By Pratyush Nath Upreti
On 15th January 2020, the first phase of the ‘Economic and Trade Agreement’ [here after trade deal] between the United States (US) and China was released. The draft came in light of recent proxy trade wars between two countries where both the countries imposed tariffs and retaliatory tariffs among each other’s. The general outline of phase one of the trade deal was conceptualized in an 86 page long ‘Fact Sheet’ released in December 2019. This note provides a succinct overview of some key provisions of the IP Chapter.
IP Chapter is the first chapter in the trade deal. In general, the Chapter is divided into eleven sections that cover areas related to trade secrets, patents, IP enforcement, e-commerce, geographical indications, trademarks, copyright, and related rights among others.
The section on ‘Trade Secrets and Confidential Business Information’ incorporates a broad definition of trade secrets. It covers ‘concerns or relates’ to trade secrets and almost any other information of commercial value, which disclosure can have the effect to cause ‘substantial harm’ to the ‘competitive position’ of the complainant. In addition, there is a key provision on the burden of proof. The relevant provision emphasizes that if trade secrets holder provides prima facie evidence, including circumstantial evidence of a reasonable indication of trade secret misappropriation, the burden of proof will shift to the accused party in civil proceeding (Article 1.5). Additionally, there should be no requirement to establish actual losses as a prerequisite to initiating a criminal investigation (Article 1.6). Another noteworthy section is on ‘Pharmaceutical-Related Intellectual Property’, that obliges China to create a system of ‘pre-market notification’ where it requires to provide notice to the patent holder, licensee or holder of marketing approval about people seeking to market generic version during the term of applicable patent, approved product or method of use for seeking approval (Article 1.11).
Besides, the IP Chapter commits China to allow disagreement to the US or other trading partners on China’s list of geographical indications agreed in the agreement with another trading partner (Article 1.15). The relevant section obliges China to take measures to reduce online infringement including ‘notice and takedown’. China commits to requiring ‘expeditious takedowns’, ‘eliminate liability for takedown notices submitted in good faith’, among others (Article 1.13).
To conclude, IP chapter departs from the recent US-led trade negotiations in Mexico and Canada (USMCA) and the earlier version of the Trans-Pacific Partnership Agreement. The general reading of the IP chapter gives the impression that the ultimate aim of the US was to bring structural reforms in China’s IP regime through transplanting the US IP norms.
 See Pratyush Nath Upreti and María Vásquez Callo-Müller, ‘Phase One US-China Trade Deal: What Does It Mean for Intellectual Property?’ (2020) 4 GRUR International: Journal of European and International IP Law (forthcoming).
By Mauritz Kop
Data sharing or rather the ability to analyse and process high quality training datasets (corpora) to teach an Artificial Intelligence (AI) model to learn, is a prerequisite for a successful Transatlantic AI ecosystem. But what about intellectual property (IP) and data protection?
In our turbulent technological era, tangible information carriers such as paper and storage media are declining in importance. Information is no longer tied to a continent, state or place. Information technology such as AI is developing at such a rapid, exponential pace that the legal problems that arise from it are to a large extent unpredictable.
- Legal dimensions of data
Data, or information, has a large number of legal dimensions. Data sharing is associated with IP law (right to prohibit and reimburse), fundamental rights (privacy, data protection, freedom of expression and other constitutional rights), fiscal law (taxation), contract law and international commercial law (e-commerce, trade treaties, anti-trust law, consumer protection). In addition, the handling of personal data has ethical, social and techno-philosophical facets.
Legal ownership of data does not exist
In most European countries, the law of property is a closed system. This means that the number of proprietary rights in rem, which are rights enforceable against everyone, are limited by law. Legal ownership of data therefore does not yet exist. From a property law point of view, data cannot be classified as ‘’res’’, as an intangible good or as a thing in which property rights can be vested. Data does have proprietary rights aspects and represents value.
