Report of the Swiss Group​^*

1. In recent years, the amount of data created, recorded, collected and used all around the world has exploded, and in very diverse fields (automotive, health, building, banking, marketing, etc.). This phenomenon is due, among others, to the development (i) of technologies to record and process data (e.g. sensors, computers) and (ii) of electronic communication and platforms. One of the major consequences of the accumulation of a huge corpus of data is the development of artificial intelligence (AI) and machine learning, because “data is the raw material of AI and the emergence of new uses and applications depends on it.​¹” Data has become so important that it is has been considered as an infrastructure​². 2. An increasing number of modern data applications give rise to legal questions about the protection of data, both in terms of protecting one’s own rights and investments, and in terms of avoiding infringement of rights of others. For instance, data mining processes can be expensive and generate extremely valuable outcomes, but often rely on uses of previously existing data that may be difficult to square with existing exceptions and limitations of IP rights. 3. At the moment, there is uncertainty about rights in data, because in most jurisdictions the legislation doesn’t give clear answers to two major questions: Who owns the data (mere data and database)? Who can access the data? 4. The lack of harmonisation creates legal uncertainty around the use and exploitation of data collections, which can deter investment and innovation. Furthermore, uncertainties around ownership can be a barrier to effective trade and transfer of data. 5. Given the increasing role and value of data in all forms of innovation across industries and countries, it may be time for this to change. 6. However, this does not necessarily mean increasing the overall scope of protection. Although protection may encourage investment, there are also valid concerns about stifling competitive innovation by creating new monopolies on information. 7. The TRIPs Agreement contains provisions on the protection of (i) compilations of data, and (ii) trade secrets​³. First, the TRIPs Agreement contains a provision relating to the protection of compilations of data. Article 10.2 relating to “compilations of data” provides that: “Compilations of data or other material, whether in machine readable or other form, which by reason of the selection or arrangement of their contents constitute intellectual creations shall be protected as such. Such protection, which shall not extend to the data or material itself, shall be without prejudice to any copyright subsisting in the data or material itself”. Under this provision, protection of compilations of data is mandatory. But the second sentence of this Article provides that the protection of compilation of data cannot extend to the data itself. The construction of this second sentence is still open, particularly as to whether it prohibits any protection of mere data by an IP right. Second, the TRIPs Agreement contains a provision dedicated to trade secrets that can apply to data. Article 39 (section 7: Protection of undisclosed information) provides that:

“1. In the course of ensuring effective protection against unfair competition as provided in Article 10bis of the Paris Convention (1967), Members shall protect undisclosed informationin accordance with paragraph 2 and data submitted to governments or governmental agencies in accordance with paragraph 3. 2. Natural and legal persons shall have the possibility of preventing information lawfully within their control from being disclosed to, acquired by, or used by others without their consent in a manner contrary to honest commercial practices so long as such information:

• (a)
  is secret in the sense that it is not, as a body or in the precise configuration and assembly of its components, generally known among or readily accessible to persons within the circles that normally deal with the kind of information in question;
• (b)
  has commercial value because it is secret; and
• (c)
  has been subject to reasonable steps under the circumstances, by the person lawfully in control of the information, to keep it secret.

3. Members, when requiring, as a condition of approving the marketing of pharmaceutical or of agricultural chemical products, which utilize new chemical entities, the submission of undisclosed test or other data, the origination of which involves a considerable effort, shall protect such data against unfair commercial use. In addition, Members shall protect such data against disclosure, except where necessary to protect the public, or unless steps are taken to ensure that the data are protected against unfair commercial use”.

8. This study question addresses the issue of rights in data, in particular IP rights in data. 9. It examines the extent to which data already enjoys protection under current IP and other laws, as well as any gaps or overlaps, such as those that may exist with regard to databases under copyright law and trade secret law. 10. It addresses whether there is a need for a new sui generis right in certain kinds of data, or whether current laws and contractual agreements are sufficient; and the potential right holder, object and scope of protection of any new right. 11. It also addresses whether such a right in data might undercut the existing system of intellectual property rights, unduly restrict the public domain and fundamental rights, distort competition, and hinder scientific research. 12. This study question does not address legal issues of privacy and personal data, i.e. information relating to an identified or identifiable natural person. Legislation and policy issues relating to personal data protection should not be be taken into consideration to answer this questionnaire. 13. This study question raises health data as one example of a data-intensive industry where the issues of rights and access to data are important. This question does not address any issues relating to procedures for obtaining legal approvals for products or procedures, such as pharmaceutical approvals and the like. 14. This study question also raises the topic of Public Sector Information (PSI) as another large source of data to which access may be desired by certain parties for commercial or other purposes. 15. In the context of this study, the term “Data” means any mere information (individual item) of any kind, not aggregated and not arranged in a systematic or methodical way, that is recorded and stored by electronic or other means. 16. The terms “Database” and “Dataset” mean a collection of independent works, data arranged in a systematic or methodical way and individually accessible by electronic or other means​⁴. 17. AIPPI has adopted three resolutions in the past:

• – Resolution on database protection at national and international level (Q182) in 2004​⁵.
  This resolution stated, inter alia:

  “AIPPI recommends that all countries provide for the protection of databases which require substantial investment by means of the sui generis right or other proprietary right subject to the provisions set out below” (first paragraph of the resolution).

  But the recital stated that: “Such proprietary right should not extend to the information and data contained in the database” (recital b).

  One aim of this new study question is to determine whether the recommendation of AIPPI to protect databases has been followed by national/regional legislations.

  Another aim is to determine whether the protection of data should now be extended to mere data.

  Other relevant prior work includes:
• – AIPPI’s Resolution on “Exceptions to copyright protection and the permitted uses of copyright works in the hi-tech and digital sectors” (Q216 and Q216B) in 2010​⁶.
• – AIPPI’s Resolution on “Copyright in artificially generated works” in 2019​⁷.

18. While these Resolutions address related issues, they do not directly ad-dress the broader issue of protection of data and data ownership. Thus, it is considered appropriate to continue this avenue of study with this study question on rights in data. 19. Under national and regional legislation, mere data has traditionally fallen outside the scope of IP protection. 20. But in many jurisdictions, databases can be protected by copyright, if they fulfil the general requirements for protection under copyright, and/or trade secret or unfair competition laws. 21. Furthermore, some jurisdictions have adopted other regimes to provide some measure of protection to investments in creating, collecting and organising databases. These regimes use different instruments to pursue different goals, with different economic rationales. For instance, in the EU, collections of data that are the product of substantial investment have been protected by a sui generis database right since 1996, with little demonstrable success in incentivising investment​⁸. In 2017, the European Commission floated the idea of creating an EU-wide ‘data producer’s right’ that would protect industrial data, but no follow up initiatives were advanced. 22. The aim of this Study Question is to decide whether such specific regimes providing protection of databases are desirable and sufficient, and whether mere data should also be protected in certain circumstances. 23. The process of creating and using mere data and of databases can be divided into three steps. 24. Data exists when it is (i) recorded and (ii) stored in a device. Many industries have been highlighted as big data intensive industries. This is, for instance, the case for automotive and healthcare industries. Example 1: the automotive industry (data generated by sensors). The automotive sector may be trending toward a situation where any device consists of two assets: the physical equipment itself and the data generated from its operation. Automotive manufacturers assemble each vehicle from many components acquired from subcontractors. These components include data sensors, recorders, and communication units. Who is the owner of the data produced during vehicle operation? The subcontractor, the automotive manufacturer, the owner of the vehicle (e.g. a leasor or employer), or the end user?​⁹ Similar issues arise, for example, with collection of vehicular traffic data and weather data. 25. This raises the question of the rights on mere data. In most of the jurisdictions, mere data seems not to be protected by copyright or another IPR. However, the data from millions of cars, taken together, has immense value. Control of and access to mere data are typically controlled by contract. In most cases, such data is subject to secrecy obligations, thereby preventing access to that data by competitors and the public. Particularly in industries with a small number of large players, this may result in an “information monopoly” that provides significant economic advantage. The question then is to consider whether control, access and use of mere data should be subject to a specific legal regime, e.g. a new sui generis right, with specific prerogatives for the owner and specific exceptions and limitations to the monopoly. 26. The structuring and annotation of mere data is crucial, for instance for machine learning. The transformation of mere data to a suitable form for training an AI modelusually requires skill and effort, and can be extremely time consuming and expensive​¹⁰. 27. This raises two issues: (i) the protection of the resulting database by IP rights, and (ii) the access to and use of the mere data to be stored in the database. 28. (i) Database protection A database can be subject to IP rights if it meets the corresponding legal requirements. Indeed, it is generally considered that a database which, by reason of the selection or arrangement of its contents, constitutes the author’s own intellectual creation shall be protected as such by copyright. Databases can be protected by others means. For instance, in the EU, a database which shows that there has been qualitatively and/or quantitatively a substantial investment in either the obtaining, verification or presentation of its contents can be protected by a sui generis right. The data producer can prevent extraction and/or reutilization of the whole or of a substantial part, evaluated qualitatively and/or quantitatively, of the contents of that database​¹¹. A database can also be protected as trade secret or by unfair competition law if it fulfils the corresponding legal requirements. The issue is to determine whether the current regimes are satisfactory or if some modifications and harmonization are desirable in light of the rapid developments in this field. 29. (ii) Access to mere data. Mere data used by the database producer to create the database may have been produced by third parties. Example 2: Works generated by AI (machine learning) Existing works (e.g. paintings) are selected to be integrated in a database. This database is used by an AI system to create an artificially generated work​¹². Source works can be protected by IP right, e.g. copyright. Indeed, when a literary, audio-visual or musical work is integrated in a database, for instance to create AI training data, its copyright protection may extend to such use. Example 3: Applications based on social media Many applications are based on statistical studies of information available on social media. An AI system may be used to examine, for example, the huge number of messages (texts, photos, etc.) posted every day on social media and detect certain relationships or trends in order to provide statistics in very diverse domains (health, marketing, etc.). In these examples, the crucial issue is the lawful access to data. Is there a need for clarification regarding the use of mere data available on internet? Should specific exceptions to IP rights be implemented, for instance for data mining?​¹³ 30. In other circumstances, third parties may wish to obtain access to existing databases as a whole. Example 4: Health data (access to a database) Research in the area of health, and valuation of new medical products and processes, may be greatly facilitated by access to existing databases containing health-relating information. However, such data are confidential and highly valuable information. They may be protected by an IP right (copyright, sui generis right) or by trade secret. A researcher may wish to obtain access to such databases for furtherance of health-related studies or initiatives. The potential desire of third parties to access existing databases is far from limited to this one example. Other examples include browsing and shopping histories, location data, music and video preferences, etc. 31. When a database is used, for instance by an AI system, the result (invention, work, etc.) may be protected by an IP right (patent, copyright, related right, etc.). The issue is to determine whether or not the owner of the mere data and/or the database could claim any right on the result of the AI process. For instance, the resolution on Copyright in artificially generated works (2019) states that:

“AI generated works should only be eligible for protection by Copyright if there is human intervention in the creation of the work and provided that the other conditions for protection are met. AI generated works should not be protected by Copyright without human intervention. This principle is considered to apply to the Working Example as follows: (...) “the AI generated work should be eligible for Copyright protection where there are human data selection criteria for the input into the AI”.

