TY - JOUR AU - Hetmank,, Sven AB - Abstract When it comes to the impact of AI on the concept of authorship and inventorship, opinions may differ widely. While some believe that IP will cope with the forthcoming challenges without major adjustments or alterations, others see the system as a whole at stake. The truth may lie in-between, depending on whether one is looking at patent law or at copyright law. I. Introduction It is not a new phenomenon that software and computers are used for creating intellectual achievements. As early as of 1991, a WIPO symposium at Stanford University discussed the implications of Artificial Intelligence (AI)1 for intellectual property law.2 More and more complex self-learning systems, which can produce an innovative or creative performance, are becoming part of everyday life. Thus, it seems to be within reach that there may be creations and inventions without substantial human input. New products are already being developed in “intelligent laboratories” without significant human intervention but only with the help of self-learning simulation programs.3 Another example is the AI created artwork “Portrait of Edmond Belamy” that has been recently sold for 432,500 $ at auction house Christie’s.4 These developments pose a challenge to the IP framework: IP law is based on the assumption that IP rights are granted as a reward for human achievements. This raises the question whether AI will shift paradigms with regard to authorship or inventorship and which options and challenges there are when adjusting the current IP law system (IV. and V.). Before addressing these questions, it is necessary to evaluate the current legal framework (II.) and the arguments against and in favour of IP protection for AI-generated output (III.). II. The current legal framework: IP protection for AI-generated outputs? 1. Patent Law When examining a filed patent application, the concept of inventorship may come into focus under three perspectives: first with regard to the invention as the subject matter of a patent; second with regard to the act of inventing as one of the crucial requirements of patentability; and, third, with regard to the inventor as the one to whom the right to a patent shall belong. a) “Invention” and “Inventive Step” In view of the first two aspects, one might reach the conclusion that the involvement of AI in the inventing process might not pose a huge challenge to the current patent system because of three main reasons. First of all, the use of AI might be kept undisclosed, which means that the question of inventorship will not arise in practice as long as there is no obligation to disclose how the invention was made. Second, in contrast to copyright law the wordings of the patent acts do not require or establish a particular threshold of human input for obtaining a patent.5 The requirements of “invention” and “inventive step” have to be assessed in an objective manner,6 which means that it does not matter how the invention was made, whether it was purely accidental or with the help of AI. What counts is the objective outcome. This objective nature of the inquiry is indicated in every step that has to be taken when examining the patent application. For example, the “state of the art” has to be determined from an objective view. Furthermore, the “person skilled in the art” is a notional person and not a real person. The same applies to the inquiry of the inventive step. The examiner at the Patent Office does not ask how the invention was made, but only whether an imaginary person with average skills in the field of the invention would have found the invention or not. This is the reason why the requirement of inventiveness is more accurately defined by the term “non-obviousness”. It is not the inventive doing of the inventor which needs to come into focus. It is the result of the inventive act which has to be non-obvious for a person skilled in the art. A further indication for the fact that the inventive step is of an objective nature might be seen in the burden of proof, which does not lie with the applicant: it is the examiner who has to show that the claimed product or method was obvious and he can do so only by taking an objective point of view. In the end, this objective nature of the inquiry is also in line with the justifications for patent protection. A patent is given to a useful result and the enrichment of the state of the art by disseminating useful information, no matter how the invention was made. What counts, is the technical progress that is provided by the inventor and not the actual process by which the invention came about. As a third reason, while the current patent law system will cope with the involvement of AI in the inventing process, one might say that it is necessary, but also sufficient that the invention was recognized and disclosed as a useful invention by a human being. The argument brought forward is that the AI is only a tool of a human inventor. The AI might have made the invention, but it still needs someone who is able to recognize the inventive capability of the outcome.7 A little child might find a new groundbreaking substance when doing unauthorized experiments in the kitchen as well as one could have a dream about a new structure of an unknown substance. In both cases, it needs someone who recognizes the inventive capability and who brings the finding into shape for the application. For that reason, the use of AI might be seen as a tool, at least, if there is a human being who put the AI purposively into action and who recognized and understood the relevance of the outcome. b) “Inventor” Things might get much more difficult when it comes to identifying the “inventor”. The term “inventor”, which is used in the patent acts of all jurisdictions, undoubtedly refers to an individual. Therefore, it is widely taken for granted that only individuals can be the owner of an inventor’s right. While, as pointed out above, the requirements of “invention” and “inventive step” are highly objectified, the identification of the inventor is