet on this point. This could invite misleading statements on authorship and/or inventorship of AI-generated output in the future.Footnote 108 Transparency obligations, such as the watermarking requirement imposed on providers of certain AI systems (including general-purpose AI models) under the EU AI Act, may bring us closer to a solution in this regard, likely combined with a “General-Purpose AI Code of Practice” that is to be drafted under the auspices of the AI Office at the EU level.Footnote 109
11.4 Miscellaneous Topics
In addition to the above, the interface between AI and IP has many other dimensions. Without any pretense of exhaustivity, this section treats some of them briefly, namely the issues surrounding IP infringement by AI systems, the potential impact of AI on certain key concepts of IP law and the growing use of AI in IP practice.
11.4.1 IP Infringement
First, in order to train an AI algorithm, a significant amount of data is often required. If (part of) the relevant training data is subject to IP protection, the reproduction and/or communication to the public thereof in principle requires authorization by the owner, subject to the applicability of relevant exceptions and limitations to copyright. The question thus arises whether actively scraping the internet for artists’ work to reuse in the context of, for example, generative AI art tools constitutes an infringement. At the time of writing, several legal proceedings are pending on this question across the globe.Footnote 110 Importantly, the EU AI Act (1) confirms the applicability of text and data mining exceptions to the training of general-purpose AI models, subject to a potential opt-out on the part of rightholders; and (2) mandates the drawing up and public availability of “a sufficiently detailed summary about the content used for training of the general-purpose AI model.”Footnote 111 Further, in order to ensure that authors, performers and other rightholders receive fair and appropriate remuneration for the use of their content as training data, contractual solutions may be envisaged.Footnote 112
Also after the training process, AI systems may infringe IP rights. By way of example, an AI program could create a song containing original elements of a preexisting work, thus infringing the reproduction right of the owner of the copyright in the musical work at issue. An inventive machine may develop a process and/or product that infringes a patent, or devise a sign that is confusingly similar to a registered trademark, or a product that falls within the scope of a protected (un)registered design. This in turn leads to further contentious matters, such as whether or not relevant exceptions and/or limitations (should) apply and whether fundamental rights such as freedom of expression may still play a role.Footnote 113
11.4.2 Impact of AI on Key Concepts of IP Law
Next, the rise of AI may significantly affect a number of key concepts of IP law that are clearly tailored to humans, in addition to the concepts of “authorship” and “inventorship.” First in line in this regard is the inventiveness standard under patent law, which centers around the so-called “person skilled in the art” (PSA).Footnote 114 This is a hypothetical person (or team) whose level of knowledge and skill depend on the field of technology.Footnote 115 If it is found that the PSA would have arrived at the invention, the invention will be deemed obvious and not patentable. If the use of inventive machines becomes commonplace in certain sectors of technology, the PSA standard will evolve into a PSA using such an inventive machine – and maybe even an inventive machine as such.Footnote 116 This would raise the bar for inventive step and ensuing patentability, since such a machine would be able to innovate based on the entirety of available prior art.Footnote 117 Taken to its logical extreme, this argument could shake the foundations of our patent system. Indeed, if the “artificially superintelligent” PSA is capable of an inventive step, everything becomes obvious, leaving no more room for patentable inventions.Footnote 118 We therefore need to start thinking about alternatives and/or supplements to the current nonobviousness analysis – and maybe even to the patent regime as a way to incentivize innovation.Footnote 119
Questions also arise in a trademark law context, such as how the increased intervention of AI in the online product suggestion and purchasing process may be reconciled with the anthropocentric conception of trademark law, as apparent from the use of criteria such as the “average consumer,” “confusion,” “imperfect recollection” – all of which are criteria that have a built-in margin for human error.Footnote 120
11.4.3 Use of AI in IP Practice
Finally, the clear hesitancy of the IP community toward catering for additional incentive creation in the AI sphere by amending existing IP laws may be contrasted with apparent enthusiasm as to the use of AI in IP practice. Indeed, the increased (and still increasing) use of AI systems as a tool in the IP sector is striking. The ability of AI systems to process and analyze vast amounts of data quickly and efficiently offers a broad range of opportunities. First, the World Intellectual Property Organization (WIPO) has been mining the possibilities offered by AI with regard to the automatic categorization of patents and trademarks as well as prior art searches, machine translations, and formality checks.Footnote 121 Other IP offices are following suit.Footnote 122 Second, AI technology may be applied to the benefit of registrants. On a formal level, AI technology may be used to suggest relevant classes of goods and services for trademarks and/or designs. On a substantive level, AI technology may be used to aid in patent drafting and to screen registers for existing registrations to minimize risk. AI technology may assist in determining the similarity of trademarks and/or designs and even in evaluating prior art relating to patents.Footnote 123 AI-based IP analytics and management software is also available.Footnote 124 Finally, AI-powered applications are used in the fight against counterfeit products.Footnote 125
11.5 Conclusion
The analysis of the interface between AI and IP reveals a field of law and technology of increasing intricacy. As the term suggests, “intellectual” property law has traditionally catered for creations of the human mind. Technological evolutions in the field of AI have prompted challenges to this anthropocentric view. The most contentious questions are whether authorship and inventorship should be extended to AI systems and who, if anybody, should acquire ownership rights as to AI-generated content. Valid points may be raised on all sides of the argument. However, we should not unreservedly start tearing down the foundations of IP law for the mere sake of additional incentive creation.
In any case, regardless of the eventual (legislative) outcome, the cross-border exploitation of AI-assisted or -generated output and the pressing need for transparency of the legal framework require a harmonized solution based on a multi-stakeholder conversation, preferably on a global scale. Who knows, maybe one day an artificially super-intelligent computer will be able to find this solution in our stead. Awaiting such further hypothetical technological evolutions, however, the role of WIPO as a key interlocutor on AI and IP remains paramount, in tandem with the newly established AI Office at the EU level.Footnote 126
Open access