If the “human data selection criteria” confers originality to the final AI generated works, the selector could possibly claim copyright on the final work. No. Under Swiss law, mere data is not subject to any intellectual or other property right that awards exclusivity or comparable rights. Such intellectual or other property rights only exist with respect to the content embodied in or represented by data. Hence, whether or not data benefits from protection under intellectual or other property rights depends upon the type of content of the data at issue. By way of illustration, data may represent a copyrightable work (e.g., if the data represents a literary work or a computer program with individual character). If so, the work is protected under copyright law, and the copyright owner is awarded exclusivity as provided for by copyright law. However, it has to be highlighted that these rights do not protect the mere data as such, but a specific type of content embodied in and represented by such data. Not applicable, as there is no intellectual or other property right that protects mere data as such. Not applicable, as there is no intellectual or other property right that protects mere data as such. Not applicable, as there is no intellectual or other property right that protects mere data as such. Not applicable, as there is no intellectual or other property right that protects mere data as such. Not applicable, as there is no intellectual or other property right that protects mere data as such. As mentioned above, there is no intellectual or other property right that protects mere data as such. However, it may be worth noting that with respect to content that is embodied in or represented by data that constitutes a copyrightable work, Swiss copyright law provides for an exception with respect to text and data mining for research purposes (cf. Article 24d of the Swiss Copyright Act, CopA). Under this exception, it is permissible to reproduce a work for the purposes of scientific research if the copying is due to the use of a technical process and if the work to be copied is lawfully accessed. In this context, it does not matter whether the research is for academic or commercial purposes. No. The situation outside the field of intellectual property or other property rights is comparable to the one within that field (cf. answer to question 1 above): There are no other Swiss laws outside that field of intellectual property or other property rights that offer protection to mere data as such and such other rights only exist with respect to the content embodied in or represented by data. For instance, the Swiss Federal Unfair Competition Act (UCA) prohibits the exploitation of a work result that has been entrusted to a person if such exploitation violates a duty of loyalty or trust (article 5(a) and (b) UCA). It further constitutes an act of unlawful competition under the UCA to take over and exploit as such the market-ready work result of a third party without reasonable own effort by means of technical reproduction processes (article 5(c) UCA). The UCA also prohibits the exploitation of trade secrets that have been unlawfully obtained (article 6 UCA). The unlawful disclosure or exploitation of a trade secret may further constitute a criminal offense (article 162 of the Swiss Penal Code, SPC). Hence, protection under the UCA and the SPC is available with respect to the unlawful disclosure and/or exploitation of certain work results and trade secrets. To the extent that trade secrets or other data is protected by contractual obligations, such as confidentiality or non-use obligations, protection may also be available under the contract law. In addition, there are a number of sector specific laws that award protection to certain information, such as with respect to data contained in applications for approval of pharmaceutical products (article 12 of the Federal Act on Medicinal Products and Medical Devices, TPA) and agrochemical products (article 52 of the Ordinance on the Placing on the Market of Plant Protection Products). In each case, however, the relevant provisions of Swiss law do not protect mere data as such, nor do they grant exclusivity or similar rights. They merely offer limited protection against specific, often narrowly defined unlawful uses of the content embodied in the data at issue, provided that the applicable prerequisites are fulfilled. Not applicable; as mentioned above, there is no other right that protects mere data as such. Not applicable; as mentioned above, there is no other right that protects mere data as such. Not applicable; as mentioned above, there is no other right that protects mere data as such. Not applicable; as mentioned above, there is no other right that protects mere data as such. Yes, as a matter of principle, data is marketable under Swiss law. There are no Swiss law provisions that would limit or restrict the marketability of mere data as such under Swiss law. Any such limitations would only arise with respect to specific content embodied in such data (e.g., under data protection laws). However, for lack of legal protection by way of an intellectual or other property right, there is no exclusivity or similar right in mere data. Hence, mere data may merely be marketed by way of restricting access and use of such data by imposing contractual obligations, such as confidentiality obligations, on the recipient. Not applicable. As mentioned above, there is no other right that protects mere data as such; hence, there are neither any specific exceptions. Yes. While there is no sui generis or specific legal protection for databases under Swiss law, a database may qualify as a collected work and thus be subject to copyright protection insofar as it is an intellectual creation with individual character with regard to its selection and arrangement (article 4 CopA). In addition, a database may potentially be copyrightable as a linguistic or scientific work (article 2(2)(a) CopA; e.g., an encyclopedia) or as a derivative work (article 3(1) CopA), in each case, provided that it is an intellectual creation with individual character. It should be noted that copyright protection of databases as collected work or as any other type of copyrightable work does not protect the mere data as such that is contained in the database. Hence, even if the database may, subject to having individual character, constitute a copyrightable work, the mere data that builds the database as such is not protected by copyright. In other words, the protection of a database as a collected work is a purely structural protection (cf. Decision of the Swiss Federal Court 4A_482/2013 E. 3.2.3). Protection may be awarded under copyright law, provided that the database has individual character and constitutes a copyrightable work. To benefit from copyright protection, the database has to qualify as copyrightable work. This requires that it is a literary or artistic intellectual creation with individual character. To constitute an intellectual creation, the work has to be a creation of a human mind and based on the human will. In other words, a work must be an expression of a statement of thoughts (Decision of the Swiss Federal Court 130 III 168, c. 4.5). If a database were to be created by a machine alone (e.g., by artificial intelligence), without there being any sufficient involvement of a human in the process of its creation, it would not be protected under copyright. Individuality differs from banality or routine work; it results from the variety of decisions taken by the author, from surprising and unusual combinations, so that it seems impossible that a third party faced with the same task could have created an identical work (Decision of the Swiss Federal Court 136 III 225, c. 4.2). With regard to the protection of a database as a collected work in particular, it is required that the database as such amounts to a new work, i.e., to an intellectual creation with individual character, and is not merely the mechanical addition or sequencing of data. By way of example, the sequencing of temperature values from different locations that are measured on an ongoing basis every sixty seconds in a database would not be sufficient for the database to constitute a copyrightable work. Similarly, an alphabetical telephone directory would not qualify as a copyrighted work. Instead, the individual character must result from the selection and arrangement of the data that is in the database. This also means that the data as such does not benefit from copyright protection, even if it is included in a copyrightable database. As a matter of principle, copyright vests in the author, who is the natural person who has created the work (article 6 CopA). If two or more persons contribute as authors to the creation of the work, they jointly own the copyright (article 7(1) CopA). Hence, the initial owners of the copyright in a database will be the author(s) who created such database. They may assign the copyright to a third party upon its creation or thereafter. For instance, employees may assign the copyright in works they create to their employer. The prohibited acts with respect to databases that are protected as copyrighted works are the same as for any other copyrighted work. Copyright confers upon the author the exclusive right to his/her work (article 9(1) CopA) so that the author may decide whether, when and how the work is used (article 10(1) 1 CopA). In particular, the author has the exclusive right to reproduce copies of the work and to recite, perform or present the work (article 10(2) CopA). Further, the author has the right to decide whether, when and how the work may be altered, and whether, when and how the work may be used to create a derivative work or may be included in a collected work (article 11(1) CopA). As mentioned above, the protection of a database as a collected work only concerns the collection as such, but not the mere data included in the collection. If only a part of the collected work is copied, the copyright is infringed only if the copied part as such is protected under copyright in itself, either because it still represents an individual, copyrightable selection, or because it adopts the individual arrangement of the original collection (cf. Cour de justice Geneva, SMI 1994, 183). In contrast, the unauthorized reproduction of the mere data does not infringe upon the copyright in the collection. Yes, there are a number of exceptions to copyright protection. For instance, there are exceptions for private use, for creating archive and backup copies, for creating temporary copies, for quotations, etc. As already mentioned above, Swiss copyright law further provides for an exception with respect to text and data mining for research purposes (article 24d CopA). Under this exception, it is permissible to reproduce a work for the purposes of scientific research if the copying is due to the use of a technical process (e.g., data mining) and if the work to be copied is lawfully accessed. In this context, it does not matter whether the research is for academic or commercial purposes. There is no sui generis or similar specific legal protection for databases under Swiss law. However, the owner of a database may under certain circumstances benefit from protection awarded against unlawful third party use of such database under unfair competition law or trade secret law. It is important to note that the relevant provisions of Swiss law do not protect the database as such, nor do they grant any exclusivity or similar rights. They merely offer limited protection against specific, narrowly defined unlawful uses of certain content, provided that the applicable prerequisites are fulfilled. Protection may be available under unfair competition law and under trade secret law. In addition, protection may be available under certain sector-specific laws. The UCA prohibits the exploitation of a work result that has been entrusted to a person if such exploitation violates a duty of loyalty or trust (article 5(a) and (b) UCA). It further constitutes an act of unlawful competition under the UCA to take over and exploit as such the market-ready work result of a third party without adequate own effort by means of a technical reproduction process (article 5(c) UCA). A database may potentially qualify as work result. It is important to note, however, that article 5 UCA does not protect work results as such, but is only intended to prevent specific unlawful conduct with respect to such work results, such as to avoid free-riding. The UCA also prohibits the exploitation of trade secrets that have been unlawfully obtained (article 6 UCA). The unlawful disclosure or exploitation of a trade secret may further constitute a criminal offense under Swiss criminal law (article 162 of the Swiss Penal Code, SPC). Thus, if and to the extent that a database is a trade secret that is unlawfully disclosed or exploited, protection may be available under the UCA and/or the SPC. Finally, there are certain sector specific laws that award protection to specific information. For instance, data relating to the marketing authorization for pharmaceuticals (article 12 TPA) and agrochemical products (article 52 of the Ordinance on the Placing on the Market of Plant Protection Products) are protected from exploitation by third parties under certain circumstances. The requirements depend upon the type of protection that is sought. To fall within the scope of article 5 UCA, a database has to constitute a work result, i.e., a tangible result of a minimal investment in the creation of the work result. Further, such work result must be exploited in the specific manner sanctioned by article 5 UCA in order to constitute unfair competition. For instance, article 5(c) UCA is triggered only if the market-ready work result is taken over and exploited as such by means of a technical reproduction process without adequate own effort. To benefit from protection as a trade secret, the database has to be kept secret and the owner needs to intend to keep, and to have a reasonable interest in keeping, it secret. Further, protection under trade secret laws is only granted if the trade secret is disclosed or exploited by a person who is not authorized to do so, which typically requires that such person is bound to confidentiality and breaches such obligation, or knowingly abuses such breach by a third party. This depends upon the type of protection that is sought. It may be the person who has created or who is entitled to use the work result as its own, or the person whose trade secrets are at issue. This depends upon the type of protection that is sought. Typically, the rights mentioned above prohibit the unlawful exploitation and/or disclosure of the work results resp. trade secrets at issue. No. Yes. The Federal Act on Freedom of Information in the Administration (FoIA) that applies to the Swiss federal administration and to certain public and private bodies outside the federal administration stipulates the general principle that any person has the right to inspect official documents and to obtain information about the content of official documents, subject to certain exceptions. Thus, the right to access official documents is intended to be the rule, whereas secrecy shall be the exception. In addition, the Swiss Federal Council adopted on November 30, 2018 an “Open government data strategy in Switzerland for the period 2019 to 2023”. Its aim is to grant public access to all open data of the federal administration on the portal “opendata.swiss”, and to develop legislation to provide for such access. Yes. Insofar as the FoIA applies and provides for access to public documents, there is such right. In order to implement an open government data strategy, further amendments to the existing law and regulation may be required. Yes, health data is governed by several statutes. For instance, health data is protected under the Swiss Federal Data Protection Act (DPA), which qualifies health data that relates to an identified or identifiable person as sensitive personal data. The processing of such sensitive personal data is subject to certain additional limitations compared to “normal” personal data, such as specific information duties and the need to justify disclosures to third parties. Furthermore, health data may be protected by way of professional secrecy obligations applicable to doctors and other health practitioners (article 321 SPC) and to persons involved in human research pursuant to the Federal Human Research Act (article 322 SPC), both of which require the relevant professionals to maintain certain information, which often includes health data, secret. The Human Research Act (HRA) further governs the collection and processing of health-related data in the context of research concerning human diseases and concerning the structure and function of the human body. In particular, the HRA requires the data subjects to be informed about the processing of their health data and the measures taken to protect such data. Furthermore, the HRA stipulates a need to obtain consent for certain processing activities in respect of health-related personal data. Health data is further addressed in the Swiss Federal Act on the Electronic Patient File. This law essentially requires that such electronic patient file can only be established with the informed consent of the patient, which can be withdrawn by the patient at any time. There is a controversial discussion ongoing as to whether the electronic patient file should be made mandatory and whether and under which circumstances third parties should have an access right to the information stored in the file. The federal government also looks into whether and how it can improve the sharing of health-related personal data between research institutes and universities to foster personalized medicine. So far, no specific measures have been implemented to that end. There is no general right of third parties to have access to health data; quite to the contrary, as a matter of principle, health-related personal data must not be disclosed to third parties, except with the consent of the data subject, or if there is a specific other justification applicable. The data subject his- or herself has certain access rights under the DPA. We believe that there is no imperative need to improve the current law and practice relating to rights in data. Even if mere data as such is not protected and even if there exists no sui generis protection for databases under Swiss law, Swiss law offers reasonable protection against specific unlawful use (most importantly, under unfair competition and trade secret law) and we believe that the limited protection granted to databases under copyright law is appropriate. In particular, we believe that there is no need to introduce a new sui generis protection for mere data or for databases and we neither believe that the threshold for a work to be considered a copyrighted work should be eased in order to facilitate copyright protection of databases or even of mere data. Overall, we believe that Swiss law reasonably strikes the balance between the different conflicting interests at stake, such as the interest in the protection of proprietary information, copyrightable works and work results, the interest in preventing the monopolization of ideas and concepts, the interest in enabling access to data and information for public benefit, and the interest in reasonable protection of personal data. What is more, we believe that Swiss law has proven to be sufficiently flexible and technology-neutral to enable it to deal with new developments, such as the increased importance of data as a commercial value. Where case law is partially criticized as too restrictive (e.g., in the context of adjudicating alleged unlawful uses of work results under article 5(c) UCA, that is often perceived as being very much in favor of the person allegedly copying such work result), the letter of the existing law allows reasonable room for courts to take such criticism into account, if needed. We do not believe that a specific intellectual or other property right or any other right to protect mere data should be introduced into Swiss law. We do not believe that a specific intellectual or other property right or any other right to specifically protect databases should be introduced into Swiss law. We do not believe that there is a need to amend rules on contract law with a view to protecting mere data or databases. If any amendments with respect to the protection of mere data or of databases were to be contemplated, they should focus on solving practical issues, such as the segregation of and access to data held by a company for a third party in the company’s bankruptcy (e.g., in case of data stored and processed by an outsourcing provider for a third party client, such third party client should have the right to receive its data from the bankruptcy estate). Yes, if rights to protect mere data or databases were to be introduced into Swiss law (which we believe is not needed), such rights should be harmonized. Due to its omnipresent nature and global replicability, we believe it would be inappropriate for each jurisdiction to have a different rule set with respect to rights protecting mere data and databases. We do not believe that mere data should be subject to a specific protection. In particular, we do not think that there should be an intellectual property right in mere data. What may be considered is the harmonized implementation of protection of data created or collected by an individual or a legal entity against specific, narrowly defined unauthorized uses by third parties (e.g., cases of unlawful misappropriation, unlawful access and exploitation of such data). In our view, unfair competition and trade secret laws would be appropriate to deal with such issues, given that they typically do not award absolute protection and exclusivity as intellectual property rights do (which might, in case of mere data, allow the monopolization of mere data, for which we believe there is no justification), but they prevent specific unlawful conduct by third parties. Any such right should be clearly defined, offering protection against narrowly defined unlawful uses only. We believe that article 5(c) UCA could serve as a model for a concept that offers limited protection against the unauthorized reproduction of a work result by means of a technical reproduction by a third party without adequate own investment. Any such law should carefully balance the interests at stake, including, without limitation, the economic value of the work result, its confidentiality, and the public interest in enabling imitation of third party work results that are not protected by intellectual property rights. We believe it should not matter whether the data was generated by humans or by a machine. The person benefitting from such right should be the individual or legal entity who has created or collected the data or invested in its use or commercial exploitation. Prohibitions should be narrowly defined, to cover specific unlawful uses and specific acts that violate the principle of good faith, contractual or other obligations. We do not think that exclusivity rights as they are granted under intellectual property rights should be replicated for mere data. In particular, we believe that any such prohibition should not hinder the independent generation of the same data by a third party, even if it results in the exact same result. Exceptions may be considered, but they should be carefully balanced and take into account all relevant interests at stake, including, without limitation, the interest of the person benefitting from the right in being awarded the relevant protection, and the public interest in preventing the monopolization of mere data. For instance, exceptions for private use, access to personal data and use for research purposes may be considered. Given that we believe that there should not be an intellectual property right in mere data, there is no need for a prohibition of contractual override. Contracts should be a means of enabling third parties to use the data and defining the permitted use; the right should grant remedies against the breach of such obligations by the counterparty. We do not believe that databases should be subject to a specific protection. In particular, we do not think that there should be an intellectual property right in mere data. We believe that the same considerations as with respect to mere data should apply (cf. our answers to questions 14 to 19). Cf. our answer to question 15. Cf. our answer to question 16. With respect to databases specifically, the owner of the right should be the person or legal entity who created the database; in this context, the creation of the database may encompass any step undertaken it its creation, from the definition of the idea (goal, duration, extent, etc.), the type of data, its collection, compilation and evaluation. Cf. our answer to question 17. Cf. our answer to question 18. Cf. our answer to question 19. We believe that the implementation of open data strategies by governments and public research organizations are desirable, provided, however, that the rights of commercial enterprises remain unaffected by such strategies. In other words, the implementation of open data strategies by governments should not impair the right of commercial enterprises to maintain their trade secrets. With a view to the rights discussed above (cf. our answers to questions 15 to 25), we do not believe that there is a need for a specific regime. In any case, health-related personal data will have to be protected appropriately. Our answers include views of in-house counsels working in the field of life sciences, public health and industrial applications.

Report of the Swiss Group​^*

Please answer the below questions with regard to your Group’s current law and practice. Yes. No. N/A If you have answered NO to Q1)a), please skip Q2)-Q6) and proceed to Q7); If you have answered YES to Q1)a), please proceed to Q2). No. In a landmark dated decision dated 1991, the Swiss Federal Supreme Court recognized the descriptive use defense in the context of a word mark: the owner of the Swiss mark “Valser”, which means “from Vals” (a village located in the canton of Graubünden, in Switzerland) requested the cancellation of the subsequent marks “Optima Valsertal”, “Piz Ault Valsertal” and “Primus Valsertal” (“Valsertal” meaning “Vals Valley”), on the ground that they create a likelihood of confusion with its mark. All signs were registered in relation to non-alcoholic beverages, and in particular mineral water. The owner of the attacked smarks responded that it was entitled to indicate that its own products originated from the Vals Valley. The Court held that geographical names such as “Valser” are descriptive and thus belong to the public domain unless they are used as a fanciful denomination or if they have acquired, through intensive use, a secondary meaning. The “fanciful denomination” doctrine was later elaborated to include, for instance, the use of a geographical place in a symbolic way (such as “Etna” [an Italian volcano] for bunsen burners) and “Alaska” for menthol cigarettes, or if it refers to a place which the covered goods evidently cannot originate from – such as “Mont-Blanc” for pens. The Court then ruled that even if “Valser” had acquired a secondary meaning, its owner could not prevent competitors from indicating the geographical origin of their goods. However, such competitors have to make sure to sufficiently distinguish their marks from the prior trade mark. Like in cases of homonyms, it is necessary to balance the conflicting interests. As an obiter dictum, the Court mentioned that the interests of the local resident in the use of the name of a place would be given less weight than the right of the name holder to use his own name as a means of identification (BGE 117 II 321). Descriptiveness of a trade mark can result from word elements as well as from device elements, or even a combination thereof. In general opinion, descriptive motifs are in public domain. E.g. an arrow pointing downwards in a female bust describes the digestive effects of the products in question; due to the differences in color and the graphical representation of the bust and the arrow, there is no relevant likelihood of confusion between the trade marks depicted below (BVGer No. B-5557/2011):     Accordingly, the descriptive use defense is not only available in respect of word trade marks but may be invoked in respect of figurative and combined marks also. In principle, descriptive signs belong to the public domain and are to be kept free for trade. Therefore, there is basically no likelihood of confusion if the conflicting signs only share a descriptive element. However, additional aspects might lead to a likelihood of confusion; according to the Swiss Federal Administrative Court, e.g. the combined trade mark “TheBodyShop” has been infringed by the combined trade mark “TheFaceShop” for identical goods (cosmetics), because the signs shared not only elements belonging to the public domain (“the”, “shop”) but also the structure comprising “the” at the beginning, “shop” at the end as well as an element of equal length referring to the body or a part of the body in the middle (BVGer No. B-2711/2016). Likewise, the following trade marks, both covering goods of class 14, were found by the same Court to be too close:     The reasoning is that although the term “nice watch” belongs to the public domain, the similar graphical representation of the point of the letter “i”, the fonts and structure of the marks create a likelihood of confusion (BVGer No. B-1481/2015). As indicated earlier, the Swiss Federal Supreme Court held that even if a name of a place has acquired distinctiveness, i.e. if it does no longer belong to the public domain, the owner of such a trade mark cannot prohibit more recent competitors from indicating the geographical origin of their goods – provided that adequate measures are taken to ensure a sufficient distinction (BGE 117 II 321). Swiss trade mark law does not provide for such specific conditions; however, the duty of good faith and the prohibition of abuse of rights are cardinal principles of Swiss law and ensue from art. 2(1) and 2(2) of the Swiss civil code (“[E]ach person shall be required to exercise his rights and fulfill his duties in good faith. The manifest abuse of a right shall not be protected by law”) and 2 of the Swiss Act against unfair competition (“[A]ny behavior or business practice that is deceptive or that in any other way infringes the principle of good faith and which affects the relationship between competitors or between suppliers and customers shall be deemed unfair and unlawful”).

• ☒ the reputation of the invoked trade mark
• ☒ the use affects the value of the invoked trade mark by taking unfair advantage of its distinctive character or reputation
• ☒ the use gives the impression that there is a commercial connection with the trade mark owner
• ☐ the use discredits or denigrates the invoked trade mark
• ☒ the product is presented as an imitation or replica of the product bearing the invoked trade mark
• ☒ the way the sign is used, namely purely descriptive and in accordance with honest practices (which is notably not met if there are similarities in other aspects)
• ☐ the party invoking the defence was aware of the invoked trade mark
• ☐ other, namely …

In principle, any kind of evidence can be presented to support the defence. However, surveys aiming to prove e.g. the absence of likelihood of confusion between two marks, have in theory no relevance under Swiss law since assessing a likelihood of confusion is a question of law, not of fact (BGE 126 III 315, 318; BVGer No. B-4A_167/2019).

• ☒ court proceedings (civil)
• ☒ court proceedings (administrative)
• ☒ court proceedings (criminal)
• ☒ IP office
• ☐ other, namely N/A

No No. Please consult with relevant in-house / industry members of your Group in responding to Part III. Yes. Harmonization is always welcome in trade mark law since disputes are increasingly international (cross-border). If YES, please respond to the following questions without regard to your Group’s current law or practice. Even if NO, please address the following questions to the extent your Group considers your Group’s current law or practice could be improved. Yes. No. The Swiss group sees no compelling reasons as to why the descriptive use defence should be limited to word marks. The Swiss group is of the opinion that a descriptive use defense shall be available only if the use is purely descriptive and if there are no similarities in other aspects between the marks at stake. Yes.

• ☒ the reputation of the invoked trade mark
• ☒ the use affects the value of the invoked trade mark by taking unfair advantage of its distinctive character or reputation
• ☒ the use gives the impression that there is a commercial connection with the trade mark owner
• ☐ the use discredits or denigrates the invoked trade mark
• ☒ the product is presented as an imitation or replica of the product bearing the invoked trade mark
• ☒ the way the sign is used, namely purely descriptive and in accordance with honest practices (which is notably not met if there are similarities in other aspects)
• ☒ the party invoking the defence was aware of the invoked trade mark
• ☐ other, namely …

The Swiss group opines that no harmonization is needed about the kind of evidence that can be filed, since there are profound differences between common-law and civil jurisdiction in this respect – and in particular the admissibility and relevancy of survey evidence aiming to assess actual or likely consumer confusion.