Data that represent IP subject matter
Data that represent IP subject matter are protected by IP rights. Data that embody original literary or artistic works are protected by copyright. New, non-obvious and useful inventions represented by data are protected by patents. Data that epitomize independently created new and original industrial designs are safeguarded by design rights. Confidential data that have business or technological value are protected by trade secret rights.
Sui generis database rights
Hand-labelled, annotated machine learning training datasets are awarded with either a database right or a sui generis database right in Europe. Although the 1996 Database Directive was not developed with the data-driven economy in mind, there has been a general tendency of extensive interpretation in favor of database protection. A database right can be qualified as either a neighboring (ancillary or related) right (however shorter in duration i.e. 15 years), or a true sui generis IP right, but not as a full copyright. A sui generis database right is an IP right with characteristics of a property right, and is awarded after a substantial investment in creating and structuring the database, be it money or time, has been made. Businesses usually consider hand-labelled, tagged training corpora to be an asset that they can license or sell to another company. This applies to the AI system’s output data as well. As all IP rights, (sui generis) database rights are subject to exhaustion. In the USA, no sui generis database right exists on augmented input or output data. What Europe and the USA do have in common, is that any existing IP rights on input data need to be cleared before processing.
Feeding training data to the machine qualifies as a reproduction of works, and requires a license. The training corpus usually consists of copyrighted images, videos, audio, or text. If the training corpus contains non-public domain (copyrighted) works or information protected by database rights -and no text and datamining (TDM) exception applies- ex ante permission to use and process must be obtained from the rightsholders (for both scientific, commercial and non-commercial training purposes).
Clearance of machine learning training datasets
Unlicensed (or uncleared) use of machine learning input data potentially results in an avalanche of copyright (reproduction right) and database right (extraction right) infringements. Some content owners will have an incentive to prohibit or monetize data mining. Three solutions that address the input (training) data copyright clearance problem and create breathing room for AI developers, are the implementation of a broadly scoped, mandatory TDM exception (or even a right to machine legibility) covering all types of data (including news media) in Europe, the Fair Learning principle in the USA and the establishment of an online clearinghouse for machine learning training datasets. Each solution promotes the urgently needed freedom to operate and removes roadblocks for accelerated AI-infused innovation.
The TDM exceptions where originally not created with machine learning training datasets in mind. Prominent scholars advocating the introduction of robust TDM provisions to make Europe fit for the digital age and more competitive vis-a-vis the United States and China are Bernt Hugenholtz and Christophe Geiger. The ‘Joint Comment to WIPO on Copyright and Artificial Intelligence’ addresses –inter alia– challenges related to machine learning and the much needed freedom to use training corpora. This ‘amicus brief’ discusses solutions such as individual and collective TDM licenses/exceptions, whether for commercial or scientific objectives.
On the other side of the Ocean, Mark Lemley and Bryan Casey introduced the concept of Fair Learning. The authors contend that AI systems should generally be able to use databases for training whether or not the contents of that database are copyrighted. Permitting copying of works for non-expressive purposes will be -in most cases- a properly balanced, elegant policy-option to remove IP obstacles for training machine learning models and is in line with the idea/expression dichotomy.
A third solution could be the establishment of an online clearinghouse for machine learning training datasets. An ex ante or ex post one-stop-shop resembling a collective rights society, however on the basis of a sui generis compulsory licensing system. A framework that would include a right of remuneration for rights holders, but without the right to prohibit data usage for commercial and scientific machine learning purposes. With a focus on permitted, free flow of interoperable data.
Public versus private data
Another legal dimension that we can distinguish is on the one hand public (in the hands of the government) machine generated (non) personal data, and private (in the hands of the business community) machine generated (non) personal data. By machine generated data, we mean in particular information and data that are continuously generated by edge devices in the Internet of Things (IoT). These edge devices are connected via edge (or fod) nodes (transmitters) to data centers that together with edge servers form the cloud. This architecture is known as edge computing.