not. Especially in cases of joint inventorship where different persons are involved in the inventing process, courts are paying much attention to the precise intellectual input of everyone who claims to be an inventor. Mere routine work such as starting the AI and reading the output may be considered insufficient. However, the requirements of intellectual and inventive input in order to establish “inventorship” have been reduced by the courts in cases where it would otherwise not have been possible at all to identify an inventor.8 This jurisdiction of the German Federal Court of Justice was established in a case where two individuals achieved the invention in conjunction but the single contribution of every contributor did not exceed the routine work of an average skilled person.9 In such a case, the courts might identify an inventor in order to determine the individual to whom the invention should belong, even if this “inventor” did not contribute any intellectual input. This indicates that the allocation of the patent is strictly to be separated from the requirement of inventive step. If the invention is patentable and involves an inventive step, the only question that remains to be answered is who should be entitled to the invention, not whether anyone should be entitled to the invention at all. This jurisdiction could also be of some importance for AI generated outcome in cases where we have at least a minimum of human participation. After all, it also has to be stressed that the patent offices in Europe do not verify the declaration of the applicant concerning the inventor. Apart from the question whether a false declaration entails criminal liability or not,10 it is up to the true inventor to claim his inventorship11 and one might suppose that AI would not fight for its rights. However, there might be other individuals who could try to claim at least a share in the patent, for example the developer of the AI or the provider of the data used by the AI. It can be assumed that in those cases co-inventorship has to be determined according to the general rules, which means, for example, that the contribution must contain at least a minimum of purposive customization or selection, while purely abstract contributions are not sufficient. Although these questions might be very difficult to answer, the current patent law seems to be well prepared and might cope with the forthcoming changes in the technical world. 2. Copyright law The same questions also arise with regard to “works” generated by AI. It is well established that the software which is the basis of the AI can be protected by copyright.12 The conditions for and the scope of protection have been harmonized by the Software Directive13 as early as 1991. However, it is an intriguing question whether also the output of AI can enjoy copyright protection. The only generally accepted fact is that the products generated by a software are not protected by the copyright in the software, unless the outcome is predetermined by the programme and thus can be attributed to the software engineer.14 For this reason, the scope of the copyright in the software is not comparable to the scope of process patents, which also include products produced directly by the patented process. a) EU copyright law Although large areas of copyright law have been harmonized by the copyright directives, the EU legislation does not provide for a general definition as to who can be considered an author. However, there are specific provisions in the Software Directive and the Database Directive15 with regard to attribution of authorship. These specific provisions do not give a clear picture as to whether non-human agents can be an author under copyright law. It follows from Art. 2 Software Directive and Art. 4 Database Directive that copyright may not only be attributed to natural persons, but also to legal persons, where provided for by national legislation. If the wording of the directives was interpreted extensively, one could argue that if the national law recognizes an AI as a legal person, also the copyright could be attributed to the AI. However, when turning to the prerequisites of protection, the need for a human contribution becomes obvious. Although the legal requirements for copyright protection have not been harmonized on a general level, the specific provisions of Art. 1 (3) Software Directive, Art. 6 Database Directive and Art. 16 Directive 2006/11616 with regard to photographs make clear that an artefact must be an “author’s own intellectual creation” in order to qualify for copyright protection. Starting with the 2009 Infopaq decision, the Court of Justice has extended this concept, which must be given an autonomous and uniform interpretation throughout the European Union,17 beyond these three directives in its jurisprudence also to other works.18 In the 2011 Painer decision, the Court of Justice further elaborated the concept. This case raised the question whether a portrait photograph of a school girl taken by an Austrian photographer qualified for copyright protection. The Court held that “an intellectual creation is an author’s own if it reflects the author’s personality. That 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 (…). By making those various choices, the author of a portrait photograph can stamp the work created with his ‘personal touch’.”19 It can be deduced from this judgement that the Court requires some kind of human input if it states that the creation must reflect the author’s personality. Although the Court did not yet have to decide on the protectability of of computer-generated works, it can be assumed that purely computer-generated works are not eligible for copyright protection.20 b) Germany According to Sec. 2 (2) German Copyright Act, works have to be a “persönliche geistige Schöpfung”, i.e. the author’s personal individual creation. The German copyright law is based on an anthropocentric copyright concept. There is a broad consensus that copyright protection requires a certain level of human involvement. This means