• ☒ court proceedings (civil)
• ☒ court proceedings (administrative)
• ☒ court proceedings (criminal)
• ☒ IP office
• ☐ other, namely …

Report of the Swiss Group​^*

The inventor is the natural person originating the technical creation constituting an invention, i.e., producing and recording the essential elements of the technical teaching – regardless of how it was done or which tools were used (Art. 3 PatA​¹; Decision S2018_003 of the Federal Patent Court, 24 August 2018, § 9; M. M. Pedrazzini / Ch. Hilti Europäisches und schweizerisches Patent- und Patentprozessrecht, Bern 2008, 477). The above definition is generally accepted as the “creative principle” (“Principe créateur” or “Schöpferprinzip”). The natural person mentioned as the inventor in the Patent Register benefits from the legal presumption that he/she is indeed the inventor. Nevertheless, because the accuracy of the entry of the inventor is not examined by the Swiss Federal Institute of Intellectual Property​², this presumption may be reversed by the actual inventor in a civil action (Art. 74 § 6 PatA). The civil courts review the inventorship status as part of proceedings requesting an assignment of the right to a patent. The burden of proof lies with the party alleging an incorrect designation of inventorship to demonstrate sufficient probability that the person designated as inventor was not the originator of the invention. The designated inventor is given the possibility to counterargue. Similarly, in the context of an employment relationship, contractual obligations do not determine inventorship, i.e., whether an employee had a contractual obligation to make an invention within the meaning of Art. 332(1) CO​³. Rather, the assessment is to be based on the activity actually performed by the employee and his/her position in the company (Cantonal Court Nidwald, March 11, 2008, Z 05 62 [sic! 2010, 41 et seqq., “Result-Verfahren”]). Yes. The applicant must provide the Swiss Federal Institute of Intellectual Property (IPI) with written confirmation of the name of the inventor (Art. 5(1) PatA). The person named by the applicant is mentioned as the inventor in the Patent Register, in the publication of the application, in the grant of the patent, and in the patent specification (Art. 5(2) PatA). The above applies mutatis mutandis where a third party provides an enforceable judgment establishing that he/she is the actual inventor, rather than the person named by the applicant (Art. 5(3) PatA). As noted above, the accuracy of the designation of the inventor in the application or in the register is not examined by the IPI. Such verification can only occur before the special civil courts competent in patent matters, at the request of a third party (Art. 74(6) PatA). Usually, the designation of the inventor is made in a separate document indicating his/her first and last names, as well as his/her domicile (Art. 34(1) Patent Ordinance​⁴). The designation of the inventor can be included in the patent application (Art. 34(2) PatO). If the designation of the inventor was not filed together with the request, it may still be filed within 16 months of the filing date or priority date, whichever expires earlier (Art. 35(1) PatO). In case of a divisional application (Art. 57 PatA), the inventor(s) can be designated within two months from filing, if the above 16-month term does not expire later (Art. 35(2) PatO). If the designation of the inventor is not filed within the required deadlines, the IPI must refuse the patent application (Art. 35(3) PatO; Federal Supreme Court of Switzerland​⁵, decision 110 II 70). However, an incorrect or incomplete designation of inventor(s) does not inevitably lead to nullity or unenforceability of the patent. Rather, a request for the provision of a declaration that the applicant is the author of the subject invention (Art. 74(6) PatA) or an action for rectification by the actual inventor(s) is issued (Cantonal High Court Zürich​⁶ of 29 January 1991, §2; Commercial Court Zürich​⁷ of 28 August 2007, published in ZR 107 (2008) No. 16, E. 3.2). The applicant or patentee may request the rectification of the designation of the inventor. The request must be accompanied by a declaration of consent of the person wrongly designated as the inventor (Art. 37(1) PatO). If the person wrongly designated as inventor is already listed in the publication(s) of the IPI or is already entered in the Patent Register, the correction shall also be entered in the Register and be published (Art. 37(2) PatO). A request may be filed that the identity of one or more of the inventor(s) should not be mentioned. The renunciation by the inventor of his/her right to be desginated in the patent register and in the publications of the IPI (Art. 6 PatA) shall only be taken into consideration, if the applicant submits a declaration of renunciation by the inventor to the IPI no later than 16 months from the filing or priority date, whichever is earlier (Art. 38(1) PatO). The declaration of renunciation must contain the reference number of the patent application, be dated, and bear the signature of the inventor (Art. 38(2) PatO). The declaration of renunciation that complies with the requirements and the designation of the inventor shall be filed separately; the existence of such titles shall be mentioned in the file (Art. 38(4) PatO). Therefore, these documents will not be immediately accessible in the event of a request for inspection of the file by third parties. There are no regulations or official guidelines specifically applicable to inventions made using AI. The State Secretariat for Education, Research and Innovation​⁸ (SERI) issued a collective report of the federal administration on Artificial Intelligence on 13 December 2019. The report provides that the legal framework in Switzerland is a priori adequate, including the integration of emerging applications and new business models using AI. Therefore, according to SERI, fundamental adjustments to the legal framework are not necessary at this stage, including patent law; see page 103 of the report “Rapport du groupe de travail interdépartemental ‘Intelligence artificielle’ au Conseil fédéral” (French version, no English version is available). No. While not explicitly codified in Swiss law, various provisions exclude or prevent the consideration of an AI entity as an inventor in a patent application. First, Art. 5(2) PatA provides that “the [natural] person named by the patent applicant shall be mentioned as the inventor in the Patent Register [...]” (emphasis added). Second, Art. 34(1) PatO provides that the inventor is to be designated in a separate document together with his/her given name, family name and domicile. Third, various acts before the IPI or Swiss courts require the signature of an inventor, such as for example the correction of inventorship. Moreover, the transfer of rights from an inventor to the applicant requires a declaration of intention by the inventor which can only be provided by a natural person. No, for the same reasons as outlined above.

• (a) Using AI to design a particular type of product or process, when the resulting patentable invention is of the type of product or process intended (e.g., a car designer who wishes to design a car body might start with a general shape, and then use AI to perfect aerodynamic or other characteristics leading to a patentable invention. Here, AI is being used as a tool to help invent, but the intent for the result lies with the user);
• (b) Using AI to achieve a particular intended goal, when a resulting patentable invention made using the AI is not directly related to that intended goal (e.g., an AI system is developed to go through social media data looking for one thing and then discovers a useful relationship leading to a patentable invention that was not an original objective of the system);
• (c) Designing or contributing to the design of the AI algorithm that is used in a) or b);
• (d) Selecting the data or the source of the data that is used to train the AI algorithm used in a) or b);
• (e) Generating or selecting the data or the source of the data that is input to the trained AI algorithm used in (a) or (b); and
• (f) Selecting one from a large number of outputs produced by the AI of (a) or (b) and recognizing it to be a patentable invention.

Most AI inventions concern software-implemented inventions, as opposed to hardware-implemented AI. Currently, this group considers that software-based AI inventions do not fundamentally differ from other computerized techniques, at least as far as intellectual property protection is concerned. Although the group acknowledges that computerized techniques lead to less-and-less tangible human involvement in the inventive process, the group fails to see a good reason to apply a specific regime to inventions made using AI. Therefore, inventorship requirements for inventions made using AI should be the same as for inventions made using any other type of tool. With this in mind, the Swiss group has come to the conclusion that, in all of the above scenarios, humans may potentially be considered inventors provided that their respective contributions meet the requirements outlined above in A1. Whether or not a particular human contribution is sufficient to be acknowledged as inventorship has to be assessed on a case-by-case basis. In particular, this will likely be the case when:

• (a) using AI to design a particular type of product or process, when the resulting patentable invention is of the type of product or process intended; or
• (b) using AI to achieve a particular intended goal, even when a resulting patentable invention made using the AI is not directly related to that intended goal.

Indeed, in each of the above cases, inventorship would likely not be questioned if the human contributors had used another tool (e.g., another computer-aided technique or a mechanical tool) to arrive at the invention. Concerning item (c), i.e., designing or contributing to the design of the AI algorithm that is used in (a) or (b), a similar question arises with respect to computer programs used to create copyrightable material. The authors of the corresponding software are typically not regarded as authors of the works subsequently created by persons using such software, provided that the creative influence of the software authors has no steering impact on such works. Similarly, designing an AI software is unlikely to be sufficient to acknowledge (co)inventorship of an invention made using this AI software. Nevertheless, there may be scenarios where the authors of the AI software (the “AI authors”) or, more generally, the AI entity, may be considered to have sufficiently contributed to the development of an invention. Obviously, this may be the case when the AI authors are also users of the AI entity. In addition, this may also be the case where the AI entity was designed for the specific goal of developing an invention as in (a) or (b) above. Using the same criteria, inventorship may be acknowledged for a human contributor in the scenarios indicated in (d), (e), and (f) above. Where the human contribution is limited to the mere selection of data used to train the AI algorithm, key aspects to consider will be the extent to which this selection impacts the training of the AI entity (and consequently the invention) and/or whether the data selection performed qualifies, in itself, as an inventive contribution. Similar considerations may apply to item (e), which concerns data inputs to the trained AI algorithm (e.g., for inference purposes). However, because the scenario assumed in (e) relates to the utilization of the AI entity after it has been trained (e.g., as in supervised approaches), it is less likely to involve inventive contributions from human users in practice. The case evoked under item (f), i.e., selecting a particular output from a large number of outputs produced by the AI entity and recognizing it to be a patentable invention, may also potentially be acknowledged as inventorship; a parallel can be made with selection inventions. In sum, a sufficient causal link must exist between the human contributions made during the inventive process and the resulting invention, as with any other type of invention. As discussed earlier, Swiss patent law does not provide an explicit definition of an inventor. Therefore, in the event of disputes as to the existence and/or identity of an inventor, the decisions fall to the courts. Thus, depending on requests and allegations of the parties and the case at hand, the courts may have to decide on the minimum degree of involvement required for a natural person to be considered as an inventor of an invention made using AI. No. The involvement (or the contribution) of a computer in (to) a claimed invention is not, on its own, a reason to exclude an invention from patentability in Switzerland. The same conclusion can be drawn, by analogy, for an AI-based contribution made during the inventive process, see e.g., the previous AIPPI Report from the Swiss group: Report of Swiss group, Patentability of computer implemented inventions,sic! 2017, 672 et. seqq. Using AI in the inventive process does therefore not represent a red flag under the Swiss patent law. Yes. In Switzerland, the existing patent law is formulated in a technology-neutral manner and has proven to be sufficiently flexible to reasonably address inventorship issues concerning new technologies in the past, be they related to computer-implemented inventions or other types of technological evolutions. Hence, this group, in agreement with the current opinion of SERI, does not currently believe that the advent of AI requires a change to the existing law see 3. The designation of the inventor does not seem to be directly linked to the policy of fostering innovation. Rather, it is seen merely as a right of the inventor, which can be compared to moral rights of creators of copyrighted works. The incentive for innovation arises from the duration of the monopoly conferred by patent protection and the possibility of effectively enforcing a granted patent. Although there is no demonstrated support that the recognition of an AI entity as inventor or co-inventor would have a positive effect on fostering innovation, this would likely not undermine the policy of fostering innovation in general. Articles 5 and 6 of the PatA seek to implement Art. 4^ter of the Paris Convention for the Protection of Industrial Property, London text and correspond to Arts. 62 and 81 of the European Patent Convention. They respond to a postulate which is of particular interest to inventors working for an employer (Message from the Federal Council to the Federal Assembly concerning the revision of the Law on Patents for Inventions (of 25 April 1950), FF 1950 I 933 (967)). Both of these articles, introduced under the heading of “safeguarding the honour of the inventor”, as well as Art. 4^ter Paris Convention, deal exclusively with the moral rights of the inventor. Historically, the designation of the identity of the inventor has not been motivated by any incentive mechanism for creativity and innovation. It is a recognition of a moral right, or a scientific or industrial honor. It is a moral reward that derives from the inventor’s personal efforts. However, unlike the author of a copyrighted work, the inventor of a patented invention does not have the right to the integrity of the work. Rather, the rights of the inventor are limited to the right to be named in the sense of Art. 5 PatA. The purpose of designating an inventor in a patent application, namely to provide recognition and motivation, does not seem to apply to the case where an AI entity is designated as inventor. However, the potential designation of an AI entity as inventor would presumably not conflict with the potential rights of a third party. Therefore, unless it would prevent the designation of a human inventor, the group does not see a risk that the hypothetical recognition of inventorship by an AI entity in general would undermine the purpose of naming the inventor in Switzerland. As such, this group considers it important to safeguard provisions recognizing that even limited contributions of human inventors should result in their designation as inventors of an AI-based invention. As mentioned above, it is left to the courts to decide what is a sufficient contribution to an invention to be acknowledged as inventorship. Thus, even in the above hypothetical scenarios, a non-designated inventor would have the opportunity to request rectification of the designation of the inventor before a court. Even where an employer fails to designate a human inventor (due to compensation obligations, etc.), the legal system offers the possibility to remedy the deficiency. If YES, please respond to the following questions without regard to your Group’s current law or practice. Even if NO, please address the following questions to the extent your Group considers your Group’s current law or practice could be improved. Yes. This group, in particular members from industry, consider it important that law regarding inventorship is harmonized on an international level. A lack of harmonization could potentially lead to situations where patent applications of the same family would designate different inventors in different countries, which would likely lead to confusion. In this group’s opinion, further harmonization is desired in respect of minimal requirements for humans to be named as inventors, be it alone or alongside AI entities (should they be considered inventors at all). This group does not see a need to distinguish such requirements from those currently applied to the determination of inventorship (or co-inventorship) of an invention made using other types of computer-related inventions (e.g., application programs such as CAD solutions meant to run on standard computer platforms) or any other tool. In the case of AI, such a contribution may potentially take many forms, including designing and/or programming the AI system, selecting suitable input data (e.g., including both training and inference data for supervised training pipelines), pre-processing (e.g., cleaning) such data, if necessary, and parameterizing the AI system (e.g., for the training phase). In addition, human activities may be further necessary to select relevant contents from the outputs produced by the AI system, and/or clean up such contents, if not to improve them a posteriori. Such contributions may potentially qualify as inventive contributions which would in turn make the contributor an inventor. No, this groups fails to see a clear need for this, inasmuch as humans are always involved in the development of an invention made using AI in practice, as noted earlier. Controversial scenarios relating to inventions made without any human contributions at all are evoked in 18. To date, only human beings can be considered as inventors of patentable inventions. To be considered an inventor or co-inventor, an AI entity would first have to be acknowledged with or assigned a particular status in the law. Only then would it make sense to question the extent to which this AI entity should be named as inventor; see also 11.

• (a) Using AI to design a particular type of product or process, when the resulting patentable invention is of the type of product or process intended (e.g., a car designer who wishes to design a car body might start with a general shape, and then use AI to perfect aerodynamic or other characteristics leading to a patentable invention. Here, AI is being used as a tool to help invent, but the intent for the result lies with the user);
• (b) Using AI to achieve a particular intended goal, when a resulting patentable invention made using the AI is not directly related to that intended goal (e.g., an AI system is developed to go through social media data looking for one thing and then discovers a useful relationship leading to a patentable invention that was not an original objective of the system);
• (c) Designing or contributing to the design of the AI algorithm that is used in (a) or (b);
• (d) Selecting the data or the source of the data that is used to train the AI algorithm used in (a) or (b);
• (e) Generating or selecting the data or the source of the data that is input to the trained AI algorithm used in (a) or (b); and
• (f) Selecting one from a large number of outputs produced by the AI of (a) or (b) and recognizing it to be a patentable invention.

As noted above in reference to question 6, any sufficient human contribution to a patentable invention (which is reflected in at least one claim of the corresponding patent application) may potentially be considered to impart inventorship or co-inventorship, as with any other type of invention. Courts will have a final say. No. The same question has been addressed in respect of inventions made using computers; there is no need to distinguish AI contributions from other computer-aided contributions. As a result, a contribution from a computerized process in general or an AI entity in particular should not serve as a basis to exclude an invention from patentability. As expressed earlier, the group is of the opinion that AI tools, at least those currently available, do not fundamentally differ from other, computer-based tools, at least as far as intellectual property is concerned. Therefore, much of the discussion on AI inventorship appears to be based on hypothetical tools (often fantasized as human-like intelligence systems), which, however, do currently not exist. Nevertheless, the group recognizes that the advent of AI gives rise to less apparent, tangible contributions of human inventors. Specifically, there are situations where the actual contributions of human inventors do not exceed contributions that could have equally been made by an AI entity, as in, e.g., chemical compounds identified using automated screening tools. In such situations, one may be of the opinion that the threshold of inventorship has been reduced to an extent comparable to what software tools may achieve. Thus, a more general question concerns the definition of minimal requirements for inventorship. Courts may have to deal with such questions in the near future. Finally, assuming that inventions may be achieved without any human intervention at all, a potential issue arises from the requirement to designate a natural person as an inventor in a patent application, which may potentially lead to a dilemma. Namely, an applicant may be forced to designate a natural person that has not sufficiently contributed to an invention because the AI entity used to achieve the invention cannot legally be designated as inventor. While an incorrect inventor designation would not immediately invalidate a patent in Switzerland, this may be the case in other jurisdictions. Such questions call for a debate as to the need for a sui generis right for AI-related inventions and, more generally, for computer-implemented inventions. Without doubt, intermediate intellectual property rights, e.g., between mere copyrights and patents, would be useful for applicants active in the field of information technology. Ultimately, such rights may not necessarily require the designation of an author or an inventor. Healthcare.