Mandatory TDM exceptions are a sine qua non for machine learning in Europe. A right of fair, remunerated text and data use to train an AI system needs to be mandatory and without opt outs. Would a broadly scoped TDM exception be an optional limitation, with room for Member States to implement their own rules, the Digital Single Market will become fragmented instead of harmonized. A right to machine legibility that drastically improves access to data, will greatly benefit the growth of the European AI-ecosystem.
Besides implementing broader scoped TDM exceptions, it is opportune that the EU Database Directive 96/9/EC shall be reformed by the EU Commission to prevent that data generated by connected edge devices qualifies for sui generis database right protection. Edge computing data must not be monopolized.
- Technical dimensions of data in machine learning
Most AI models need centralized data. In the current, dynamic field of machine learning, hand-labelled training datasets are a sine qua non for supervised machine learning, which uses regression and classification techniques to solve its prediction and optimization problems. This process mimics biological cognition. In contrast, unsupervised machine learning, which utilizes association and clustering (pattern recognition) techniques, uses unlabelled (unstructured) datasets as an input to train its algorithms to discover valuable regularities in digital information. Semi-supervised learning employs a combination of structured and unstructured training datasets to feed our thinking machines.
Data in machine learning can be discrete or continuous, numerical and categorical. AI systems that utilize deep learning techniques for predictive analysis and optimization, contain deep layers of artificial neural networks, with representation learning. Artificial deep neural networks (ANN’s and DNN’s) rudimentarily mimic the architecture of human biological brains and are comprised of simplified, artificial neuron layers. Anno 2020 DNN’s do not yet have axon’s, soma, dendrites, neurotransmitters, plasticity, cerebral cortices and synaptic cores. In the field of AI, data mining, statistics, engineering and neuroscience converge.
Deep reinforcement learning
Reinforcement learning does not require existing input datasets. Instead, the model learns from data from simulations and games using a reward system based on continuous feedback. Deep reinforcement learning systems, such as AlphaGo, are not easy to train. Too many correlations in the data interfere with its goal-oriented algorithms’ stable learning process. Inference applies the capabilities of a pre-trained deep learning system to new datasets, to predict its output in the form of new, useful real-world values and information.
Transfer learning is a machine learning method that seeks to apply a certain solution model for a particular problem to another, different problem. Applying a pre-trained model to new (and smaller) datasets can turn a one trick pony into the ultimate synthetic multitasker.
Evolutionary computing uses genetic optimization algorithms inspired by neo-Darwinian evolution theory. Genetic algorithms can be used standalone, or to train ANN’s and DNN’s and to identify suitable training corpora.
The approaches described above are all centralized machine learning techniques. Federated learning, in contrast, trains algorithms that are distributed over multiple decentralized edge devices in the Internet of Things. These mobile devices -such as your smartphone- contain local data samples, without exchanging their data samples. The interconnected IoT devices collaboratively train a model under a central server. Federated Learning is a scalable, distributed machine learning approach which enables model training on a large corpus of decentralized data. ‘’Federated learning embodies the principles of focused data collection and minimization, and can mitigate many of the systemic privacy risks and costs resulting from traditional, centralized machine learning and data science approaches.’’ It brings the code to the data, instead of bringing the data to the code. In other words, there is no need for sharing data.
- Data: contracts, property law and trade secrets
IP on training data and data management systems is subject to both property law aspects and proprietary rights in rem that are enforceable against everyone. Data is not a purely immaterial, non-physical object in the legal (not the natural-scientific) meaning of the word. However, if a party to a dataset transaction has acquired a contractual claim right in exchange for material benefits provided by him, there is a proprietary right. This proprietary right in rem is subject to transfer, license and delivery.
The attitude of the parties, and their legal consequence-oriented behaviour when concluding contracts about datasets and their proprietary aspects may perhaps prevail over the absence of a clear legal qualification of data (or information) in the law. In this case, party intentions go beyond the legal void. In other words, legislative gaps can be remedied by contracts.