computer-generated works without any human involvement do not qualify for copyright protection.21 This is due to the historic foundations of the droit d’auteur systems. The droit d’auteur systems were heavily influenced by the concept of art of the 19th century and thus by the idea of the poor, but creative genius, who must be protected.22 It cannot be denied that these ideas only partially correspond to today's reality. Today, the emotional bond between author and work has been loosened – especially with regard to works of a technical nature such as software. Furthermore, in many cases, the creative effort is not achieved by the creative genius, but by a team of authors, who may also be employees. Nevertheless, the design of today’s copyright law is still based on an anthropocentric foundation, as the protection of droit moral and the term of protection, referring to the author’s death, show.23 It is well established that creators may use technical tools such as software and randomizers. Therefore, so-called “computer-assisted” or “computer-aided” works are protected by copyright, if the author is still involved in substantial decisions with regard to the final design of the work.24 However, it is disputed how influential the human contribution must be with regard to the final design. This question does not only arise with regard to AI output, but has already been discussed with regard to other non-human achievements such as the famous Monkey Selfie case. The action filed by animal welfare organisation PETA on behalf of macaque Naruto raised the question, whether a monkey, that had used a photographer’s camera to take a picture of itself, could own a copyright in the photograph.25 There is broad consensus that the monkey itself cannot be a rights owner.26 However, it has been argued that a copyright might have arisen in favour of the photographer David Slater.27 He had arranged the situation that resulted in the pictures by setting up his camera equipment in such a way that a selfie picture might come about. It was also him who selected this picture out of hundreds of pictures to be published. So it is undeniable that there was some human involvement. However, the crucial question is whether this human involvement would be sufficient for copyright protection. This is to be answered in the negative, since the photographer had too little impact on the final shape of the work. This reasoning is transferable to computer-generated works. The pivotal question here is again whether the human involvement in a specific case is sufficient for copyright protection. It is not enough if a human being causally induces a creation, but does not exert any control over the actual shape of the output.28 Some German scholars argue that when an artist has used a randomizer and thereby created several versions of poems or melodies, the decision to select and accept one of these different versions as the final work would be a sufficient human involvement for copyright protection.29 It is certainly true that the decision to use certain apparatus or tools can constitute one element of creativity. Nevertheless, if the human involvement is restricted to this decision and the selection of a final version, it is claimed here that this limited degree of human involvement is not sufficient – in terms of both the conditions of protection and the allocation of copyright – for justifying copyright protection. The case is different when an artist exerts more influence on the final outcome by determining, e.g., some of the parameters of the creative act undertaken by the computer. Thus the case lies with regard to the stained-glass window of the Cathedral in Cologne created by Gerhard Richter. After deciding that the window should be composed of small squares and that there should be certain repetitive patterns, Gerhard Richter used a randomizer to choose the colours from a defined colour range and to decide on the arrangement of the colours in the window.30 Arguably, this kind of human contribution is sufficient for copyright protection. Until now, there has been little discussion whether the same principles apply with regard to neighbouring rights or whether a lesser degree of human involvement is required. The neighbouring right in photographs according to Sec. 72 German Copyright Act does not presuppose an intellectual creation, but only a minimum level of technical merits. Under this provision, also photographs taken by an automated recording process can be protected under Sec. 72 Copyright Act because of the technical achievement involved.31 Therefore, one can argue very well that a photographer may be granted a related right for photos made by automated recording, even though he did not have complete control of the process of creation. c) UK law The UK Copyright, Designs and Patents Act 1988 (CDPA 1988) in Sec. 9 (3), Sec. 178 explicitly provides for copyright protection of computer-generated works, even if they are created by a computer without the involvement of a human author.32 When the British legislature passed these provisions in 1988, he intended to provide for a protection for automated creations, one example being satellite photographs.33 It seems that the British legislature has resolved this issue by expanding copyright protection, whereas under German law photographs taken by an automated recording process can only be protected by the neighbouring right of Sec. 72 UrhG. Arguably, it is easier for British law to integrate the protection of non-human creations into copyright protection than this would be for droit d'auteur systems, as the 1988 CDPA is more committed to the protection of investments and as the protection of moral rights is less pronounced. However, it is doubtful whether this legislation complies with the prerequisites of protection as they have been construed by the CJEU. Arguably, at least with regard to photographs this legislation is compatible with the EU acquis, since Art. 6 Directive 2006/116 allows the Member States to provide for protection of photographs which are not the