Peter Bigler​^*

Das Thema des diesjährigen AIPPI Swiss Day, der im Anschluss an die Generalversammlung der AIPPI Schweiz am 20. Juni 2019 im Hotel St. Gotthard in Zürich stattfand, lautete «Artificial Intelligence – Exploitation through IP Rights». Der auf Englisch durchgeführte Anlass stiess beim Publikum – wie bereits der letztjährige Anlass zur Blockchain-Technologie – auf reges Interesse. Die öffentliche Diskussion um künstliche Intelligenz bzw. «Artificial Intelligence» (kurz «AI») der letzten Monate hat gezeigt, dass mit der Technologie ein breites Spektrum von Hoffnungen, Erwartungen und Ängsten verbunden ist: Hoffnungen auf intelligente Computerprogramme, die Juristen Recherche und Redaktion von Rechtsschriften erleichtern oder sogar abnehmen, aber auch Ängste vor der «Wegautomatisierung» lukrativer Teile der Anwaltstätigkeit oder gar düstere Zukunftsszenarien über ausser Kon­trolle geratene künstliche Intelligenz. Das Programm des diesjährigen Swiss Day konzentrierte sich diesbezüglich auf die wirtschaftlichen Anwendungen künstlicher Intelligenz sowie deren ­Regelung und Verwertung mittels Immaterialgüterrechten. Die Tagung wurde durch Dr. Lorenza Ferrari Hofer, Rechtsanwältin in Zürich und Präsidentin der AIPPI Schweiz, eröffnet. Sie begrüsste die Teilnehmer und übergab das Wort anschliessend an Dr. Reinhard Oertli, Rechtsanwalt in Zürich, der das dicht bepackte Programm vorstellte und durch die Veranstaltung führte. Den Anfang machte Prof. Andrea Danani vom IDSIA Dalle Molle Institute for Artificial Intelligence​¹. Danani begann seine Präsentation mit dem Titel «Innovation and Business through Artificial Intelligence – an introduction for IP lawyers» mit einer kurzen Vorstellung seines Instituts. Dieses wurde 1988 durch den Unternehmer Angelo Dalle Molle ins Leben gerufen, der über eine Stiftung Forschung auf dem Gebiet der Computertechnologie förderte. Heute ist das Institut Teil der Fachhochschule der italienischen Schweiz​² und widmet sich Themen wie Machine Learning, Deep Neural Networks und Data Mining. Im Anschluss daran erläuterte Danani den Anwesenden zunächst die Grundlagen und technischen Begriffe von Artificial Intelligence. Das ultimative Ziel seien Maschinen und Systeme, die nahtlos mit Menschen interagieren könnten, z. B. indem sie menschliche Sprache verstehen sowie aus eigener Erfahrung lernen und so ihre Aufgabenerfüllung verbessern könnten. Dafür seien jedoch riesige Mengen an Rechenleistung nötig. Waren solche früher ­ausschliesslich auf Supercomputern verfügbar, böten Privatcomputer und insbesondere auch tragbare Geräte wie Smartphones und Tablets seit ca. 2010 endlich auch ausreichend Leistung, um Programme mit künstlicher Intelligenz auch bei den Endanwendern laufen zu lassen. Dadurch sei es in den letzten Jahren zu massiven Fortschritten gekommen. Im Zentrum stehe dabei das sog. Machine Learning: Computer­programme bzw. deren Algorithmen hätten bis in die 1980er-Jahre in der Regel nur genau das gekonnt, was Programmierer vorgesehen hatten. Für komplexere Aufgaben wie z. B. die Bild­erkennung sei dies unzureichend, da nicht jede Eventualität vorhergesehen und programmiert werden könne. Beim Machine Learning würden deshalb ­grosse Datenmengen so strukturiert, dass sich daraus Modelle und aus diesen Modellen wiederum Vorhersagen ab­leiten liessen. Ein damit «trainierter» Algorithmus sei so in der Lage, aus den Datenmengen weitere Vorhersagen ­abzuleiten. Eine Steigerung hiervon sei das sogenannte Deep Learning, bei dem ein Netzwerk von Algorithmen (üb­licherweise als neuronales Netzwerk bezeichnet) selber Datenmengen strukturiere und aus eigenen Fehlern lerne. Der Referent präsentierte danach einige Beispiele: Den Anfang machte der «Long Short-Term Memory (LSTM)»-Algorithmus​³, der 1997 von Forschern des Instituts veröffentlicht wurde. Der heute weit verbreitete Algorithmus habe zahlreiche Anwendungen, darunter insb. die Bilderkennung. Er werde z. B. in der Medizin dazu eingesetzt, Bilder von Zellkulturen auf Krebszellen zu untersuchen. Die Erfolgsquote liege mit 80 bis 90% z. T. höher als diejenige von erfahrenen Ärzten. Dass jedoch auch solche Algorithmen z. T. an ihre Grenzen stossen können, zeigte Danani mit einem unterhaltsamen Beispiel, bei dem Bilder von Chihuahua-Hunden von denjenigen von Rosinen-Muffins unterschieden werden sollen. Was auf den ersten Blick selbstverständlich erscheint, erweist sich bei genauerer Betrachtung als Herausforderung, da Fell- und Teigfarbe je nach Foto fast identisch sind und im Muffin steckende Rosinen bei flüchtiger Betrachtung leicht für Augen gehalten werden können​⁴. Danani präsentierte anschliessend eine Tour d’Horizon über die Projekte, an denen sein Institut beteiligt ist. Die Beispiele reichten von Flugsoftware für Drohnen, die mittels Flugvideos selbständig lernt, Hindernisse zu umfliegen, über die Optimierung von Arzneimittelzusammensetzungen und vir­tuelle Gesundheitsassistenten, die den Nutzern die wahrscheinlichsten Auswirkungen des eigenen Lebenswandels aufzeigen, bis hin zu Software für Armeen, die den Kommandierenden in unübersichtlichen Gefechtssituationen strategische Empfehlungen liefern soll. Aus dem Publikum kam anschliessend die Frage, wie das Institut die Rechteverteilung an den zum Einsatz gelangenden Algorithmen bzw. deren Arbeitsergebnissen handhabt. Der Referent antwortete darauf, sein Institut behalte in der Regel die Rechte an der Software bzw. den Algorithmen, während deren Arbeitsergebnisse denjenigen Unternehmen gehörten, die die künstliche Intelligenz eingesetzt hätten. Danani schloss sein Referat mit der Feststellung, künstliche Intelligenz sei nicht bloss ein kurzzeitiges Phä­nomen, sondern werde unseren Alltag nachhaltig verändern. Man müsse sich aber stets bewusst sein, dass das hinter künstlicher Intelligenz stehende ­Machine Learning lediglich ein «Handwerk» sei, keine unfehlbare Wissenschaft. Künstliche Intelligenz sei stets nur so gut wie die Daten, mit denen sie gefüttert worden sei. Deshalb sei es wichtig, dass der Mensch deren Entscheidprozesse bei Bedarf nachvollziehen könne, anstatt ihr blind zu vertrauen. Er sprach damit u. a. indirekt Probleme wie etwa der «Algorithmic Bias»​⁵ oder Fragen der Haftung für Fehler künstlicher Intelligenz an, die immer wieder Thema der öffentlichen Diskussion um künstliche Intelligenz sind. Das zweite Referat der Tagung mit dem Thema «AI and copyright/trademarks – the international and EU approach» wurde durch Prof. Dr. Jan Bernd ­Nordemann, Rechtsanwalt in Berlin, Deutschland, präsentiert. Nordemann betonte dabei gleich zu Beginn, dass künstliche Intelligenz im Urheberrecht allgemein weit mehr Fragen aufwerfe als im Markenrecht. Zur Veranschau­lichung präsentierte er zwei Beispiele: Den Anfang machte Google Clip, eine (mittlerweile bereits wieder aus dem Vertrieb genommene) Weitwinkel­kamera, die z. B. an einer Party auf­gestellt werden kann und mithilfe künstlicher Intelligenz selber entscheidet, welche Motive eines Schnappschusses würdig sind. Als zweites Beispiel stellte Nordemann das Projekt «The next Rembrandt»​⁶ vor. Bei diesem lernte eine Software anhand detaillierter Scans der bedeutendsten Werke des niederländischen Malers Rembrandt, was ein typisches Porträt des Künstlers ausmacht, und erstellte anschliessend selber einen «neuen» Rembrandt. Bei beiden Beispielen stelle sich mangels (direkten) menschlichen Inputs un­weigerlich die Frage nach dem Schöpfer bzw. der Urheberin im Sinne des Urheberrechts. Seine Präsentation gliederte Nordemann in vier Bereiche: (I.) das Rechtemanagement mithilfe künstlicher Intelligenz; (II.) den rechtlichen Schutz künstlicher Intelligenz; (III.) Werke und Arbeitsergebnisse, die durch künstliche Intelligenz geschaffen wurden, und (IV.) Rechtsverletzungen durch den Einsatz künstlicher Intelligenz. Das Rechtemanagement mittels künstlicher Intelligenz werde, so die Prognose Nordemanns, die Anwalts­tätigkeit nachhaltig verändern. Gerade im Markenrecht würden Recherchetools bereits zunehmend «intelligenter», was Anwälten die Recherche und Beurteilung potenZieller Kennzeichenkonflikte erleichtere. «LegalTech»-Produkte wie z. B. Bilderkennungsdienste könnten Anwälten z. B. weiter dabei helfen, ­Urheberrechtsverstösse im Internet zu identifizieren und verfolgen. Was den rechtlichen Schutz künstlicher Intelligenz bzw. deren Infrastruktur anbelange, gälten z. B. im Markenrecht die allgemein bekannten Regeln. Der urheberrechtliche Schutz biete dagegen einige bisher nicht gekannte Herausforderungen. So stelle sich z. B. die Frage, ob Software, die sich selbst verbessere, auch für die von ihr selbst geschaffenen Teile Urheberrechtsschutz beanspruchen könne. Ein weiterer ­Problemkreis betreffe die Daten, mit denen die fragliche künstliche Intelligenz ­trainiert worden sei. Hier existierten z. T. Spezialregelungen wie etwa die EU-Datenbankrichtlinie​⁷. Schliesslich müsse gefragt werden, wie das Ergebnis des «Trainings» der künstlichen Intelligenz geschützt werden könne, sei es als Software oder Datenbank. Was den Schutz der Arbeitsergebnisse künstlicher Intelligenz anbelange, verwies Nordemann auf die beiden Eingangsbeispiele. Denn Urheberrechtsschutz setze in den meisten Ländern einen Urheber bzw. eine Schöpferin in Form einer natürlichen Person voraus​⁸. Genau diese Voraussetzung fehle aber im Falle künstlicher Intelligenz in der Regel. Interessant sei in diesem Zusammenhang die Formulierung im britischen Urheberrechtsgesetz​⁹. Dieses kenne für computergenerierte Werke eine Spezialregelung, die an die «nötigen Vorbereitungen» zur Schaffung des Werks anknüpfe und damit weiter gefasst sei als das reine Schöp­ferprinzip​¹⁰. Doch auch hier bedürfe es letztlich eines menschlichen Inputs. Schliesslich komme es mit zunehmender Verbreitung künstlicher Intelligenz auch immer häufiger zu Situationen, in denen die fragliche Software selbst durch ihr Verhalten Rechte Dritter verletzen würde. Nordemann präsentierte auch hierzu verschiedene ­Beispiele aus der Praxis, die denkbare Verletzungen des Marken-, Lauterkeits- und Urheberrechts illustrieren: Das erste betraf einen Gerichtsfall aus dem Vereinigten Königreich. Der Versandhändler Amazon hatte, wenn der Nutzer auf dessen Website nach bestimmten Marken suchte, stattdessen Produkte von Drittanbietern bzw. eigene Produkte angeboten​¹¹. Das «Keyword-Advertising» und die damit verbundenen (insb. markenrechtlichen) Probleme hängen zwar nicht primär mit dem Einsatz künstlicher Intelligenz zusammen. Der Referent warf hier aber die Frage auf, wie solche Fälle zu beurteilen wären, wenn z. B. eine virtuelle Assistentin wie Amazons Alexa selbständig eine Produkteauswahl trifft. Das zweite Beispiel stammte aus Deutschland und handelte von einer Software für Arztpraxen, die die vom Arzt eingegebenen Originalmedikamente selbständig durch entsprechende Parallelimport-Produkte ersetzte​¹². Das OLG Hamburg erblickte in diesem Verhalten einen Verstoss gegen das Lauterkeitsrecht. Das dritte Beispiel zeigte schliesslich ein tragbares Hilfsgerät für sehbehinderte Personen. Dieses nutzt eingebaute ­Kameras sowie künstliche Intelligenz dazu, dem Träger die unmittelbare ­Umgebung zu beschreiben, darunter Objekte, Texte sowie Personen. Der Referent stellte hier die Frage in den Raum, ob solche Geräte, wenn sie z. B. in einem Museum urheberrechtlich geschützte Werke beschreiben oder vorlesen, in die entsprechenden Urheberrechte eingreifen. Die Beispiele des Referenten zeigten, dass künstliche Intelligenz grundsätzlich die gleichen Rechteverstösse begehen kann wie etwa natürliche Personen. Bei künstlicher Intelligenz kommen jedoch zusätzlich Fragen der Verhaltenszurechnung (z. B. zum Hersteller oder Endnutzer) hinzu. Nordemann stimmte im Fazit seiner Präsentation Danani zu, dass künstliche Intelligenz die Welt nachhaltig verändern dürfte. Für Juristen könne die Technologie je nach Fachgebiet unterschiedliche Auswirkungen haben. Das Markenrecht an sich sei bereits weitgehend bereit für Anwendungen künstlicher Intelligenz. Die darin tätigen Anwälte könnten einerseits zwar auf neue und hilfreiche Arbeitswerkzeuge hoffen, es drohe andererseits aber auch ein Wegfall gewisser Tätigkeiten (und damit Einkommensquellen) aufgrund der erwarteten Automatisierung. Alles in allem werde aber auch mit künstlicher Intelligenz genügend Arbeit für Markenrechtsspezialisten bleiben. Auch die Verwaltung von Urheberrechten sowie insb. die Verfolgung von Urheberrechtsverletzungen dürften von der Unterstützung durch künstliche Intelligenz profitieren. Das Urheberrecht selber hinke den neuesten technischen Entwicklungen aber hinterher und werde entsprechend Lösungen für die sich stellenden neuartigen Probleme wie die Rechte an von künstlicher Intelligenz generierten Inhalten finden müssen. Der Referent schloss sein ausgezeichnetes Referat mit einer Aussage des US-amerikanischen Musikers William Adams (auch bekannt unter seinem Künstlernamen will.i.am). Dieser vergleicht den Stand künstlicher Intelligenz heute mit demjenigen des Internets 1987. Adams meint, damals hätte sich niemand erträumen lassen, welchen Stellenwert das Internet in unserer Gesellschaft heute einnimmt. Gleiches gelte auch für künstliche Intelligenz. Nachdem sich das zweite Referat marken- und insb. urheberrechtlichen Fragen gewidmet hatte, fokussierte das dritte Referat auf das Patentrecht. Dr. Beat Weibel, Chief IP Counsel der Siemens AG in München, Deutschland, referierte zum Thema «AI and patentable inventions». Künstliche Intelligenz berühre im Patentrecht im Wesent­lichen drei Teilgebiete. Erstens sei hier wiederum an Recherche- und Forschungswerkzeuge zu denken, die ­(Patent-)Anwälte bei ihrer Arbeit unterstützen würden. Er selber wolle seine Präsentation aber auf die beiden an­deren Aspekte konzentrieren, nämlich zum einen den patentrechtlichen Schutz von Erfindungen, die künstliche Intelligenz zum Gegenstand haben, und zum anderen den Schutz von mittels künstlicher Intelligenz geschaffenen Erzeugnissen. Weibel zeigte zunächst einen kurzen Film darüber, wie Siemens künst­liche Intelligenz im Betrieb einsetzt. Die Anwendungsbereiche reichten dabei von Forschung und Entwicklung bis hin zu Unterhalt und Wartung der fertigen