Legal ownership, or property, is different from an IP right. IP is a proprietary right in rem. An IP right can entail a right to use data, in the form of a license.
Extra layers of rights will not bring more innovation
Raw non personal machine generated data are not protected by IP rights. Introducing an absolute data property right or a (neighboring) data producer right for augmented machine learning training datasets, or other classes of data, is not opportune. Economic literature has made clear that there are no convincing economic, or innovation policy arguments for the introduction of a new layer of rights, especially due to the absence of an incentive and reward problem for the production and analysis of datasets.
Moreover, additional exclusive rights will not automatically bring more innovation. Instead, it will result in overlapping IP rights and database right thickets. The introduction of a sui generis system of protection for AI-generated Creations & Inventions is -in most industrial sectors- not necessary since machines do not need incentives to create or invent. Where incentives are needed, IP alternatives exist. Finally, there are sufficient IP instruments to protect the various components of the AI systems that process data, create and invent. Because of theoretical cumulation of copyrights, patents, trade secrets and database rights, protection overlaps may even exist.
Public Property from the Machine
Non-personal data that is autonomously generated by an AI system and where upstream and downstream no significant human contribution is made to its creation, should fall into the public domain. It should be open data, excluded from protection by the Database Directive, the Copyright Directive and the Trade Secrets Directive.
These open, public domain datasets can then be shared freely without having to pay compensation and without the need for a license. No monopoly can be established on this specific type of database. I would like to call these AI Creations “Res Publicae ex Machina”  (Public Property from the Machine). Their classification can be clarified by means of an official public domain status stamp or marking (PD Mark status). Freedom of expression and information are core democratic values that -together with proportionality- should be internalized in our IP framework. Reconceptualizing and strengthening the public domain paradigm within the context of AI, data and IP is an important area for future research.
Data as trade secret
In practise however, to safeguard investments and monetize AI applications, companies will try hard either to keep the data a trade secret or to protect the overall database, whether it was hand-coded or machine generated. From an AI perspective, the various strategies to maximize the quality and value of a company’s IP portfolio can differ for database rights, patents and trade secrets on the input and output of an AI system. Moreover, this strategy can differ per sector and industry (e.g. software, energy, art, finance, defence).
As legal uncertainty about the patentability of AI systems is causing a shift towards trade secrets, legal uncertainty about the protection and exclusive use of machine generated databases is causing a similar shift towards trade secrets. Although it is not written with the data driven economy in mind, the large scope of the definition of a trade secret in the EU means that derived and inferred data can in theory be classified under the Trade Secrets Directive. This general shift towards trade secrets to keep competitive advantages results in a disincentive to disclose information and impedes on data sharing.
In an era of exponential innovation, it is urgent and opportune that both the Trade Secrets Directive, the Copyright Directive and the Database Directive shall be reformed by the EU legislature with the data-driven economy in mind.
- EU open data sharing initiatives
Data can be shared between Government, Businesses, Institutions and Consumers. Within an industry sector or cross-sectoral.
Important European initiatives in the field of open data and data sharing are: the Support Centre for Data Sharing (focused on data sharing practices), the European Data Portal (EDP, data pooling per industry i.e. sharing open datasets from the public sector, the Open Data Europe Portal (ODP, sharing data from European institutions), the Free flow of non-personal data initiative (including the FFD-Regulation, cyber security and self-regulation) and the EU Blockchain Observatory and Forum.
A European initiative in the strongly related field of AI is the European AI Alliance, established by the EU Commission. An international project on AI and -inter alia- training data is the “AI and Data Commons” of the ITU (International Telecommunication Union).
EU Data Strategy
On February 19 2020 The EU Commission published its ‘EU Data Strategy’. The EU aims to become a leading role model for a society empowered by data and will to that end create Common European Data Spaces in verticals such as Industrial Manufacturing, Health, Energy, Mobility, Finance, Agriculture and Science. An industrial package to further stimulate data sharing follows in March 2020.