author’s own intellectual creation. Any other kind of creation apart from photographs would have to me this standard in order to qualify for copyright protection. However, it is claimed that the protection for computer-generated works only amounts to a related right due to the special provisions on authorship and term. If this assessment is correct, the standard of originality would have to be determined by UK law and not by the EU acquis.34 Moreover, on closer inspection this provision does not create much legal certainty. It dispenses with the need to attribute the creation to a human creator. But the question still remains in whom the copyright to a computer-generated work would be vested. According to Sec. 9 (3) CDPA 1988, this is the person by whom the arrangements necessary for the creation of the work are undertaken. However, it is unclear who this person may be, when several parties are involved in the process of creation, e.g. the programmer, or the person who uses or controls the AI, or the person who feeds the system with the required data.35 It also unclear, how the originality of computer-generated works has to be determined: Does the work have to be original in the sense of not being copied, does it have to be novel, or does it have to be equivalent to an author’s own intellectual creation, thus showing a substantial amount of skill, labour and effort, if it had been produced by a human author?36 3. Protection of AI-generated works by database right? Whereas for the time being copyright protection will have limited relevance for computer-generated products, arguably also the sui generis right in databases may apply. If, e.g., an AI compiles data on a weather phenomenon or DNA sequences, the compilation may be protected as a database under the Database Directive. In order to be protected under Art. 7 Database Directive, a compilation requires a substantial investment in either the obtaining, verification or presentation of the contents. As the sui generis database right is intended to protect an investment,37 it seems convincing that there is no need for a human compiler. It is unclear though, who should be regarded as the maker of a database compiled by AI and thus as the rightholder. It would be conceivable that in the case of computer-generated data collections, a database right would arise in favour of the person who either develops the AI with the goal of compiling the database or in favour of the person who actually uses the AI. It is an intriguing question whether works traditionally protected by copyright such as novels, films and music can also be considered a database. In line with Art. 1 (2) Database Directive, a database is a collection of ‘independent works, data or other materials’. According to the Court, this means that the data have to be separable from one another without their informative, literary, artistic, musical or other value being affected.38 In its Esterbauer decision from 2015, the Court has applied quite a generous standard by holding that “the autonomous informative value of material which has been extracted from a collection must be assessed in the light of the value of the information not for a typical user of the collection concerned, but for each third party interested by the extracted material”.39 On that basis, the CJEU has considered analogue topographic maps to be databases within the meaning of Art. 1 (2) Database Directive. This judgment raises the questions whether the database right can also be applied to other kinds of works. The scope of application of the database right certainly cannot be stretched so far as to apply to novels or musical compositions as such, since single words or letters or musical notes on their own do not possess any informative value. Things may be different with regard to e.g. a recording of a piece of music or single frames of a film. As the music sampling case Metall auf Metall shows, even a short sound sequence can have an artistic relevance as well as a market value and thus can be “separable” from the other sounds contained in the recording.40 So the scope of application of the database right may be broader than it seems at first glance. Thus, one can try to argue that a map compiled by AI or a film made by AI is protected by the sui generis right. It has been called into question whether the Database Directive safeguards the balance between the legitimate interests of manufacturers and of lawful users of databases.41 Against this background, also this possibly very broad scope of application should be taken into account when discussing the need for reform of the Database Directive. III. Arguments against and in favour of IP protection for products generated by AI Having examined the status quo, the question arises whether there should be IP protection of products generated by autonomous agents. This question deserves a more thorough analysis involving also empirical studies and an economic perspective, which cannot be achieved within this contribution. It must be taken into account that – as every restriction of competition – also the granting of an IP Right, which may have monopoly-like effects, needs to be justified and that the advantages must outweigh the disadvantages. On the one hand, it has been stressed that AI is not susceptible to the incentives which are provided by IP Rights42 and that thus other tools of stimulating investments might be sufficient.43 Furthermore, IP rights might foster market concentration and stifle the entry of new ventures into the market, because competitors with access to AI will have huge advantages over those without. A further aspect might be seen in the negative impacts on human efforts to create or to invent. It might be seen as a kind of devaluation of human efforts if automatic creations were rewarded in the same way. On the other hand, IP rights on AI generated output might be necessary to stimulate investments in the research and development of AI, which is one of the most important political missions of the EU in order to keep pace with other economies, e.g., China