Produkte. So konnte etwa die Effizienz einer Gasturbine dank künstlicher Intelligenz um 20 % gesteigert und deren Lebensdauer beachtlich verlängert ­werden. Während des Betriebes der Produkte soll entsprechend trainierte Software unter dem Stichwort «preventive maintenance» zudem zu erwartende Defekte an Maschinen vorhersehen und so eine rechtzeitige Wartung statt aufwändiger Reparaturen ermöglichen. Im Anschluss kam Weibel zum Kernstück seiner Präsentation, zur ­Patentierbarkeit von Erfindungen, die künstliche Intelligenz zum Gegenstand haben. Solche Erfindungen stellten den Patentanmelder vor eine Reihe von Problemen: Zum einen seien sie ihrer Natur nach stets computerimplementierte Erfindungen und hätten zudem oft einen engen Bezug zu Geschäftsmethoden und der Aufbereitung und Wiedergabe von Daten, weshalb ihnen die grundsätzlichen Ausschlusskriterien der Patentierbarkeit nach Art. 52 Abs. 2 EPÜ entgegenstünden. Zum anderen stellten die Entscheide künstlicher Intelligenz oftmals eine «Blackbox» dar, was zur Beanstandung mangelhafter Offenbarung in der Patentanmeldung (Art. 83 EPÜ) führen könne. Der Referent zeigte am Beispiel des EPA, mit welcher Begründung derartige Patente in der Praxis oft bemängelt bzw. verweigert werden. Es sei deshalb zwingend nötig, die Erfindung nicht nur in den Ansprüchen, sondern auch in der Beschreibung und den Zeichnungen in einen ausreichenden technischen Kontext zu stellen. Für die anwesenden Praktiker besonders interessant war, dass Weibel in der Folge anhand konkreter Formulierungsbeispiele und EPA-Entscheide zeigte, wie sich Beanstandungen überwinden lassen und wie z. B. gemischte Ansprüche (d. h. Ansprüche die sowohl technische als auch nichttechnische Merkmale aufweisen) formuliert werden sollten. Was die ausreichende Offen­barung anbelange, seien auch Fluss­diagramme, die den groben Ablauf der Datenver­arbeitung abbildeten, sehr hilfreich. Allerdings sei die Offenbarung bei Erfindungen rund um künstliche Intelligenz per se heikel. Denn der Funktionsablauf der Erfindung, d. h., wie z. B. eine künstliche Intelligenz von den Ausgangsdaten zum Ergebnis gelange, sei gerade der eigentliche Wert der Erfindung. Als Faustregel sei es deshalb sinnvoller, die konkreten Anwendungen zu patentieren als die Technologie selber. Dies gelte umso mehr, als Teile der Technologie (z. B. die verwendeten Algorithmen) oft ohnehin z. B. als Open Source erhältlich seien und entsprechend für jeden zugänglich. Schliesslich würde in der Praxis auch häufig eine Kombination von Anspruchsarten («claim-sets», z. B. betreffend Training der Software, der Speicherung der Daten, Vorrichtung zur Ausführung etc.) verwendet, um einen möglichst umfassenden Schutz der Erfindung sicherzustellen. Dies sei zwar aufwendiger und damit teurer, zahle sich aber später bei der Durchsetzung der Patente aus. Den dritten und letzten Teil seiner Präsentation widmete Weibel der ­Patentierbarkeit von durch künstliche ­Intelligenz generierten Ergebnissen. Neben den bereits in den voran­ge­henden Referaten diskutierten (ur­heber­recht­lichen) Werken wie etwa dem «neuen» Rembrandt drehe sich die Diskussion im Patentrecht primär um Gegenstände wie z. B. Bauteile, die von einer Software nach gewissen Vorgaben entwickelt wurden. Dabei ergebe sich jedoch oft das Problem, dass die an der Entwicklung beteiligten Experten nicht nachvollziehen könnten, warum die Software ein Bauteil genau so gestaltet hat. Er illustrierte dies am Beispiel einer Aufhängung, die von einer Software mit künstlicher Intelligenz verbessert wurde. Die Ingenieure waren bei der Betrachtung des Endergebnisses überrascht, wie die Software die einzelnen Befestigungspunkte der Aufhängung miteinander verband. Für die Frage der Patentierbarkeit bedeute dies Folgendes: Neuheit und Nichtnaheliegen seien im Allgemeinen regelmässig erfüllt. Allerdings fehle es streng genommen wiederum am menschlichen Erfinder, dem das Patent zustehe​¹³. Maschinen bzw. Software könnten mangels Rechtspersönlichkeit offensichtlich keine Erfinder sein. Als Lösung schlug Weibel vor, den Erfinderbegriff auf juristische Personen auszuweiten. Als Erfinder der Arbeits­ergebnisse würde damit diejenige (juristische) Person gelten, welche die frag­liche Software oder Massnahme betreibt. In der anschliessenden Diskussion wurde der Vorschlag des Referenten rege diskutiert. Aus dem Publikum wurde diesbezüglich zur Vorsicht gemahnt, weil ein solches System zu einer Flut von Patenten führen könnte. Dem wurde entgegengehalten, dass nicht jede Erfindung auch zwingend patentiert werden müsse. Wo z. B. eine Software zahlreiche verschiedene Lösungen für ein Problem präsentiere, dürfte sich zudem automatisch die Frage stellen, ob die jeweiligen (kleinen Abweichungen) noch ausreichen, das Endergebnis erfinderisch erscheinen zu lassen. Schliesslich wurde aus dem Publikum auch noch eine weitere interessante ­Anwendungsmöglichkeit für künstliche Intelligenz angedacht. Denn genauso wie eine Software eingesetzt werden könne, um ein Bauteil mit bestimmten Parametern zu erschaffen, so könne sie auch eingesetzt werden, um genau jene Parameter und damit den Patentschutz eines Dritten zu umgehen. Künstliche Intelligenz dürfte also sowohl auf die Patentanmeldung als auch auf allfällige Verletzungsprozesse interessante Auswirkungen haben. Der letzte Abschnitt der Tagung widmete sich der Vorstellung konkreter Anwendungsfälle. Hierzu referierten Sonia Cooper, Senior Patent Attorney bei Microsoft in London, Vereinigtes Königreich, sowie Léonard Bouchet, Responsable Données et Archives bei RTS – Radio Télévision Suisse in Genf. Den Anfang machte Bouchet, der über den Einsatz künstlicher Intelligenz bei RTS berichtete. Bouchet zeigte das ­Archivsystem des RTS. Früher hätten Mitarbeiter von Hand unzählige Stunden Videomaterial sichten, verschlagworten und mit Beschreibungen versehen müssen, damit Journalisten bei Bedarf das richtige Material, z. B. einer bestimmten Person, finden konnten. Hier versuche man nach und nach, diese Arbeiten durch künstliche Intelligenz erledigen zu lassen, die diese Arbeiten oft sogar effizienter erledige als Menschen. Die Arbeit für die mensch­lichen Mitarbeiter falle jedoch nicht weg, sondern verlagere sich vielmehr: weg von der Kategorisierung an sich hin zur Verwaltung der anfallenden Datenmengen sowie zur Kontrolle und wo nötig Korrektur der Ergebnisse. Ein ­weiterer Anwendungsbereich von Bild­erkennungssoftware sei etwa, dass ein Journalist ein Bild einer Person hoch­laden könne, worauf das System diese identifiziere und entsprechend weiteres Bild- und Videomaterial präsentiere. Vorteil des neuen Systems sei, dass viel grössere Datenmengen katalogisiert und damit zugänglich gemacht werden könnten, als dies bisher von Hand der Fall war. Der Einsatz künstlicher Intelligenz berge aber auch neue Herausforderungen. Mit der gesteigerten Datenmenge werde es auch schwieriger, die jeweiligen Rechteinhaber zu identifizieren, da die Software z. B. den Abspann eines Videos finden und die darin aufgeführten Informationen verstehen lernen müsse. Seitens RTS freue man sich jedoch über die zahlreichen Möglichkeiten, die künstliche Intelligenz biete, und arbeite diesbezüglich auch aktiv mit anderen Medienunternehmen zusammen. Als Zweite stellte Cooper den Einsatz von künstlicher Intelligenz bei ­Microsoft vor. Als einer der weltweit grössten Software- und Technologie­konzerne nutze Microsoft künstliche Intelligenz in einer Vielzahl seiner ­Produkte, auch wenn dies oft für den Endabnehmer gar nicht ersichtlich sei. Die öffentliche Angst vor künstlicher Intelligenz erachtete Cooper als unbegründet. Denn künstliche Intelligenz ergänze menschliche Intelligenz lediglich. Sie zog den Vergleich zur industriellen Revolution, seit der Maschinen die Muskelkraft des Menschen ergänzten und unterstützten. Analog werde künstliche Intelligenz den menschlichen Geist ergänzen und unterstützen. Als konkreten Anwendungsfall zeigte Cooper den Alltag eines blinden Softwareingenieurs bei Microsoft. Dieser trägt «Smart Glasses», d. h. eine Brille mit eingebauter Kamera. Die zuge­hörige Software analysiert die Auf­nahmen und beschreibt dem Träger anschliessend Umgebung, Objekte, ­Personen und Texte. Die Referentin ­demonstrierte die Software anschlies­send mit ihrem eigenen Smartphone. Sie richtete dessen Kamera aufs Publikum, worauf die Software ihr die Situation treffend als «eine Gruppe von Personen, auf Stühlen sitzend» beschrieb. Diese und andere faszinierende Anwendungen künstlicher Intelligenz seien im Wesentlichen nur möglich geworden, weil drei Faktoren in den letzten Jahren zusammengekommen seien: Big Data, d. h. die Sammlung enormer Datenmengen, die zum Training künstlicher Intelligenz eingesetzt werden können; die Verfügbarkeit der nötigen Rechnerleistung, wie sie Danani im Eingangsreferat bereits angesprochen hatte, sowie fortschrittliche Algorithmen, die aus den Daten Sinn gewinnen könnten. Mit dem ersten Punkt, der Sammlung und Strukturierung grosser Datenmengen («data mining»), sprach Cooper zum Schluss ihres Referates auch einen der potenziellen Konfliktherde an. Denn bei der Sammlung und Verarbeitung solcher Datenmen-gen, z. B. Fotos von Personen oder Gegenständen, stellten sich unweigerlich urheber- und persönlichkeitsrechtliche Fragen. Diesbezüglich stellte sie die Frage in den Raum, ob das Urheberrecht tatsächlich solche Tätigkeiten be- oder gar verhindern solle, obwohl die Verwendung der Daten nur zum Training der Software erfolge. Die Veranstaltung wurde wie üblich durch eine Podiumsdiskussion abgerundet. Nach einer kurzen Einführung durch Dr. Jürg Simon, Rechtsanwalt in Zürich, führte Oertli den regen Austausch zwischen den Referenten und dem Publikum. Oertli stellte zunächst die Frage, ob die Schweiz im Angesicht künstlicher Intelligenz eine neue Art Immaterial­güterrecht benötige. Aus dem Publikum wurde diese Frage verneint, sie tauche bei jeder neuen Technologie auf. Allerdings könne man sich fragen, ob Big Data allenfalls neue Regelungen nötig mache. Andererseits bestehe in der Schweiz im Lauterkeitsrecht mit Art. 5 lit. c UWG eine Bestimmung, die die Verwertung fremder Leistungen sanktioniere. Diese Bestimmung könnte bei entsprechender Auslegung auch in ­solchen Situationen Schutz bieten. Nordemann fand die Idee grundsätzlich interessant, weil sie denjenigen schützen würde, der in die Datensammlung investiert hat. Andere Voten aus dem Publikum zeigten sich skeptisch, dass der Weg über das UWG der richtige sei. Er gebe zwar derjenigen Person Rechte, die die Datensammlung generiert habe. Die Inhaber der Ursprungsdaten (z. B. die abgebildeten Personen etc.) erhielten dadurch jedoch keinerlei Rechte. Ferrari Hofer stellte danach zur Debatte, ob denn Daten per se einen Wert hätten. Denn eine Eigenschaft künstlicher Intelligenz sei es ja gerade, dass sie aus riesigen, für sich genommen wertlosen Datenmengen Struktur, Sinn und damit Wert generieren könne. Das bisherige System der Immaterialgüterrechte belohne ja auch diejenigen, die durch Kombination etwas von Wert (sei es eine Erfindung oder ein künst­lerisches Werk) geschaffen hätten. Aus dem Publikum wurde als Alternative weiter vorgeschlagen, gewisse Verhaltensweisen mit Schutzrechten zu belohnen, also z. B. die Selektion der Daten und das anschliessende Training der künstlichen Intelligenz. Oertli nahm diesen Aspekt auf und kam auf die von Cooper ange­sprochene Problematik zurück, nämlich die Verwendung urheberrechtlich geschützten Materials zum Training von künstlicher Intelligenz. Nordemann antwortete, das Urheberrecht sei in ­solchen Fällen klarerweise betroffen. Die Frage sei, ob allenfalls Ausnahmen bzw. Schranken greifen würden und ob es deren weitere für künstliche Intel­ligenz benötige. Cooper ergänzte, künstliche Intelligenz sei ein Werkzeug für ganz spezifische Problemlösungen. Der ­klassische Urheberrechtsschutz ­mache in solchen Situationen wenig Sinn. Daten gewönnen erst durch die Verarbeitung an Wert, ein Schutz für Daten per se würde demgegenüber die technische Innovation stark behindern. Auf Rückfrage von Danani ergänzte Nordemann, bei der Geltendmachung von Urheberrechtsverletzungen würden sich ohnehin schwierige Beweisfragen stellen, da man einer künstlichen Intelligenz üblicherweise nicht ansehe, mit welchen Daten sie trainiert worden sei. Zum Schluss der Diskussion fragte Oertli die Teilnehmenden, was der ­voraussichtlich nächste grosse Schritt auf dem Gebiet künstlicher Intelligenz sei. Danani und Cooper waren sich ­einig, dass personalisierte Medizin, d. h. das Massschneidern von Wirkstoffen und Behandlungen auf die Patienten, unser Leben positiv verändern werde. Bouchet warnte jedoch auch davor, künstlicher Intelligenz blind zu vertrauen, insb. im Hinblick auf autonome Systeme wie Fahrzeuge oder gar Waffensysteme. Zum Schluss dankte Ferrari Hofer den Referenten für die qualitativ hochstehenden Beiträge und schloss die Veranstaltung. Das Fazit der Veranstaltung dürfte sein, dass künstliche Intelligenz die Tätigkeit von Immaterialgüterrechtlern massgeblich verändern wird, sowohl was die verwendeten Werkzeuge anbelangt als auch den Inhalt der Arbeit an sich. Es hat sich aber auch gezeigt, dass sich die neuen Möglichkeiten zu gros­sen Teilen in das bestehende System der Immaterialgüterrechte einordnen lassen. Der Wandel dürfte für die Praktiker also eher Evo- statt Revolution sein. Was die rechtliche Regelung künstlicher Intelligenz bzw. deren Arbeitsergebnisse anbelangt, scheint bei den bestehenden Schutzrechten punktuell Anpassungsbedarf zu bestehen, z. B. was deren Abhängigkeit von den Leistungen einer natürlichen Person (als Schöpfer bzw. Erfinderin) anbelangt. Grösserer Handlungsbedarf besteht dagegen im Urheberrecht. Die Anwendungsbeispiele und die an­schlies­sende Podiumsdiskussion haben gezeigt, dass insb. bei der Handhabung von Datensammlungen, ohne die keine künstliche Intelligenz trainiert werden kann, eine Vielzahl an immaterialgüter-, datenschutz- und eigentumsrechtlichen Fragen zusammentreffen. Daten, die verschiedentlich schon als «Rohstoff des 21. Jahrhunderts» bezeichnet wurden, dürften also Immaterialgüterrechtler auch in diesem Kontext noch eine Weile beschäftigen.