In addition, the EU Commission has appointed an Expert Group to advise on Business-to-Government Data Sharing (B2G). In its final report, the Expert Group recommends the creation of a recognized data steward function in both public and private sectors, the organization of B2G data-sharing collaborations and the implementation of national governance structures by Member States. The aim of B2G data sharing is to improve public service, deploy evidence-based policy and advise the EU Commission on the development of B2G data sharing policy.
In its 2019 Policy & Investment Recommendations, the High-Level Expert Group on Artificial Intelligence (AI-HLEG) also devoted an entire section to fostering a European data economy, including data sharing recommendations, data infrastructure and data trusts. Finally, in a recent report, the German Opinion of the Data Ethics Commission made 75 authoritative recommendations on general ethical and legal principles concerning the use of data and data technology.
Given that data are generated by such a vast and varied array of devices and activities, and used across so many different economic sectors and industries, it is not easy to picture an all-inclusive single policy framework for data.
Dutch vision on B2B data sharing
At the beginning of this year, the Dutch government published a booklet about the Dutch Digitization Strategy, in which it sets out its vision on data sharing between companies. This vision consists of 3 principles:
- Principle 1: Data sharing is preferably voluntary.
- Principle 2: Data sharing is mandatory if necessary.
- Principle 3: People and companies keep a grip on data.
The Dutch Ministry of Economic Affairs is currently exploring the possibilities of encouraging the use of internationally accepted FAIR principles in sharing private data for AI applications. FAIR stands for (Findable, Accessible, Interoperable, Reusable). The Personal Health Train initiative builds on FAIR data principles.
Recent Dutch initiatives in the field of data sharing are the Dutch Data Coalition (self-sovereignty of data), aimed at cross-sectoral data sharing between companies and institutions, the Dutch AI Coalition (NL AIC) as well as some hands-on Data Platform and Data Portal projects from leading academic hospitals, Universities of Technology and frontrunning companies.
- Mixed datasets: 2 laws (GDPR & FFD Regulation) in tandem
More and more datasets consist of both personal and non-personal machine generated data; both the General Data Protection Regulation (GDPR) and the Regulation on the free flow of non-personal data (FFD) apply to these “mixed datasets”. The Commission has drawn up guidelines for these mixed datasets where both the FFD Regulation and the GDPR apply, including its right to data portability. Based on these two Regulations, data can move freely within the European Union.
Market barriers for early-stage AI-startups
The GDPR thoroughly protects the personal data of EU citizens. In some cases however, GDPR legislation is also hampering the European internal market with regard to the rapid rollout of AI and data startups (SME’s). This applies in particular to a smaller group of early-stage AI-startups who often lack sufficient resources to hire a specialized lawyer or a Data Protection Officer. Therefore, these companies are hesitant to do anything spectacular with personal data, and otherwise in large public-private consortia in which one operates ‘gründlich’, but where it takes (too) long to create the necessary trust among the participants. This hinders the innovative performance of early-stage AI-startups. In that sense, complex data protection rules do not encourage ambitious moonshot thinking, creative, revolutionary AI and data field experiments and the design of clever products that solve real-world problems. It is paramount that the whole field has a good grasp on the legal dimensions of their data. And that there are no significant restrictions and market barriers in that important early stage. Sharing data is simply a necessary condition for a successful AI ecosystem.
A second axiom that has the potential to inhibit rapid scientific advances in the EU -in case of expected large risks or unknown risks- is the precautionary principle. EU lawmakers have a tendency to minimize risk and prevent all possible negative scenarios ex ante via legislation. It doesn’t make drafting directives and regulations faster. Rigid application of the precautionary principle in EU law promotes excessive caution and hinders progress. It remains at odds with accelerated technological innovation.