or the US.44 While it is true that the AI is completely indifferent to incentives, the human developer is not. However, if the basic assumption is accepted that IP rights are providing for effective incentives to stimulate investments,45 the question arises whether it would be sufficient to reward the human developer of the AI with a patent or another IP right on the AI itself instead of providing for a patent or another IP right for the AI generated output. To answer this question, it must first be considered that it is still not clear whether AI is patentable or not, because AI is largely based on algorithms and “algorithms as such” are not patentable.46 Second, even if it would be possible to obtain an IP right in the AI, this might not be a sufficient incentive, because the AI would not be of much value, if thirds parties could freely exploit the innovative or creative output of the AI. The assumption is obvious that investors would be reluctant to invest in the development of AI if they were unable to exclude third parties from using the outcome, e.g., if AI generated inventions might spread freely once revealed. In this case the developer of the AI might not capture enough of the market success to amortize his investments. For that reason, if the political goal is to provide incentives for investments in the development of AI, these arguments support the assumption that not only the AI itself, but also the AI generated output has to be protected by IP rights. A further argument might be found in the fact that patent law is providing incentives not only for innovation, but also for disclosing newfound technologies. Therefore, patents on AI generated output might be necessary to incentivize disclosure and dissemination of those inventions as well as to disclose the use of AI. All in all, these first thoughts show that the assessment whether an IP protection is needed and how this protection should be organized will not be an easy task. IV. Adjusting the current IP law system – options and challenges Even if one should come to the conclusion that more arguments speak in favour of an IP protection of the AI itself as well as the AI-generated output, there might be further hurdles to take, when it comes to fitting this protection into the current IP law system. 1. Copyright law a) The first option would be to abandon the anthropocentric focus of European continental copyright law and to accept that purely AI-generated output without any human involvement can also enjoy copyright protection. This option would abolish the concept of the author’s own intellectual creation and instead focus on the question whether the level of creativity displayed in a work is sufficient, regardless of whether this is human or computer creativity.47 However, in its pure form this option is unworkable since the copyright regime is adjusted to the conditions of human authors. For example, the copyright protection lasts 70 years after the author’s death – a provision, which obviously cannot be applied to robots. These problems could be avoided if the term of protection began with the day of publication. However, if this rule applied to non-human “works” as well as to human works, this would lead to the detrimental effect that unpublished works would not enjoy copyright protection any longer. A crucial argument against such a reform is also that the legislature would have to provide for long transition periods. A further challenge would be the question, how to handle the moral rights provided for by the droit d’auteur systems, which are difficult to justify for AI. Furthermore, the question arises to whom the right should be allocated – to the person programming the software, to the person providing for the data, which is analysed by the AI, or to the person actually applying the AI? In order to solve this problem, Sec. 9 (3) CDPA 1988 allocates the authorship in computer-generated work to “the person by whom the arrangements necessary for the creation of the work are undertaken”, as has been pointed out above. This provision may work with regard to satellite photos. However, with regard to AI, the situation might be more complex, as all of the persons named above may be considered to undertake arrangements. Or the opposite situation might arise: the AI might have such a high level auf autonomy that there is no human being anymore, who has undertaken the arrangements necessary for the creation of a specific work and to whom the copyright could be allocated. Or would it be worthwhile to acknowledge the AI as a legal person and to allocate the IP right to the AI itself? There are some reasons why it would not be a good solution to create an ePersonhood in this respect, since this would neither solve the question who could grant licences nor who would enforce the IP right in case of infringement. Basically the IP right should be allocated in such a way as to provide for an incentive to invest in the development of AI – and probably an allocation to the AI itself would not achieve this objective. Therefore, if both human works and AI-generated works were to enjoy the same protection, this would necessarily entail a radical reform of the current copyright regime, which would result in a significant reduction of protection. Such a radical reform of the copyright system seems to be undesirable or even impossible due to the required transition periods. For these reasons, the integration of AI-generated products into the existing copyright regime would face insurmountable problems. Furthermore, it is doubtful that the products of AI systems without human authors deserve a fully-fledged copyright protection. b) The second option would be not to take any actions with the result that AI-generated output remains unprotected and thus in the public domain, if no neighbouring rights, such as the database rights apply. Whether this is a desirable option, largely depends on the economic assessment. c) The third option would be to create a different legal regime for AI output, either a specific “robot copyright”, a