Fabian Wigger​^*

Das Modeunternehmen «G-Star» bezichtigte die portugiesische «Cofemel», ebenfalls ein Modeunternehmen, der Urheberrechtsverletzung, weil diese Jeans-, Sweatshirt- und T-Shirt-Modelle übernommen habe. Cofemel begegnete diesem Vorwurf insbesondere mit dem Argument, die betreffenden Bekleidungsmodelle seien gar nicht urheberrechtlich geschützt​². Das von G-Star angerufene erstinstanzliche portugiesische Gericht bejahte die Urheberrechtsverletzung durch Cofemel ebenso wie später ein Berufungsgericht​³. Die Streitsache gelangte daraufhin zum Supremo Tribunal de Justiça, dem obersten Gericht Portugals. Da in der portugiesischen Lehre und Rechtsprechung ungeklärt war, unter welchen Voraussetzungen Werke der angewandten Kunst urheberrechtlich geschützt sind, stellte das Supremo Tribunal im Rahmen eines Vorabentscheidungsverfahrens dem EuGH zwei Fragen: Erstens, ob das Unionsrecht mitgliedstaatlichen Regelungen entgegenstehe, wonach Werke der angewandten Kunst dem Urheberrechtsschutz deshalb zugänglich sind, weil sie über ihren Gebrauchszweck hinaus einen «eigenen, ästhetisch markanten visuellen Effekt hervorrufen». Zweitens, ob das Unionsrecht mitgliedstaatlichen Regelungen entgegenstehe, wonach Werke der angewandten Kunst nur dann urheberrechtlich geschützt sind, wenn sie «nach einer besonders strengen Beurteilung hinsichtlich ihres künstlerischen Charakters unter Berücksichtigung der in den kulturellen und institutionellen Kreisen herrschenden Auffassung die Einordnung als ‹künstlerische Schöpfung› oder ‹Kunstwerk› verdienen»​⁴. Nachstehend werden die Ausführungen des EuGH zu den beiden Vorlagefragen vollständig und unverändert​⁵ wiedergegeben. Deren Lektüre lohnt sich insbesondere deshalb, weil sie mit einer kompakten Zusammenfassung der bisherigen Rechtsprechung zum unionsrechtlichen Werkbegriff eröffnen. 26 Mit seiner ersten Frage möchte das vorlegende Gericht wissen, ob Art. 2 lit. a RL 2001/29 dahin auszulegen ist, dass er einer nationalen Vorschrift entgegensteht, nach der Modelle wie die im Ausgangsverfahren in Rede stehenden Bekleidungsmodelle urheberrechtlich geschützt sind, weil sie über ihren Gebrauchszweck hinaus einen eigenen, ästhetisch markanten visuellen Effekt hervorrufen. 27 Nach Art. 2 lit. a RL 2001/29 sind die Mitgliedstaaten verpflichtet, für Urheber das ausschliessliche Recht vorzusehen, die Vervielfältigung ihrer Werke zu erlauben oder zu verbieten. 28 Der Begriff «Werk», auf den diese Bestimmung Bezug nimmt, findet sich auch in Art. 3 Abs. 1 und Art. 4 Abs. 1 RL 2001/29, die das ausschliessliche Recht des Urhebers eines Werks auf ­öffentliche Wiedergabe und auf Verbreitung betreffen, in Art. 5 dieser Richtlinie, der die Ausnahmen und Beschränkungen in Bezug auf diese ausschliesslichen Rechte betrifft, sowie in den Art. 6 und 7 dieser Richtlinie, die technische Massnahmen bzw. Informationsmassnahmen zum Schutz dieser ausschliesslichen Rechte betreffen. 29 Der in allen diesen Bestimmungen enthaltene Begriff «Werk» stellt, wie sich aus der ständigen Rechtsprechung des Gerichtshofs ergibt, einen autonomen Begriff des Unionsrechts dar, der einheitlich auszulegen und anzuwenden ist und zwei Tatbestandsmerkmale hat. Zum einen muss es sich bei dem betreffenden Gegenstand um ein Original in dem Sinne handeln, dass er eine eigene geistige Schöpfung seines Urhebers darstellt. Zum anderen ist die Einstufung als «Werk» Elementen vorbehalten, die eine solche Schöpfung zum Ausdruck bringen (vgl. in diesem Sinne EuGH vom 16. Juli 2009, C-5/08, Rn. 37 und 39, «Infopaq» sowie EuGH vom 13. November 2018, C-310/17, Rn. 33 und 35–37, «Levola»​⁶ sowie die dort angeführte Rechtsprechung). 30 Hinsichtlich des ersten Merkmals kann nach ständiger Rechtsprechung des Gerichtshofs ein Gegenstand erst bzw. bereits dann als Original angesehen werden, wenn er die Persönlichkeit seines Urhebers widerspiegelt, indem er dessen freie kreative Entscheidungen zum Ausdruck bringt (vgl. in diesem Sinne EuGH vom 1. Dezember 2011, C-145/10, Rn. 88 f. und 94, «Painer» sowie EuGH vom 7. August 2018, C-161/17, Rn. 14, «Renckhoff»​⁷). 31 Wurde dagegen die Schaffung ­eines Gegenstands durch technische Erwägungen, durch Regeln oder durch andere Zwänge bestimmt, die der Ausübung künstlerischer Freiheit keinen Raum gelassen haben, kann nicht davon ausgegangen werden, dass dieser Gegenstand die für die Einstufung als Werk erforderliche Originalität aufweist (vgl. in diesem Sinne EuGH vom 1. März 2012, C-604/10, Rn. 39, «Football Dataco» und die dort angeführte Rechtsprechung). 32 Hinsichtlich des zweiten in Rn. 29 des vorliegenden Urteils genannten Merkmals hat der Gerichtshof klargestellt, dass der Begriff «Werk» im Sinne der RL 2001/29 zwangsläufig einen mit hinreichender Genauigkeit und Objektivität identifizierbaren Gegenstand voraussetzt (vgl. in diesem Sinne EuGH vom 13. November 2018, C-310/17, Rn. 40, «Levola»). 33 Zum einen müssen nämlich die Behörden, die mit dem Schutz der dem Urheberrecht innewohnenden Ausschliesslichkeitsrechte betraut sind, den so geschützten Gegenstand klar und genau erkennen können. Dasselbe gilt für Dritte, gegenüber denen der Urheber dieses Gegenstands den Schutz beanspruchen kann. Zum anderen setzt das Erfordernis des Ausschlusses jedes – der Rechtssicherheit schädlichen – subjektiven Elements bei der Identifizierung des geschützten Gegenstands ­voraus, dass dieser auf objektive Weise ausgedrückt worden ist (vgl. in diesem Sinne EuGH vom 13. November 2018, C-310/17, Rn. 41, «Levola»). 34 Wie der Gerichtshof hervorge­hoben hat, entspricht es nicht dem ­Erfordernis der Genauigkeit und Objek­tivität, wenn eine Identifizierung im Wesentlichen auf naturgemäss subjektiven Empfindungen der Person beruht, die den fraglichen Gegenstand wahrnimmt (vgl. in diesem Sinne EuGH vom 13. November 2018, C-310/17, Rn. 42, «Levola»). 35 Wenn ein Gegenstand die in den Rn. 30 und 32 des vorliegenden Urteils genannten Merkmale aufweist und ­daher ein Werk ist, muss er in dieser ­Eigenschaft gemäss der RL 2001/29 urheberrechtlich geschützt werden, wobei der Umfang dieses Schutzes nicht vom Grad der schöpferischen Freiheit seines Urhebers abhängt und daher nicht geringer ist als derjenige, der allen unter die Richtlinie fallenden Werken zukommt (vgl. in diesem Sinne EuGH vom 1. Dezember 2011, C-145/10, Rn. 97–99, «Painer»). 36 In Anbetracht dieser Rechtsprechung ist – um die erste Frage des vorlegenden Gerichts beantworten zu können – in einem ersten Schritt zu bestimmen, ob Modelle überhaupt als «Werke» im Sinne der RL 2001/29 eingestuft werden können. 37 Hierzu ist zunächst festzustellen, dass nach Art. 17 Abs. 2 der Charta der Grundrechte der Europäischen Union geistiges Eigentum geschützt wird. 38 Aus dem Wortlaut dieser Bestimmung ergibt sich, dass Gegenstände, die geistiges Eigentum darstellen, nach dem Unionsrecht geschützt sind. Dies bedeutet jedoch nicht, dass solche Gegenstände oder Kategorien von Gegenständen allesamt in identischer Weise geschützt werden müssen. 39 So hat der Unionsgesetzgeber verschiedene abgeleitete Rechtsakte erlassen, durch die gewährleistet werden soll, dass das geistige Eigentum geschützt wird, und zwar insbesondere zum einen urheberrechtlich geschützte Werke im Sinne der RL 2001/29 und zum anderen Muster und Modelle, die entweder unter die RL 98/71 – die auf Muster und Modelle anwendbar ist, die in einem oder mit Wirkung für einen Mitgliedstaat eingetragen sind – oder unter die VO 6/2002 – die auf Muster und Modelle (dort Geschmacksmuster genannt) anwendbar ist, die auf Unionsebene geschützt sind – fallen. 40 Dadurch hat der Unionsgesetz­geber zum Ausdruck gebracht, dass Gegenstände, die als Muster oder ­Modell geschützt sind, grundsätzlich nicht Gegenständen gleichgesetzt werden können, die durch die RL 2001/29 geschützte Werke darstellen. 41 Diese gesetzgeberische Entscheidung steht im Einklang mit der Berner Übereinkunft, an deren Art. 1–21 sich die Union, obwohl sie nicht Vertragspartei der Berner Übereinkunft ist, nach Art. 1 Abs. 4 des WIPO-Urheberrechtsvertrags [WCT], dem sie beigetreten ist, halten muss (vgl. in diesem Sinne EuGH vom 13. November 2018, C-310/17, Rn. 38, «Levola» und die dort angeführte Rechtsprechung). 42 Art. 2 Abs. 7 RBÜ gestattet es den Vertragsparteien nämlich, gewerblichen Mustern und Modellen einen besonderen Schutz zu gewähren, der sich von dem Schutz für die unter dieses Übereinkommen fallenden Werke der Literatur und der Kunst unterscheidet und diesen gegebenenfalls ausschliesst, und die Voraussetzungen für einen solchen Schutz festzulegen. Gleichzeitig schliesst diese Bestimmung auch eine kumulative Anwendung der beiden Schutzarten nicht aus. 43 In diesem Kontext hat sich der ­Unionsgesetzgeber für ein System entschieden, bei dem der Schutz, der ­Mustern und Modellen vorbehalten ist, und jener, der durch das Urheberrecht gewährt wird, einander nicht ausschlies­sen. 44 In Bezug auf Muster und Modelle heisst es nämlich in Art. 17 Satz 1 RL 98/71, dass das nach Massgabe dieser Richtlinie durch ein in einem oder mit Wirkung für einen Mitgliedstaat eingetragenes Recht an einem Muster oder Modell geschützte Muster oder Modell auch nach dem Urheberrecht dieses Staates von dem Zeitpunkt an schutzfähig ist, an dem das Muster oder Modell geschaffen oder in irgendeiner Form festgelegt wurde. Anschliessend wird in Satz 2 dieses Artikels ausgeführt, dass der Umfang eines solchen urheberrechtlichen Schutzes und die Bedingungen für dessen Gewährung sowie die erforderliche Gestaltungshöhe von dem einzelnen Mitgliedstaat festgelegt werden. Für die auf Unionsebene geschützten Muster und Modelle enthält Art. 96 Abs. 2 VO 6/2002 eine Art. 17 RL 98/71 entsprechende Regelung. 45 Diese beiden Bestimmungen sind ihrerseits im Licht des achten Er­wägungsgrundes der RL 98/71 und des 32. Erwägungsgrundes der VO 6/2002 zu sehen, in denen ausdrücklich auf den Grundsatz des «kumulativen Schutzes» als Muster oder Modell zum einen und nach dem Urheberrecht zum anderen hingewiesen wird. 46 Was das Urheberrecht angeht, so ergibt sich aus Art. 9 RL 2001/29 («Weitere Anwendung anderer Rechtsvorschriften»), der insbesondere unter Berücksichtigung all seiner Sprach­fassungen (vgl. in diesem Sinne EuGH vom 4. Februar 2016, C-659/13 und C-34/14, Rn. 122 und die dort angeführte Rechtsprechung) und im Licht des 60. Erwägungsgrundes der Richt­linie auszulegen ist, dass die Richtlinie Vorschriften des nationalen oder des Unionsrechts in anderen Bereichen, insbesondere Rechte an Mustern und Modellen, unberührt lässt. 47 So ändert die RL 2001/29 nichts am Bestehen und am Anwendungs­bereich der geltenden Rechtsvorschriften im Bereich der Muster und Modelle, was den oben in Rn. 45 erwähnten Grundsatz des «kumulativen Schutzes» einschliesst. 48 Angesichts all dieser Rechts­vorschriften ist festzustellen, dass Modelle dann als «Werke» im Sinne der RL 2001/29 anzusehen sind, wenn sie die beiden oben in Rn. 29 genannten Voraussetzungen erfüllen. 49 Anhand dieser Voraussetzungen ist in einem zweiten Schritt zu prüfen, ob Modelle, die – wie die im Ausgangsverfahren in Rede stehenden Bekleidungsmodelle – dem vorlegenden Gericht zufolge über ihren Gebrauchszweck hinaus einen eigenen, ästhetisch markanten visuellen Effekt hervorrufen, als «Werke» einzustufen sind, wobei das vorlegende Gericht wissen möchte, ob ein solches Element der ästhetischen Originalität das grundlegende Kriterium für die Gewährung des Schutzes nach der RL 2001/29 ist. 50 Hierzu ist zunächst anzumerken, dass der Schutz von Mustern und Modellen einerseits und der urheberrechtliche Schutz andererseits grundverschiedene Ziele verfolgen und unterschiedlichen Regelungen unterliegen. Wie der Generalanwalt in den Rn. 51 und 55 seiner Schlussanträge im Wesentlichen ausgeführt hat, erfasst der Schutz von Mustern und Modellen Gegenstände, die zwar neu und individualisiert sind, aber dem Gebrauch dienen und für die Massenproduktion gedacht sind. Ausserdem ist dieser Schutz während eines Zeitraums anwendbar, der zwar begrenzt ist, aber ausreicht, um sicherzustellen, dass die für das Entwerfen und die Produktion dieser Gegenstände erforderlichen Investitionen rentabel sind, ohne jedoch den Wettbewerb übermässig einzuschränken. Demgegenüber ist der mit dem Urheberrecht verbundene Schutz, der deutlich länger dauert, Gegenständen vorbehalten, die als Werke eingestuft werden können. 51 Aus diesen Gründen darf – ent­sprechend den Ausführungen des ­Generalanwalts in Rn. 52 seiner Schlussanträge – die Gewährung urheberrechtlichen Schutzes für einen als Muster oder Modell geschützten Gegenstand nicht dazu führen, dass die Zielsetzungen und die Wirksamkeit dieser beiden Schutzarten beeinträchtigt werden. 52 Daraus folgt, dass der Schutz von Mustern und Modellen und der mit dem Urheberrecht verbundene Schutz nach dem Unionsrecht zwar kumulativ für ein und denselben Gegenstand gewährt werden können, diese Kumulierung jedoch nur in bestimmten Fällen in Frage kommt. 53 Hierzu ist zum einen festzustellen, dass die ästhetische Wirkung, die ein Modell haben kann, ausweislich der üblichen Bedeutung des Begriffs «ästhetisch» das Ergebnis einer naturgemäss subjektiven Schönheitsempfindung des jeweiligen Betrachters ist. Folglich erlaubt es diese subjektive Wirkung für sich genommen nicht, auf einen mit hinreichender Genauigkeit und Objektivität identifizierbaren Gegenstand im Sinne der oben in den Rn. 32–34 erwähnten Rechtsprechung zu schliessen. 54 Zum anderen trifft es zwar zu, dass ästhetische Erwägungen Teil der schöpferischen Tätigkeit sind. Gleichwohl ermöglicht der Umstand, dass ein Modell eine ästhetische Wirkung hat, für sich genommen nicht die Feststellung, ob es sich bei diesem Modell um eine geistige Schöpfung handelt, die die Entscheidungsfreiheit und die Persönlichkeit ihres Urhebers widerspiegelt und somit dem oben in den Rn. 30 und 31 angeführten Erfordernis der Originalität genügt. 55 Daraus folgt, dass der Umstand, dass Modelle wie die im Ausgangsverfahren in Rede stehenden Bekleidungsmodelle über ihren Gebrauchszweck hinaus einen eigenen, ästhetisch markanten visuellen Effekt hervorrufen, es nicht rechtfertigen kann, solche Modelle als «Werke» im Sinne der RL 2001/29 einzustufen. 56 Nach alledem ist auf die erste Frage zu antworten, dass Art. 2 lit. a RL 2001/29 dahin auszulegen ist, dass er einer nationalen Vorschrift entgegensteht, nach der Modelle wie die im Ausgangsverfahren in Rede stehenden Bekleidungsmodelle urheberrechtlich geschützt sind, weil sie über ihren Gebrauchszweck hinaus einen eigenen, ästhetisch markanten visuellen Effekt hervorrufen. 57 In Anbetracht der Antwort auf die erste Frage erübrigt sich die Beantwortung der zweiten Frage. […] Nachstehend werden die vier zentralen Aussagen des vorgestellten Urteils zusammengefasst und – soweit sinnvoll – der Rechtslage in der Schweiz gegenübergestellt. Bislang gab es unterschiedliche Auffassungen darüber, ob die vom EuGH vorangetriebene Harmonisierung des urheberrechtlichen Werkbegriffs auch für Werke der angewandten Kunst gelten soll. Genährt würde die Kontroverse insbesondere durch den Vorbehalt zugunsten mitgliedstaatlichen Rechts in Art. 17 Satz 2 RL 98/71​⁸. Nach dieser Bestimmung obliegt es nämlich den Mitgliedstaaten festzulegen, «[i]n welchem Umfang und unter welchen Bedingungen» Geschmacksmuster (auch) Urheberrechtsschutz erfahren. Sodann wird darin festgehalten, dass diese ­mitgliedstaatliche Regelungskompetenz auch die dafür «erforderliche ­Gestaltungshöhe» mitumfasst. Zwar liess bereits die EuGH-Entscheidung «Flos» von 2011 erahnen, dass die sich aus dem besagten geschmacksmusterrechtlichen Vorbehalt ergebende mitgliedstaatliche Regelungskompetenz ihre Grenzen dort haben dürfte, wo die Harmonisierungsvorgaben des EU-­Urheberrechts beginnen​⁹. Nichtsdestotrotz ging beispielsweise der deutsche BGH in seiner Geburtstagszugs­entscheidung von 2014 noch davon aus, dass es den Mitgliedstaaten aufgrund des zitierten Vorbehalts nicht verwehrt wäre, den Urheberrechtsschutz für Werke der angewandten Kunst vom Erreichen einer besonderen Gestaltungshöhe abhängig zu machen​¹⁰. In der hier vorgestellten Entscheidung reduziert der EuGH den Regelungsgehalt von Art. 17 RL 98/71 nun im ­Wesentlichen darauf, dass das Verhältnis zwischen dem Geschmacksmusterrecht und dem Urheberrecht dem Prinzip des «kumulativen Rechtsschutzes» folgt. Für die Aufteilung der Regelungskompetenzen innerhalb des EU-Ur­heberrechts hat diese Bestimmung für den EuGH jedoch keine Bedeutung. Somit gelten die auf der RL 2001/29 fussenden Harmonisierungsvorgaben – und damit auch der unionsrechtliche Werkbegriff – uneingeschränkt auch für den urheberrechtlichen Schutz von Werken der angewandten Kunst​¹¹. Die Frage, ob die urheberrecht­lichen Schutzvoraussetzungen bei Werken der angewandten Kunst unionsrechtlich harmonisiert sind oder nicht, ist für die Rechtslage in der Schweiz freilich nicht entscheidend. Festhalten lässt sich immerhin, dass der Grundsatz des «kumulativen Rechtsschutzes» für das Verhältnis zwischen dem Designrecht und dem Urheberrecht, ja im Immaterialgüterrecht insgesamt, heute auch hierzulande anerkannt ist​¹². Der EuGH wendet auf Werke der an­gewandten Kunst dieselben Schutzvoraussetzungen an wie für andere Werk­arten auch. Insbesondere weigert sich der Gerichtshof, die vom portugiesischen Recht für derartige Gestaltungen entwickelten besonderen Abgrenzungskriterien zu übernehmen​¹³. Ebenso hat der deutsche BGH im Geburtstagszugs-Entscheid klargestellt, es sei nicht gerechtfertigt, «[…] an den Urheberrechtsschutz von Werken der angewandten Kunst höhere Anforderungen zu stellen als an den Urheberrechtsschutz von Werken der zweckfreien Kunst»​¹⁴. Er verabschiedete sich damit von seiner früheren – im Lichte der hier vorgestellten Entscheidung ­klarerweise unionsrechtswidrigen – Praxis, gemäss der der Urheberrechtsschutz von Werken der angewandten Kunst eine «besondere Gestaltungshöhe» voraussetzte (sog. «Stufentheorie»). Gleiches dürfte auch in der Schweiz gelten, wo die Stufentheorie nie richtig Fuss gefasst hat. Entsprechend beurteilt sich die urheberrechtliche Schutzfähigkeit von Werken der angewandten Kunst nach denselben Kriterien, die auch für die anderen Werk­arten gelten​¹⁵. Bedauerlicherweise hat sich das BGer im Barhocker-Entscheid jedoch um eine entsprechend klare Aussage – die vor dem Hintergrund des Geburtstagszugs-Entscheids auf der Hand gelegen hätte – gedrückt​¹⁶. Der EuGH folgert aus dem Umstand, dass Geschmacksmuster- und Urheberrecht «grundverschiedene Ziele verfolgen und unterschiedlichen Regelungen unterliegen», dass die beiden Schutzrechte «nur in bestimmten Fällen» gleichzeitig zur Anwendung gelangen​¹⁷. Dies dann, wenn eine Gestaltung sowohl den geschmacksmusterrechtlichen als auch den – strengeren und ebenfalls voll­harmonisierten – urheberrechtlichen Schutzvoraussetzungen genügt. Der EuGH scheint mithin davon auszugehen, dass das Gros der geschmacksmusterrechtlich erfassten Gestaltungen den urheberrechtlichen Schutzvoraussetzungen nicht genügt​¹⁸. Auch das Schweizer BGer ging im Barhocker-Entschied davon aus, dass es keinen «flächendeckenden» Urheberrechtsschutz für vom Designrecht erfasste Gestaltungen gäbe. Allerdings betonte es in diesem Zusammenhang – anders als der EuGH und auch der BGH​¹⁹ – nicht die funktionalen Unterschiede zwischen dem Design- und dem Urheberrecht, sondern gerade ­deren diesbezügliche Gemeinsamkeiten («[…] Der Schutz beider Gesetze gilt der kreativen Formgebung. Aus dem grundsätzlich übereinstimmenden Schutzzweck ergibt sich, […]»)​²⁰. Die verschieden strengen Schutzvoraus­setzungen erklärt das BGer derweil mit der unterschiedlichen Reichweite des Schutzes. Diese bedinge, dass «[…] die Voraussetzungen der urheberrecht­lichen Individualität höher sein müssen als die Eigenart des Designrechts […]»​²¹. Ob das BGer mit dem EuGH auch darin übereinstimmt, dass die Schutzkumulation von Design- und Urheberrecht einen Ausnahmefall darstellt, könnte zwar prima facie aus der Zweifelsfall-Regelung, wie sie etwa im Le-Corbusier-Entscheid bemüht wird​²², abgeleitet ­werden, ist aber mit Blick auf die dem Barhocker-Entscheid zugrunde liegende Grosszügigkeit bei der Anwendung der urheberrechtlichen Schutzvoraussetzungen gleichwohl fraglich. Für den EuGH widerspiegelt die Ästhetik «das Ergebnis einer naturgemäss subjektiven Schönheitsempfindung des  jeweiligen Betrachters»​²³. Aufgrund dieser subjektiven Komponente sei das Kriterium der Ästhetik weder genau noch objektiv und daher als Element urheberrechtlicher Schutzvoraussetzungen nicht geeignet​²⁴. Diese Feststellung kontrastiert mit der jüngeren Rechtsprechung des BGH, gemäss der die Ästhetik bzw. die «ästhetische Wirkung» einer Produktge­staltung für die Begründung des Urheberrechtsschutzes eine zentrale Rolle spielt​²⁵. Dies freilich weniger als eigenständiger Massstab, sondern primär als Kriterium zur Abgrenzung vom Bereich der Technik​²⁶. Urheberrechtsschutz bedingt danach das Ausnutzen eines ästhetischen und nicht eines technischen Gestaltungsspielraums. So betrachtet, bedient sich der BGH der vom EuGH verschmähten Ästhetik, um – wie vom EuGH ge­fordert​²⁷ – auf die freien kreativen ­Gestaltungsentscheidungen zu fokussieren bzw. solche technischer Natur auszublenden. In Übereinstimmung mit den Erwägungen des EuGH nimmt derweil das Schweizer BGer im Zusammenhang mit den urheberrechtlichen Schutzvoraussetzungen keinen Bezug auf die Ästhetik, hält den «ästhetischen Wert» einer Gestaltung gar für irrelevant​²⁸. Im Zentrum der bundesgerichtlichen Argumentation steht vielmehr das Kriterium der Einmaligkeit, das voraussetzt, dass sich eine Formgebung deutlich von vorbekannten Formen abhebt​²⁹.

Report of the Swiss Group​^*

The Swiss Trademark Act (“TA”), Patent Act (“PA”), Design Act (“DA”) and Copyright Act (“CA”) all provide the IP owner with the following exclusive rights (in connection with commercial purposes):

• – Manufacturing goods – whether or not the goods are thereafter placed on the market is irrelevant​¹.
• – Offering goods – the offering can happen in the context of both public and private tenders, on flyers, classified ads and storefront windows for instance, regardless of the characterization of the underlying legal relationship (sales, loan, lease, etc.). Whether the goods are then sold (or loaned /leased) is irrelevant​².
• – Placing goods on the market.
• – Importing, exporting or transiting goods – (under the TA and DA even for private purposes) the import of goods to Switzerland may be prohibited even if the same are not intended to be placed on the Swiss market​³. Likewise, in principle, exporting goods from Switzerland may be prohibited regardless of whether these are IP protected in the country of export​⁴.
• – Storing goods – this refers to possession for the purpose of commercialization and encompasses all actors in the distribution chain, such as warehouse keepers, freight forwarders, wholesalers and retailers​⁵.

Apart from that, there are some particularities in the specific areas: Trademarks: Art. 13 TA provides that trademark owners have the exclusive right to affix the mark to goods or their packaging; it is infringing irrespective of whether or not such goods are thereafter placed on the market​⁶. By granting trademark owners the exclusive right to use a mark on business papers, in advertising or otherwise in the course of trade, the law encompasses the product-related use of a mark, e.g. as a domain name, trading name or a company name​⁷. Patents: Any commercial use of the patented invention is prohibited. Carrying in transit may, however, only be prohibited if the patent owner is entitled to prohibit importation in the country of destination (art. 8[3] PA). Further, the patent owner may prevent any third party from removing the indication that said products are protected by a patent – or subject to a patent application – from products or their packaging (art. 66[c] PA). Designs: Any use of the design for industrial purposes is prohibited (art. 9[1] DA). This can include the use of an object incorporating the protected design​⁸. Further, art. 9(1) DA explicitly mentions the possession of an infringing design for infringing purposes. Copyright: The CA lists a significant number of infringing acts that are rather copyright-specific (in particular, Arts. 9, 10, 11 and 39a CA​⁹). With respect to monetary reliefs, Swiss intellectual property laws explicitly refer to the Swiss Code of Obligations (“CO”) (art. 55[2] TA; art. 35[2] DA; art. 62[2] CA; art. 73[1] PA; art. 9[2] of the Swiss Unfair Competition Act [“UCA”]). Actions for damages are therefore governed by the general provisions of art. 41 et seq. CO. Damages are defined as the involuntary decrease in net capital (reduction in assets, increase in liabilities or loss of profit), meaning the difference between the injured party’s hypothetical economic position in the absence of the wrongful conduct and its actual position (the so-called “difference theory”)​¹⁰. The injured party is obliged to prove both the existence and the amount of the damage in a sufficiently determined and specific manner​¹¹. Only where the exact amount of the damage cannot be quantified, the court will estimate the amount at its discretion in the light of the normal course of events and the steps taken by the injured party (art. 42[2] CO). This provision is not only applicable if the numerical proof of the damage is impossible, but also if it cannot be strictly proven that damage occurred at all​¹². The injured party must, to the extent possible and reasonable, assert and prove all circumstances which indicate the occurrence of the damage and permit or facilitate its estimation​¹³. However, in practice, without any sales by the infringer, the right holder frequently cannot substantiate the incurred damage or lost profit as required by law, or will be unable to prove the necessary ”causal nexus” between the infringement and the suffered damage or lost profit​¹⁴. Given the legal uncertainties and current restrictive practice concerning remuneration claims in general, most claims for financial compensation on the grounds of IP infringement or unfair competition law are settled out of court. Further, in practice, claims for accounts for profits are usually more relevant than damage claims. Costs for pre-litigation market surveillance regarding intellectual property rights are generally not regarded as reimbursable as damages​¹⁵. However, scholars recognize the legal costs necessary for remedying the infringement (e.g., warning letters, costs with regard to an out-of-court settlement or necessary pre-trial lawyer’s fees) as a reduction in assets, which should be reimbursed​¹⁶. The Swiss Federal Court (“SFC”) does acknowledge a remuneration of damages based on a hypothetical or notional licensing fee (“license analogy method”), however, only under very restrictive conditions​¹⁷. Although recent decisions indicate that the SFC might adapt a more lenient approach towards the license analogy method based on art. 62 CO​¹⁸ in the future, substantial uncertainties remain in this regard. The dilution of (in particular, famous) trademarks constitutes a damage which should be compensated. The value of the trademark after infringement has to be compared with the value it had prior to the infringement​¹⁹. The confusion of the market may also result in damages to be compensated. In such case, the injured party must specify either the costs for rectifying the market confusion or the actual loss of the value of the trademark​²⁰. In copyright law, the proof of damage and damage calculations cause great difficulties, since often no reduction in assets or increase in liabilities occurs and mostly the only possible damage is a loss of potential royalties​²¹. Furthermore, it is conceivable that a copyright infringement may lead to an erosion in reputation or a reduction of market potential, although this would be difficult to quantify​²². In copyright law, the damages can be calculated on the basis of the remuneration tariffs customary in the industry for the use of the copyrighted work, such as tariffs of collecting societies or according to industry agreements​²³. Many tariffs of collecting societies provide for a doubling of the remuneration for uses for which the necessary authorization was not obtained​²⁴. However, the SFC rejects the doubling of a tariff of a collecting society as an inadmissible form of punitive damages​²⁵. In most of the design cases, it is difficult to precisely calculate the damage that occurred​²⁶. Scholars recognize the following types of damages: If the injured party’s prices had to be reduced because of the infringing goods or if the injured party suffered a decline in turnover, the loss of profit suffered as a result should be compensated in the form of lost profit​²⁷. Market confusion costs should also be reimbursed; this includes the costs of increased information and advertising, which are necessary to counteract the resulting uncertainty in the market​²⁸. Furthermore, the dilution or depreciation of the design – i.e., the associated reduction in the value of the design and the damage to the reputation of the design owner – caused by the infringing act should also be covered by the compensation for damages​²⁹. In determining the damage according to a license analogy, a fictitious license fee, which the infringer would have had to pay if he had concluded a lawful license agreement with the design right holder, is assumed. The amount of the fictitious license fee should be based on what would normally and reasonably have been agreed between the parties to a license agreement. The customary tariffs, other already concluded and comparable license agreements and the scope of protection of the design right should also serve as further indications for the calculation​³⁰. In a decision concerning unfair competition law, it has been recognized that an injured party may claim compensation for damages resulting from market confusion​³¹ or price erosion, and claims for lost profits may include remuneration for lost follow-up sales (e.g., wear and spare parts or consumables), add-on sales or other business opportunities. In principle, any damage unlawfully and willfully or negligently caused should be compensated (art. 41 CO). Therefore, multiple infringing acts may result in multiple damage claims. Liability law requires both a natural and an adequate causal link between the infringing act and the damage claimed. The natural causal link is given if a conduct is an indispensable condition for the damage and must be demonstrated with a predominant probability​³². An adequate causal link is established if, according to the ordinary course of things and the experiences of life, the action is capable of leading to such damage​³³. Furthermore, it has to be considered that, in Swiss liability law, the prohibition of over-compensation is recognized as a fundamental principle​³⁴. Over-compensation occurs if the same person is awarded different compensation payments for the same event during the same period of time and if the sum of the payments exceeds the actual damage​³⁵. The Swiss Group is not aware of any IP-related cases where courts have dealt with “franking” issues. However, based on these general principles, if the injured party can show that, with predominant probability and according to the ordinary course of things and the experiences of life, the second infringing act (e.g., a sale) and not the first infringing act (e.g., the manufacturing) caused a certain damage, the infringed party is entitled to sue the infringer again for compensation of damages. If such additional damage is caused by the second infringing act, there is no over-compensation of the claimant. Therefore, if damages have been paid in relation to goods that have been manufactured, those goods will not be “franked” and the infringer or any acquirer of the goods may be liable again for any damages causally and additionally occurring from the fact that the same goods are subsequently sold. However, for the claimant, it might be difficult to prove that additional market disturbance occurred through the sales of goods if the manufacture of the same goods already became publicly known causing initial market disturbance. The payment of compensation for damages does not mean that the infringing goods are no longer infringing and may be freely circulated subsequently. The subsequent circulation amounts to a fresh infringement. Yes. The Swiss damage restitution system is not balanced, especially with regard to damages for infringements of IP rights: Proving such damages requires a substantial effort resulting in high court costs and attorney fees. Several aspects are hard to prove. First, it is often difficult to prove the mere existence of damage. Further, there are no sufficiently developed concepts to quantify damages for non-sales infringements. Moreover, court practice on the necessary causal link between damage and infringement is very strict. Swiss courts only grant low compensation, which usually does not cover the actual damage resulting from the IP infringement. In the Swiss Group’s view, the courts’ practice should take into consideration the specific problematics and circumstances of infringements of IP rights, such as the difficulty to prove and quantify damage resulting from the dilution of a trademark or market confusion. From this perspective, lowering the burden of proof in quantification of damages for non-sales infringements would be worth considering. The Swiss Group supports option c): both damages and an injunction should be available for non-sales infringements. The assertion of claims for damages should not result in the goods being allowed to circulate freely, as there was no consent of the IP owner to place the product on the market. This would encourage infringers to commit non-sales infringements, as it would allow them to purchase a de facto compulsory license to freely trade IP-protected products. The Swiss Group would only consider an exception to that principle where the IP owner received full compensation as if the infringer had obtained a license. In that case, the legal interest of the IP owner in future injunctions could be questioned and it would possibly amount to an abuse of rights to continue to enforce the IP rights. However, since Swiss courts regularly do not grant full compensation (i.e., a compensation comparable to the royalty which would have been possible in license negotiations), such a situation will rarely be given. In principle, the Swiss Group is of the opinion that franking and cross-border franking in particular should not be admissible. If franking were admissible, a court awarding damages should take into consideration the compensation already awarded in other jurisdictions. The court should compare the damage regimes of the different jurisdictions, especially the reasonably achievable amount of compensation in those jurisdictions. Regarding quantification, the prohibition of overcompensation is very important. The awarded compensation should not put the IP owner in a more favorable position than would be the case had no infringement occurred. However, it should also be remembered that the IP owner should not be undercompensated, which is a risk if franking is admissible. Finally, it should be noted that the Swiss Group’s proposal of taking into consideration compensation awarded in other jurisdictions should not imply that a foreign court decision stating that a certain product is franked in that country is binding for a Swiss court when assessing whether the product is franked from the perspective of Swiss law or even lead to an exhaustion of the IP right. There are no other proposals within the scope of the Study Question. As IP rights often are dealt with in a cross-border context and IP-protected goods circulate in different jurisdictions, there is a certain need for common ground or harmonization in order to reduce efforts in cross-border law enforcement. Non-sales IP infringements are, however, of minor interest in practice. Therefore, as long as franking is not permitted, there is no need for harmonization in relation to damages for non-sales IP infringements. However, the Swiss Group takes the view that the question of whether or not franking is admissible should be harmonized. The mere manufacture of infringing products without any subsequent sale or offering for sale does not lead to any damage for the IP owner. The only hypothetical damage which could be considered is potential market confusion (namely if the public became aware of the ongoing manufacture of the products). If damage occurs from additional subsequent infringing acts, these acts should be the decisive element for the quantification of damage. The calculation should be different in the case where the infringing product is used by the manufacturer himself (e.g., where a patented machine is used to manufacture goods), as in the case where the product is subsequently sold. If the infringer uses the infringing product himself, the right to only manufacture the product may have a value. This would be best calculated according to a hypothetical license fee for the mere manufacture of the product and the benefits resulting from that. If there is a subsequent sale, the quantification of the damage should be based on the methods developed for the calculation of damages for sales. The manufacturer should be liable if there is a sufficient causal link between the manufacture and the damage occurring from the sale of the infringing product, e.g., if the subsequent sale was foreseeable. If the manufacturer and the subsequent seller cooperate – if the manufacturer acts at least as an accomplice – they should be jointly liable to the IP owner for the whole damage (see also below, Question 12). Yes, see the principles outlined in Question 9 above. Damages should be quantified according to the methods developed to quantify damages occurring from sales. The manufacturer should be liable if there is a sufficient causal link between manufacturing the goods and the damage occurring from selling or exporting the infringing goods. The quantification of damages depends on the specific circumstances of the individual case. If the infringing goods are only imported or warehoused without any subsequent sale or offering for sale, no damage occurs. If the goods are subsequently sold, damages should be quantified on the basis of the methods to quantify damage occurring from sales. The import or warehousing company should be fully liable if there is a sufficient causal link between the import or the warehousing and the damage occurring from the sale of the infringing product. With respect to joint liability/contributory infringement see also Question 12 below. If the product is eventually sold, damages should be quantified on the basis of the methods to quantify damage occurring from sales. If the product was never sold, damages should be quantified on the basis of each infringing act in the chain. As a matter of principle, under Swiss law the aggrieved party is entitled to claim the entire loss from both the direct author(s) and the contributory infringer(s), such as the instigator(s) and accomplice(s), the court then having to determine whether and to what extent they have a right of recourse against each other (art. 50 of the Swiss code of obligations; for further details, see the Report issued by the Swiss Group in 2015 on the “Liability for contributory infringement of IPRs (Q204)” AIPPI Study Question). The quantification of damages should reflect the specific circumstances of the infringement at hand including eventual material products directly obtained/produced by the manufacturing process. Furthermore, the Group is of the opinion that the infringed party should generally be allowed to quantify the damage occurred on the basis of a license analogy, even if the parties would ex ante not have agreed to such licensing agreement. Please refer to the Swiss Group’s answer to Question 13. Yes. Court B should also award damages. In doing so, court B should take into consideration damages already awarded by other courts, so that no over- or undercompensation of the IP owner occurs. Yes. The infringer should be liable again for all damages additionally caused by the subsequent sale or circulation of the product. However, observing the principle of the prohibition of overcompensation, the awarded compensation should not put the IP owner in a more favorable position than he was in the absence of any infringement. The quantification of damages for non-sales infringements is an interesting but rather exotic topic. As questions regarding this topic do not actually arise in practice, there is almost no case law. The Swiss Group did not include an in-house counsel.

Report of the Swiss Group​^*

No. Neither the Swiss Patent Act (hereinafter “PatA”) nor the Swiss Patent Ordinance (hereinafter “PatO”) make any reference to a plausibility requirement. As mentioned above, plausibility is not an explicit requirement in Switzerland. If a plausibility requirement was recognised at all, which has not been the case so far, it could become relevant as part of the inventive step (PatA 1[2]) or sufficiency/enablement analysis (PatA 50[1]). No. According to the PatA, patents for inventions are granted for new inventions applicable in industry (PatA 1[1]); anything that is obvious with regard to the state of the art is not patentable as an invention (PatA 1[2]). Furthermore, the invention must be described in the patent application in such a manner that it can be carried out by a person skilled in the art (PatA 50[1]). All of those requirements are also grounds for nullity of a patent (PatA 26[1][a] and PatA 26[1][b], respectively). As indicated above, the issue of plausibility is considered under the title of inventive step and/or sufficiency of disclosure (possibly also industrial applicability), as is the case in other jurisdictions. However, see also the answers to questions 5) and 6) below. No. Swiss courts tend to follow the case law of the European Patent Office (see FSC 137 III 170 para. 2.2.1, “Alendronsäure”). In EPO proceedings plausibility is discussed primarily in the context of the inventive step and sufficiency requirements. However, the plausibility case law of the EPO has not been adopted by Swiss judgments yet. A reference to plausibility is only made in the Patent Examination Guidelines when assessing sufficiency of disclosure (Guidelines, chapter 4.2, 28), although not necessarily in the context discussed in this Question. Such guidelines for examiners issued by the Federal Institute of Intellectual Property are not a legal act and have no legally binding effect. As stated above, there is no explicit plausibility requirement in Swiss law. The Swiss Group does not exclude that considerations that are relevant in the context of “plausibility” are also taken into account in Switzerland when assessing inventive step (Art. 1[2] PatA) or sufficiency of disclosure (Art. 50[1] PatA). Yet, so far neither Swiss case law nor doctrine has referred to a “plausibility” requirement as such or explicitly mentioned a “plausibility requirement” as part of those criteria in court judgments. As a consequence, the assumption that an implied plausibility requirement implicitly applies under Swiss law or practice seems far-fetched and construed. Thus, questions 6 to 14 do not apply. n/a. n/a. n/a. n/a. n/a. n/a. n/a. n/a. No. The Swiss Group is satisfied with the current state of the law in this respect. Yes. In Switzerland, technological progress is considered important for economic growth. The patent system provides one option to incentivize technological progress. The prevalent theories on the legitimation of patent law, i.e. the reward, incentive and contract theory refer to two justification arguments: the earnings/profit argument and the disclosure argument. Patents enable the right holder to obtain a return relative to the economic benefit of its invention and to recoup his/her investments. With the grant of the patent, the inventive technical knowledge is generally accessible and can be used as the basis for further research and development (disclosure and dissemination emanating from the patent system). Switzerland employs a first-to-file system, i.e. patent rights are granted to the applicant with the earliest valid priority date. The “first-to-file system” provides motivation to file a patent application as early as possible and thus also for a disclosure of inventive knowledge as quickly as possible – provided that a “completed” invention exists. A patent application should convince a person skilled in the art that a technical problem underlying the invention has been solved. The technical problem according to the invention must be solved in its entirety by the means included in the patent application. It only discloses a solution if the person skilled in the art can reproduce the invention according to the technical teaching with constant success and without undue burden. Therefore, the Swiss patent law aims at an early disclosure of “completed” inventions. However, it does not involve a mandatory requirement of a “best mode”. No. As mentioned under point 16), the incentive for an early disclosure does not come at the expense of the patentability requirements. An explicit, standalone plausibility requirement has not yet been recognized in Swiss law and practice (see questions 1 to 5 above). An implied plausibility requirement, if any, is applied as a way to determine whether patentability requirements such as sufficiency of disclosure and inventive step are met, and therefore does not interfere with the early disclosure incentive (which itself requires fulfilment of the patentability requirements). Please consult with relevant in-house /  industry members of your Group in responding to Part III. Yes. A harmonization of principles applicable in the context of the validity of European patents that are governed by the same EPC rules appears desirable. A wider harmonization throughout different legal systems appears to be challenging. No. It is the Swiss Group’s view that at present plausibility should not be a standalone requirement. The EPO appears to consider plausibility as an aspect of two main patentability requirements, namely sufficiency and inventive step (which are validity criteria already harmonized through the EPC for all European patents). Against this background, it does make sense to clarify and harmonize the corresponding standard under sufficiency and inventive step, yet it does not seem practicable to institute plausibility as a new standalone patentability requirement. No. In general, an invention must somehow show a “surprising effect”, otherwise it could not be considered inventive. Therefore, the existence of a surprising effect should not make an invention implausible. However, when the technical effect described in the patent application is “incredible” in that it clearly contradicts the prior art as understood by the skilled person, it should be clarified in a credible way why and how the invention achieves the claimed effect irrespective of the common perception in the prior art. Pursuant to established case law and literature, patentability criteria are assessed from the viewpoint of a person of ordinary skill in the art. This applies independently of questions related to plausibility/credibility. There is no different standard applicable specifically to plausibility/credibility questions. Not necessarily, but protection (in an appropriate generalized scope) is only awarded to those promises, for which it has been (credibly) shown that they are achievable. In principle, yes. A patent application should either make plausible how a certain technical effect is achieved or show that this technical effect can be achieved. A speculative filing arises from an insufficient disclosure or lack of sufficient evidence. A non-speculative filing is achieved if at the time of filing the technical effect is made plausible through sufficient evidence, whether in the form of technical data, based on prior art or common general knowledge or by being self-evident from the patent application. Whether the invention is obvious is determined in comparison to the state of the art, including the common general knowledge. A technical effect has to be made plausible by the application as filed in order to be considered in the assessment of inventive step. How the technical effect has been shown to be plausible, namely whether it is implied or self-evident from the specification, is not the relevant test for the inventive step. No. Plausibility should not be understood as a specific prohibition of prophetic examples, but rather as a standard in the assessment of enablement. The number and nature of examples, whether prophetic or not, is not determinative in the assessment of patentability. The question is rather, whether the application as a whole contains sufficient credible and enabling support to enable the skilled person to work the invention without undue burden. Thus, while the inclusion of prophetic examples (only) may increase the hurdle for a showing of enablement (in particular for unpredictable technologies), they do not necessarily deny plausibility as such. See question 22 above. No. See question 22 above. Non-prophetic examples can facilitate a showing of enablement but are not a prerequisite. The relevant point in time should be the effective date of the application (i.e. the priority date or the filing date if no priority is claimed). A technical effect that appears plausible at the effective date, but which later proves to be wrong, may be used as a ground for refusal of the application or revocation of the patent. Conversely, if the application has already been finally refused, there would be no remedy. No. The rationale underlying the plausibility test should be the same for all types of claims. Rather it is the extent of evidence necessary to show plausibility, which may change with the nature and predictability of the technology. The burden of proof resides with the applicant in case of a pending application (or the third party in case of a third-party observation). The burden of proof resides with the opponent in case of a granted patent. In inter partes proceedings, the burden of proof may be shifted according to standard practice at the EPO. It should be noted that a first-to-file system (like that in Switzerland) creates a greater urgency for the applicant to file a patent in the early stages of the invention than a first-to-invent system, which allows more time to carry out and collect evidence to show a technical effect. Thus, it may be appropriate to have a lower standard for the examination of plausibility in countries with a first-to-file system. n/a.