neighbouring right or a sui generis right. Maybe this is what the Committee on Legal Affairs of the European Parliament had in mind when it proposed in its Report on Civil Law Rules on Robotics from January 2017 that criteria for “own intellectual creation” for protectable works produced by computers or robots should be defined.48 One of the questions to be answered would be which substantive protection requirements should apply to such IP rights for AI-generated products – should they require only a substantial investment or a creative achievement? And again, the question arises, to whom the right should be allocated. d) The decision, whether the 2nd or the 3rd option is favourable, largely depends on the economic analysis whether and to what extent IP protection is needed to foster investments in the development of AI without risking any adverse effects. However, it should be highlighted that both options – either creating a new IP regime for AI or leaving AI-generated products in the public domain – encounter one severe difficulty in practice: It is hardly possible to distinguish between works made by humans and works made by AI without having further information on the creative process. You cannot tell from the look of a translation whether it has been made by a human or by a software. In many cases, there may also be a “teamwork” between humans and AI. If, however, human creations enjoy stronger IP protection than products created by AI, then this would give an incentive to conceal that a product has been created by AI. One can argue that this question has to be resolved by the allocation of the burden of proof. At least according to German law, the burden of proof with regard to the factual basis of protectability by means of copyright lies with the person who claims to be the author.49 If too strict requirements were imposed, this would entail disadvantages for human creators. These would then have to document the creative process to prove the existence of their copyright. As an alternative option, one might think of imposing an obligation to disclose the use of AI. However, it is doubtful whether such an obligation could be controlled and enforced. This leaves us with the dilemma that on the one hand the integration of AI works into the copyright regime seems impossible. On the other hand, also the creation of a distinct legal regime for AI “works” would create complex challenges, as it will hardly be possible to distinguish between human and non-human achievements at first sight. 2. Patent law In contrast to copyright law and as indicated above, one might expect that the current patent law system will cope with the emergence of AI. However, there are certain aspects worth taking into consideration. First, it might be worth to think about (re-)establishing the concept of “company inventions”,50 which allows companies to name themselves as inventors. Under German Patent Law, this concept prevailed until 1936. The courts assessed the relation between the employee’s input and the overall performance of the invention. What the employee effortlessly gained from the experiences made in the company should not be attributed to him.51 It is obvious that the inventorship for companies might make it easier to deal with AI generated outcome where it is difficult to identify and to name an individual inventor in the patent application in order to comply with Art. 81 EPC. Second, it could be worthwhile to think about a shortening of the duration of patent protection, because in the era of AI innovation cycles might become increasingly shorter and patent thickets increasingly denser. For the same reason, the publication of a patent application 18 months after the priority date52 might be too late in a world of increasingly shorter innovation cycles. And last but not least, the capabilities of AI might give rise to rethink the requirements for the patent specifications and its structure to make them more concise and better readable for AI which might help to cope with the increasing number of filed patents in the future. V. Conclusion: Does AI shift paradigms? 1. Copyright law From a factual point of view, AI-generated “works” presumably will compete with human works. It can be assumed that AI will be able to create “works” on-demand, which cost less and are more adapted to the customers’ wishes. To certain extent, AI will replace human creators – probably not with regard to the fine arts, which reflect human emotions, but with regard to the more commercial sector such as jingles used for advertising, posters, graphic design, etc. From a legal point of view, “works” created purely by AI or without a sufficient human involvement are ineligible for copyright protection in the droit d’auteur systems, as has been shown above. It is an open question how substantial the human input needs to be to be granted copyright protection. Therefore, neighbouring rights will be of considerable relevance for the protection of non-human achievements. However, it depends on the actions taken by the legislature, whether there will be a shift of paradigms in copyright law. If the legislature decided to grant copyright protection for AI-“works”, this would imply a shift of paradigms, since radical reforms of the current copyright regime would be necessary with regard to the prerequisites of protection, to the scope of protection concerning the moral rights and the term of protection, as has been illustrated above. If the legislature decided to introduce a distinct IP protection for AI-generated “works”, e.g. by neighbouring rights, or no IP protection at all, the legal concept of authorship for now would remain unaffected. However, legal uncertainties will arise from the fact that in many cases a distinction between AI-generated output and human works will not be possible. Furthermore, the need to justify copyright protection for human works will grow, if AI-“works” are perceived to be equivalent or even “better” than human works, but enjoy a lesser degree of protection. Of course the protectability of creative works