For the purposes of this Study Question, a work created in whole or in part using artificial intelligence is referred to as an artificially-generated work or an AI-created work. The term “copyright” means the rights associated with copyright as set forth in the Berne Convention. The term “related rights” means all other copyright-type rights, e.g., “related rights”, “neighbouring rights”, “sui generis rights”, etc. The term “economic rights” means the exclusive rights of copyright granted to the author, e.g., the right of reproduction. The term “moral rights” means the rights of copyright granted to the author apart from economic rights, e.g., the right to object to distortion of the work. Of all the technological advances that attract lawyers’ attention, artificial intelligence (AI) stands as a good chance as any of proving to be genuinely transformational. There is already a lively debate about whether the advent of AI challenges the fundamental assumptions, structures and concepts of copyright law, or whether current laws will suffice as long as its practitioners understand how the technology works. In the fields of creations likely to be protectable by copyright, AI has already made it possible to create artificially-generated works in a wide variety of areas: poems, literary works, novels, news articles, music, paintings and other artworks, etc. Human intervention in the process of creating an artificially-generated work may occur in a number of different ways or, potentially, may not occur at all. An inquiry into artificially-generated works must start with an understanding of what AI is. However, AI is a broad and rapidly evolving field that defies simple definition. Often, AI is characterized generally as the ability of a computer to mimic human intelligence, such as the ability to reason and to learn from past experience. Recognizing, however, that there is no universally applicable definition of AI, for the purposes of this Study Question and in order to have a concrete frame of reference for analysis of the copyright issues, we adopt the following definition: Artificial intelligence is an entity (or collective set of cooperative entities), able to receive inputs from the environment, interpret and learn from such inputs, e.g., in view of achieving a particular goal or objective. Today, artificially-generated works are generally not created in the absence of human intervention: the creative process often involves several human individuals. The role of the human can take many forms. To start with, programmers develop AI algorithms or “entities” that can receive certain types of inputs and interpret and learn from those inputs to achieve a desired output or goal. Then, humans (e.g., users) are typically involved to determine datasets, e.g., paintings, literary works, music, etc., that are fed as input to such entities, and to set parameters of the AI entities. Moreover, humans may ultimately select one or more works from a large number of works created using such entities, e.g., based on human tastes, preferences, and opinion. To provide a concrete basis for analysis of this Study Question, the following Working Example is adopted.

Step 1: One or more AI entities are created that are able to receive inputs from the environment, interpret and learn from such inputs, and exhibit related and flexible behaviours and actions that help the entity achieve a particular goal or objective over a period of time. The particular goal or objective to be achieved is selected by a human and, for purposes of this Study Question, involves generations of works of a type that would normally be afforded copyright protection. Step 2: Data are selected to be input to the one or more AI entities. The data may be prior works, such as artwork, music or literature as in the examples above. The data also may be inputs from sensors or video cameras or input from other sources, such as the internet, based on certain selection criteria.

Step 3: The selected data is input to the one or more AI entities, which achieve the particular goal or objective over time by generating “new works” that are not identical to any prior work.

The answers to the following questions are often based on this working example. No. No. In Switzerland, copyright protection arises automatically upon creation of a work, regardless of any formality (art. 5 par. 2 Berne Convention), provided that the requirements of protection set forth in art. 2 of the Federal Act on Copyright and Related Rights (Copyright Act, hereafter CopA) of 9 October 1992 (Status as of 1 January 2017) are met. I.e., the work is automatically protected by copyright if and from the moment it qualifies as a literary or an artistic intellectual creation with an individual character, irrespective of its value or purpose. There are no other requirements for copyright protection. In particular, it is not required that the author(s) of the work are identified. Works created by anonymous or unknown authors also enjoy copyright protection under Swiss law. Moreover, Swiss law provides, for instance, that the editor of a work or the person who has published it may exercise the copyright in the work when the true identity of the author(s) of this work remains unknown (art. 8 par. 2 CopA). Yes. Swiss law defines the requirements for a work to qualify as copyrightable work in art. 2 par. 1 CopA: A work must be an “intellectual creation” (“geistige Schöpfung” in the German wording of the CopA or “creation de l’esprit” in the French wording) to be eligible for copyright protection. This implies a human origin of the work. Art. 6 CopA consistently defines an author as “the natural person who has created the work”. Accordingly, a work has to be a creation of a human mind and based on the human will. In other words, a work must be an expression of a statement of thoughts (BGE 130 III 168 E. 4.5). Thus, a “work” in the meaning of copyright law can only be a work that is created by a human being (see also R. Von Büren/M. A. Meer, in: R. von Büren/​L. David (eds.), Schweizerisches Immaterialgüter- und Wettbewerbsrecht, II/1, Urheberrecht und verwandte Schutzrechte, 3rd edn., Basel 2014, N 165). A legal person does not have the capacity to create such an intellectual creation. Only natural persons can create works eligible for copyright protection in Switzerland. As a result, an AI-generated work will only be eligible for copyright protection if it sufficiently involves a human being in its creation process. This is in line with the established practice regarding copyright protection of photographs, whereby photographs only benefit from copyright protection if they have a human author. This precludes the eligibility of photographs obtained by automatic processes, such as radar photos or surveillance cameras (FF 2018 559, 29). More generally, the theoretical justification of copyright protection in Switzerland (as in other civil law countries) makes clear that a human contribution is a prerequisite to copyright protection. Copyrights are justified by the importance of the author’s personality and also by the ownership of the fruits stemming from a human being’s work. Yes, possibly. This, however, will only be the case if the intellectual creation of any of these persons – which is required for the new work to be eligible for copyright protection – is reflected in the AI-generated work. In particular, creative decisions of a human must find their expression in the resulting work. Otherwise, the resulting work will not be eligible for copyright protection and the person(s) involved in the Working Example will not qualify as author(s) of the new work. In that sense, a “creative causal link” must be perceptible between the creative work of said natural person(s) and the resulting work. Thus, the authors, if any, of the new work can only be humans who have provided creative inputs that are linked to and reflected in the final work. The occurrence and extent of human intervention (beyond the mere design/programming of the AI entity) remains decisive in appreciating the authorship. Whether or not this is the case has to be assessed on a case-by-case basis. As far as intellectual property (IP) protection is concerned, AI (and, in particular, machine learning) can be compared to various other computerized techniques that have emerged in the last decades, such as computer-aided design, graphics and image editors, and digital audio workstations. Such techniques have already raised questions as to the eligibility of works generated therewith and the qualification of authors of such works. In particular, the authors of software implementing such techniques are typically not regarded as authors of works created by persons using such software. Given that AI favours ever less tangible human involvement in the creative process (just like other computerized techniques), one may expect the threshold of human involvement required for such works to be eligible to change further in the future. There is no general rule under Swiss law according to which the authors (call them the “AI authors”) of the program or code (the “AI code”) defining the AI entities qualify as authors of the resulting work (i.e., work subsequently generated with this program or code). Now, without the AI authors’ creative decisions reflecting in the resulting work, there is no reason for the AI authors to qualify as authors of the resulting work (see, e.g., von Büren /Meer, SIWR II/1, 3rd edn., 2014, N 297 et seqq.). Similarly, the authors of a digital audio workstation (a software) are normally not regarded as the authors of any musical work created with this workstation. Note, the distinction between the AI code and the resulting work (the new work) matters, also when the new work happens to be a computer program code of its own. That the AI code per se may possibly benefit from copyright protection (art. 2 par. 3 CopA) does not imply that the AI authors will automatically be acknowledged as the authors of the new work, even if the latter is a computer program code. Rather, also in this context, a prerequisite for the AI authors to be qualified as authors of the new work would be the existence of a creative causal link, as discussed earlier. Whether or not this will be the case depends on the particulars of the situation. Two important criteria will presumably be: (i) the extent to which the definition (the design, programming, etc.) of the AI code by the AI authors predetermines what the new work will be; (ii) the possible further involvement of the AI authors in the creative process leading to the new work, beyond the mere definition of the AI code (e.g., as users of the AI entity). In some circumstances, the AI authors may happen to be actively involved in the production of the new work, also as users of the AI entity. E.g., the AI authors could be led to make creative decisions in the process of the creation of the new work through the use of an AI entity deriving from their own AI code. In such a case, the AI authors may possibly qualify as authors of the new work, but this would rather stem from their contribution to the creation process rather than to the AI code per se. Still, experience with previous computerized techniques suggests that AI authors will, in practice, rarely qualify as authors of the new work. There is no general rule under Swiss law according to which a human who defines a goal or objective to be achieved by the AI entities does, as such, automatically qualify as an author of the new work. Thus, this human will only be regarded as an author of the resulting work if the prerequisites set out above are fulfilled. Accordingly, a human defining such a goal or objective may be considered an author of the new work provided that such definition is adequately reflected in the resulting work. In addition, this definition must involve some creative, human contribution for the resulting work to qualify as an intellectual creation and therefore as copyrightable work. Again, this must be approached on a case-by-case basis. There is no general rule under Swiss law according to which a human who selects such inputs does, as such, automatically qualify as an author of the new work. And again, the same prerequisites as set out above would need to be fulfilled for this human to be considered an author of the resulting work. The influence of the human contribution on the new work is key, i.e., the selection performed must have a sufficient impact on the resulting work and, as such, be reflected in the resulting work. Further, this selection must involve some creative contribution from this human for the resulting work to qualify as an intellectual creation. Whether or not this will be the case depends on the specific situation. In all likelihood, no. A human selecting a particular work from multiple works generated by the AI entities will likely not be regarded as fulfilling the above prerequisites, because such a selection will normally not involve a sufficient creative activity. However, the occurrence of such a human selection shall not systematically preclude rights to authorship, as for instance discussed by Fenna Hornman (“A robot’s right to copyright”, University of Tilburg, 2018). In particular, a process in which a human author sets important conditions for the AI entity to generate various works and makes a final selection may possibly involve sufficiently free and creative choices for the resulting work to carry the personal imprint of that human author. No additional actors need to be specifically considered at this point. As noted earlier, any other natural person must, irrespective of her/his role, have provided creative inputs that are sufficiently linked to and reflected in the new work (the resulting work), in order to qualify as an author of this work. This depends on the extent of human intervention in the creative process and the resulting work itself, as discussed earlier. According to art. 2 par. 1 CopA, works are protected by copyright if they are intellectual creations with an individual character. Originality in the sense of a personal imprint of the author is not explicitly required under the Swiss law. Still, despite the explicit wording of the law and a concurring statement from the Federal Supreme Court (BGE 143 III 373 E. 2b), originality is essentially perceived as a synonym of individuality both in the jurisprudence and in the literature (see, e.g., BGE 125 III 328 E. 4b). One notes, in that respect, that the individual character is, like the originality of a work, usually regarded as an attribute of an intellectual creation of a human being. Thus, only a human work can possibly have the individual character required by the law. Accordingly, a “fully artificial work” would not be regarded as having individual character. A fully artificial work is defined here as a work automatically generated, without any creative human input beyond the definition (design/coding) of the AI code, assuming that this definition is not sufficiently linked to the original expression of the subsequently generated work. Still, some human contribution will typically occur in practice, which may be sufficient for a work to be eligible for copyright protection, without it being necessary for this contribution to be reflected in all parts (stages) of the work (see, e.g., cf. von Büren/Meer, SIWR II/1, 3rd edn., 2014, N 179 with reference to BGE 134 III 171). Aside from the required human contribution to the creative process, qualities of the resulting works will have to be assessed. Additional considerations may accordingly come into play, such as the aesthetic appeal of the resulting work, which might contribute to its originality (or individual character). A parallel can for instance be made with photographs, for which the threshold for copyright protection is essentially defined by two decisions of the Swiss Federal Supreme Court, i.e., the “Bob Marley” case (BGE 130 III 168, 2003) and the “Meili” case (BGE 130 III 175, 2004). Owing to the various possible qualities of AI-generated works and extent of human intervention to achieve them, the eligibility of such works will clearly have to be assessed on a case-by-case basis. The requirements for a work to be protected by copyright under Swiss law are discussed above. As discussed earlier, this depends on the extent of human intervention in the various stages of the creation process and, in particular, whether a sufficient causal link exists between the definition (design, coding, etc.) of the AI code underlying the AI entity and the new work eventually obtained. Thus, many scenarios can be contemplated, ranging from no copyright owner at all to co-authorship between various (human) actors, these possibly including the AI authors. Assuming a sufficient causal link exists between the definition of the AI code and the AI-generated work, the first owner may possibly (though unlikely) be or include the AI authors, as discussed earlier. Rather, the persons involved in the subsequent creative stages would more likely qualify as first owner of this work, just as with works created using other computerized techniques (image editors, digital audio, etc.). No. Under Swiss law, only human beings (“natural persons”) and legal entities (“legal persons”) may hold copyrights. There are no specific rules pertaining to artificially-generated works under Swiss law. Thus, the same legal provisions as applicable to usal works would also apply in this case. Art. 29 CopA provides that the term of protection of an artificially-generated work ends either 50 or 70 years after the death of the author. The artificially-generated work may happen to be a computer program of its own, which could then be eligible for copyright protection, according to art. 2 par. 3 CopA, subject to the previous remarks in terms of required human intervention. Since for software, the protection ends 50 years after the death of the author, the limit of 50 years would logically apply in that case. But for other types of works (e.g., pictures, songs), the limit of 70 years would remain valid, keeping in mind the distinction between the AI code and the resulting work. Where the author is unknown, protection expires 70 years after the (final instalment of the) work has been published. For the purpose of this question, we note that the process of the Working Example allows some human intervention. This process can thus lead to copyrightable material. We further consider that such material may be used in a subsequent process, which also may be according to the Working Example. I.e., the subsequent process may, for example, serve a performance involving a human being. If the first process does not involve any human intervention, its output cannot enjoy copyright protection, which precludes the possibility for related rights according to Swiss law. However, if the first process involves a human contribution that leads to copyrightable material, this material may then be subject to related rights. In that respect, performers’ rights as set forth in art. 33 et seqq. CopA would only apply to the extent that they originate in an act of a performing artist, which by definition, must be a human being (cf., for example, Barrelet/Egloff, CRA 33 N 6). For example, a technical process that intonates a pre-existing musical work does not qualify as a performer and, therefore, cannot enjoy protection under art. 33 et seqq. CopA. On the contrary, a live performance involving a sufficient contribution from a performing artist may engender performers’ rights. Whether or not the action of a human being who merely initiates a technical process (e.g., involving an AI entity) and/or selects pre-existing musical works as inputs to such a process qualifies as an act of a performing artist must be examined on a case-by-case basis. Next, producers’ and broadcasters’ rights as set forth in art. 36 et seq. CopA would not apply if they do not attach to the result of a creation process but to the result of an organisational/economical endeavour (Hilty, para. 363). This does not exclude, however, that producers and broadcasters, in the context of their organisational/economical endeavours, rely on the assistance of AI processes, such as the ones underlying the Working Example. Note, under Swiss law, the term of protection of related rights is 50 years (art. 39[1] CopA). As said, this group does not regard AI as fundamentally different from other computerized techniques, as far as IP is concerned. The existing copyright law has proven to be sufficiently flexible and technology-neutral to reasonably address previous technological evolutions. Hence, currently this group does not believe that the advent of AI requires changes to the existing copyright law. Irrespective of the sophistication of AI entities, such entities remain tools (like paint brushes), which may be used to produce copyrightable material when creatively leveraged by a human person. Thus, assuming that the current law or practice can adequately cope with works obtained via previous computerized techniques, the present group does not see a need for improvement for what specifically concerns artificially-generated works. Not to the knowledge of this working group. Harmonization is often desirable. Massive amounts of data (including works) are and will continue to be produced via computerized systems in the context of, e.g., social media, the Internet of things, etc. Thus, the problem posed by works created by or via AI entities must be more generally assessed in the context of the post-internet era, where digital works can instantly travel through networks of interconnected, computerized devices. In this context, copyrights and related rights need more than ever to be clearly identifiable, hence there is a need for harmonization. However, this problem is not specifically tied to AI-generated works but more generally concerns all digital works. Thus, the present group does not see a need to harmonize copyright protection and/or related rights protection specifically for artificially-generated works. Note, this working group does not include any in-house counsel. However, most of the participants advise companies active in IT and AI and/or have some experience as in-house counsel in a company active in IT and, in particular, in AI. In-house/industry members were consulted during the course of this Study Question.