cannot be measured in a quantitative way. In contrast to patent law, the requirements for copyright protection are subjective and do not depend e.g., on artistic quality or the innovative potential of a creation. Having said this, though, the current strong copyright regime would come under pressure, if creations generated by AI should be rated as more appealing or superior to human creations. It is certainly not a satisfactory answer to grant copyright protection to a human creation regardless of its merits for the only reason that it was made by a human being. So these developments touch upon fundamental issues of the justification of copyright law and arguably may make it more and more difficult to justify copyright protection for works with a low degree of originality.53 2. Patent law With regard to patent law, it is submitted that the current patent system will cope with the emergence of AI as long as there is a minimal human contribution to the invention. However, if AI becomes state of the art, the bar of inventiveness will be raised, because a product or method might not be inventive anymore, if a person skilled in the art would have found it by using AI that was available to the public.54 Moreover, the state of the art will become more uncertain and less obvious, because the capabilities of AI are difficult to assess and much less obvious then the content of a certain document. Additionally, patent examiners might need extensive knowledge of the capabilities of current AI, which is not an easy task. And last but not least, in practice AI might be kept secret to avoid non-inventiveness of their output, which means, that developers of AI might not be willing to file a patent even if it would be possible. This contribution is an extended version of the presentation, which the authors have given at the “GRUR meets Brussels Workshop” on “AI, Robotics & IP – What’s at stake” in Brussels on 5 June 2018. Footnotes 1 For a brief critical analysis of the terminology and the underlying concept Herberger, NJW 2018, 2825 passim. 2 WIPO Worldwide Symposium on the Intellectual Property Aspects of Artificial Intelligence, 25-27 March 1991, available at (accessed 21 January 2019). 3 Medeiros, “The startup fighting cancer with AI”, Wired, 22.3.2016, available at (accessed 12 December 2017); for further examples see Blok, E.I.P.R. 2017, 69 (70), Abbott, MittDPatAnw 2017, 429 (430 ff.). 4 Cohn, Up for Bid, AI Art Signed ‘Algorithm’, New York Times, 22 October 2018, available at (accessed 17 January 2019). 5 See for example Art. 56 EPC: “An invention shall be considered as involving an inventive step if, having regard to the state of the art, it is not obvious to a person skilled in the art.” 6 See Cornish/Llewelyn/Aplin, Intellectual Property: Patents, Copyright, Trade Marks and Allied Rights, 8th ed. 2013, para. 5-31; Kroher, in: Singer/Stauder, PatG, 7th ed. 2016, Art. 56 para. 4; Moufang, in: Schulte, PatG, 10th ed. 2017, Sec. 4 para. 7; Abbott, 57 B.C.L. Rev. 1079 (1110) (2016); Blok, E.I.P.R. 2017, 69 (73). 7 See Blok, E.I.P.R. 2017, 69 (73). 8 German Federal Court of Justice (BGH), GRUR 1966, 558 (560) – Spanplatten; German Federal Court of Justice (BGH), GRUR 1978, 583 (585) – Motorkettensäge; German Federal Court of Justice (BGH), GRUR 2004, 50 (51) – Verkranzungsverfahren. 9 German Federal Court of Justice (BGH), GRUR 1966, 558 (560) – Spanplatten. 10 See on the one hand Stortnik, in: BeckOK PatR, 10th ed. 2018, Sec. 37 para. 10; Moufang; in: Schulte (supra note 6), Sec. 37 para. 22 and Beyerlein, MittDPatAnw 2003, 67 on the other hand. 11 See for example Art. 60 para. 3 and Art. 61 EPC. 12 See in detail Hartmann/Prinz, in: Rechtsfragen digitaler Transformationen, Tagungsband Herbstakademie 2018, p. 769 (774 et seqq.). 13 Directive 2009/24/EC of 23 April 2009 on the legal protection of computer programs (Codified version) OJ L 111, 5 May 2009, p. 16, replacing Council Directive 91/250/EEC of 14 May 1991 on the legal protection of computer programs OJ L 122, 17 May 1991, p. 42. 14 Dreier, FS Kitagawa, 1992, p. 869 (885). 15 Directive 96/9/EC of 11 March 1996 on the legal protection of databases, OJ L 77, 27 March 1996, p. 20. 16 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, OJ L 372, 27 December 2006, p. 12. 17 CJEU, ECLI:EU:C:2018:899, para. 33 = GRUR 2019, 73 – Levola Hengelo/Smilde Foods. 18 CJEU, ECLI:EU:C:2009:465, para. 37 = GRUR Int. 2010, 35 – Infopaq; CJEU, ECLI:EU:C:2010:816, para. 45 = GRUR Int. 2011, 148 – BSA; CJEU, ECLI:EU:C:2011:631, para. 97 = GRUR Int. 2011, 1063 – FAPL/Murphy; CJEU, ECLI:EU:C:2011:798, para. 94 = GRUR 2012, 166 – Painer/Standard; CJEU, ECLI:EU:C:2012:115, para. 38 = GRUR Int. 2012, 435 – Football Dataco/Yahoo; CJEU, ECLI:EU:C:2018:634, para. 14 = GRUR Int. 2018, 1204 – NRW/Renckhoff; CJEU, ECLI:EU:C:2015:10, para. 34 = GRUR Int. 2015, 281 – Ryanair/PR Aviation; CJEU, ECLI:EU:C:2014:25, para. 21 = GRUR Int. 2014, 285 – Nintendo/PC-Box; CJEU, ECLI:EU:C:2018:899, para. 36 = GRUR 2019, 73 – Levola Hengelo/Smilde Foods; see also Rosati, Originality in EU Copyright Law, 2013, passim. 19 CJEU, ECLI:EU:C:2011:798, para. 88 et seqq. = GRUR 2012, 166 – Painer/Standard. 20 See also Ihalainen, 13 (9) JIPLP 724 et seqq. (2018) under 4.; de Cock Buning, EJRR 2016, 310 (314). 21 Peifer, FS Walter, 2018, p. 222 (226 seq.); Loewenheim, in: Schricker/Loewenheim, Urheberrecht, 5th ed. 2017, Sec. 2 para. 45; Bullinger, in: Wandtke/Bullinger, UrhR, 4th ed. 2018, Sec. 2 para. 15 et seqq. with regard to the Berne Convention Ginsburg, 49 IIC 131 passim (2018). 22 Fierdag, Aleatorik in der Kunst und das Urheberrecht, 2004, p. 54; Barudi, Autor und Werk – eine prägende Beziehung?, 2013, p. 115 et seqq. 23 See also Peifer, FS Walter, 2018, p. 222 (225 seq.). 24 Loewenheim, in: Schricker/Loewenheim (supra note 21), Sec. 2 margin no. 38 et seq; Ahlberg, in: Ahlberg/Götting, BeckOK UrhR, Sec. § 2 para. 54 et seq.; Peifer (supra note 21), p. 222 (226); Dreier (supra note 14), p. 869 (877 et seq.); Bullinger, in: Wandtke/Bullinger (supra note 21), Sec. 2 para. 17; Fierdag (supra note 23), p. 68. 25 U.S. Court of Appeals for the Ninth Circuit, 23 April 2018, no. 16-15469 – Naruto v. Slater. 26 Loewenheim, in: Schricker/Loewenheim (supra note 21), Sec. 2 para. 42; Schulze, in: Dreier/Schulze, UrhG, 6th ed. 2018, Sec. 2 para. 9; see also United States Copyright Office, Compendium of U.S. Copyright Office Practices, 3rd ed., 29 September 2017, para. 802.5(C): “Human Authorship – To be copyrightable, musical works, like all works of authorship, must be of human origin. A musical work created by solely by an animal would not be registrable, such as a bird song or whale song. Likewise, music generated entirely by a mechanical or an automated process is not copyrightable. For example, the automated transposition of a musical work from one key to another is not registrable. Nor could a musical composition created solely by a computer algorithm be registered.” 27 Guadamuz, Internet Policy Review, March 2016, Vol. 5 Issue 1, DOI: 10.14763/2016.1.398, p. 8. 28 See also Peifer (supra note 21), p. 222 (226); Bullinger, in: Wandtke/Bullinger (supra note 21), Sec. 2 para. 15 et seqq.; A. Nordemann, in: Fromm/Nordemann, UrhG, 12th ed. 2018, Sec. 2 para. 21; Loewenheim, in: Schricker/Loewenheim (supra note 21), Sec. 2 para. 41; Dreier (supra note 14), p. 869 (881 et seq.). 29 Schulze, in: Dreier/Schulze (supra note 26), Sec. 2 para. 8. 30 See Richter, Zufall: Das Kölner Domfenster und 4900 Farben, 2008. 31 District Court (LG) Berlin, GRUR 1990, 270 – Satellitenfoto; Vogel, in: Schricker/Loewenheim (supra note 21), Sec. 72 para. 24; Schack, Urheberrecht/Urhebervertragsrecht, 8th ed. 2017, para. 723; Lauber-Rönsberg, in: Ahlberg/Götting (supra note 22), Sec. 72 para. 16. 32 Sec. 9 CDPA 1988: “(3) 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”; Sec. 178 CDPA 1988: “ʽcomputer-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”. Similar provisions can be found in the copyright laws of Ireland, New Zealand, South Africa, Hong Kong and India, cf. Ginsburg, 49 IIC 131 note 7 (2018). 33 De Cock Buning, EJRR 2016, 310 (318). 34 Bently/Sherman/Gangjee/Johnson, Intellectual Property Law, 5th ed. 2018, p. 118. 35 Dreier (supra note 14), p. 886 footnote 65; see also Ihalainen, 13 (9) JIPLP 724 et seqq. (2018) under 3. 36 Bently/Sherman/Gangjee/Johnson (supra note 34), p. 117; König, Der Werkbegriff in Europa, 2015, p. 160. 37 Dreier, in: Dreier/Schulze (supra note 26), Sec. 87a para. 1; Vogel, in: Schricker/Loewenheim (supra note 21), § 87a para. 3. 38 CJEU, ECLI:EU:C:2015:735, para. 17 = GRUR Int. 2015, 1161 – Esterbauer. 39 CJEU, ECLI:EU:C:2015:735, para. 27 = GRUR Int. 2015, 1161 – Esterbauer. 40 See the opinion of Advocate General Szpunar, ECLI:EU:C:2018:1002, delivered on 12 December 2018 – Pelham v. Hütter, Schneider-Esleben; it is, however, a different question whether the reproduction of two seconds of a rhythm sequence from a phonogram amounts to a reproduction within the meaning of Art. 2 of Directive 2001/29. 41 See e.g. the criticism pointed out in the first evaluation report of the European Commission of 12 December 2005, available at (accessed 21 January 2019). 42 Abbott, in: Sugimoto/Ekbia/Mattioli, Big Data Is Not a Monolith, 2016, available at p. 14 (accessed 30 January 2019). 43 Yanisky-Ravid/Liu, Cardozo Law Review 2215 (2222) (2017-2018), (accessed 8 February 2019). 44 See Report of the European Parliament with recommendations to the Commission on Civil Law Rules on Robotics (2015/2103(INL)). 45 In copyright law, however, there are doubts as to whether the assumed incentive effect of copyright actually exists and whether alternative incentive mechanisms could possibly lead to more efficient solutions than copyright law is able to provide, Hansen, Warum Urheberrecht, 2009, p. 135 et seqq. (150 et seqq.); Ohly, in: Depenheuer/Peifer, Geistiges Eigentum: Schutzrecht oder Ausbeutungstitel? – Urheberrecht als Wirtschaftsrecht, 2008, p. 141 (144-145). In addition to the incentive function, therefore, complementary arguments are needed for justifying the legitimacy if copyright, in particular personalistic arguments. 46 See Art. 52 Paragraph 2 and 3 of the EPC and the Guidelines for Examination of the EPC concerning “Artificial intelligence and machine learning”, (accessed 30 January 2019). 47 De Cock Buning, EJRR 2016, 310 (320). 48 Committee on Legal Affairs of the European Parliament, Report with recommendations to the Commission on Civil Law Rules on Robotics, (2015/2103(INL)), 27 January 2017, available at (accessed 21 January 2019). 49 German Federal Court of Justice (BGH), NJW 2003, 665 (667) – Staatsbibliothek; German Federal Court of Justice (BGH), GRUR 1981, 820 (822) – Stahlrohrstuhl II; German Federal Court of Justice (BGH), GRUR 1991, 449, 450 – Betriebssystem; Schulze, in: Dreier/Schulze (supra note 26), Sec. 2 para. 70 et seq. 50 In Germany the concept of the so called “Betriebserfindungen” was abolished in 1936, see Witte, GRUR 1958, 163. 51 See RGZ 139, 87, 93 = GRUR 1933, 225 (228) = JW 1933, 1394. 52 Art. 93 para. 1 of the EPC. 53 See more in depth Lauber-Rönsberg, GRUR 3/2019, 244 (251). 54 See Firth-Butterfield/Chae, Artificial Intelligence Collides with Patent Law, White Paper, World Economic Forum, p. 12, (accessed 4 February 2019). In partnership with the German Association for the Protection of Intellectual Property (GRUR) JIPLP exchanges content with GRUR Int., the leading German-language journal specialising in intellectual property law. This section features specially-selected content from GRUR Int. for the benefit of our readers. © The Author(s) 2019. Published by Oxford University Press. All rights reserved. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model) TI - The concept of authorship and inventorship under pressure: Does artificial intelligence shift paradigms? JF - Journal of Intellectual Property Law & Practice DO - 10.1093/jiplp/jpz061 DA - 2019-07-01 UR - https://www.deepdyve.com/lp/oxford-university-press/the-concept-of-authorship-and-inventorship-under-pressure-does-Ak6Oq1f7Yo SP - 570 VL - 14 IS - 7 DP - DeepDyve ER -