- California Consumer Privacy Act (CCPA 2020)
The GDPR also has some important advantages for European startups and scaleups. The advantage of the GDPR is that it is now the international standard in the field of the use of personal data when doing business internationally. Partly for this reason, California has largely taken over the spirit/contents of the GDPR, and implemented it -with a fundamental American approach- in its own regulations that better protect consumer data and safeguard the trade thereof. The California Consumer Privacy Act (CCPA 2020), state-level privacy legislation, came into force on January 1, 2020. If European startups and scaleups are completely GDPR-proof, there will be no privacy legislation anywhere in the world that will require major changes to their personal data protection policy, including the associated legal uncertainty and legal costs. This is a significant competitive advantage. From that lens, European tech startups and AI-scaleups have a head start on their competitors from outside the European Union.
- Future EU AI and Data Regulation: CAHAI & EU Commission Whitepaper
Transformative technology is not a zero sum game, but a win-win strategy that creates new value. The Fourth Industrial Revolution will create a world where anything imaginable to improve the human condition, could actually be built.
The CAHAI (Ad Hoc Committee on Artificial Intelligence), established by the Committee of Ministers of the Council of Europe is currently examining the possibility of a binding legal framework for the development, design and application of AI and data, based on the universal principles and standards of the Council of Europe on human rights, democracy and the rule of law. The CAHAI expects to be able to report by March 2020 on the possibilities and necessity of new legislation.
Both data sharing practices and AI-Regulation are high on the EU Commission’s agenda. On February 19th 2020, the EU Commission published its ‘White Paper On Artificial Intelligence – A European approach to excellence and trust’. Fortunately, the White Paper uses a risk-based approach, not a precautionary principle-based approach. The Commission ‘supports a regulatory and investment oriented approach with the twin objective of promoting the uptake of AI and of addressing the risks associated with certain uses of this new (data-driven) technology.’  In its White Paper, the Commission addresses issues concerning the scope of a future EU regulatory framework and -to ensure inclusiveness and legal certainty- discusses requirements for the use of training datasets. In addition, the Commission contends that independent audits, certification and prior conformity assessments for high risk areas like Health and Transportation, could be entrusted to notified bodies (instead of commercial parties) designated by Member States. The Commission concludes with the desire to become a global hub for data and to restore technological sovereignty.
When developing informed transformative tech related policies, the starting point is to identify the desired outcome. In the case of IP policy, that outcome would be to compose a regime that balances underprotection and overprotection of IP rights per economic sector. IP is supposed to serve as a regulatory system of stimulation of creation and innovation that uses market dynamisms to reach this objective. The goal should be no less than a Pareto optimum and if possible a Pareto improvement by incentivizing innovation, encouraging scientific progress and increasing overall prosperity.
Modalities of AI-regulation
Law is just one modality of AI-regulation. Other important regulatory modalities to balance the societal effects of exponential innovation and digital transformation are the actual design of the AI system, social norms and the market. Data governance should be less fixed on data ownership and more on rules for the usage of data.
The goal should be global open data sharing community with freedom to operate and healthy competition between firms, including unification of data exchange models so that they are interoperable and standardized in the IoT. There is an urgent need for comprehensive, cross sectoral data reuse policies that include standards for interoperability, compatibility, certification and standardization.
Against this background, strengthening and articulation of competition law is more opportune than extending IP rights. Within the context of AI-regulation and data sharing practices, there is no need for adding extra layers of copyrights, database rights, patent rights and trade secret rights.
Technology shapes society, society shapes technology
Society should actively shape technology for good. The alternative is that other societies, with social norms and democratic standards that perhaps differ from our own public values, impose their values on us through the design of their technology.
AI for Good norms, such as data protection by design and by default, as well as Accountability of controllers and processors, transparency, trust and control should be built in the architecture of AI systems and high quality training datasets from the first line of code. In practice, this can be accomplished through technological synergies such as a symbiosis of AI and blockchain technology. Crossovers can offer solutions for challenges concerning the AI-black box, algorithmic bias and unethical use of data. That way, society can benefit from the benevolent side of AI.
Robust, collaborative AI framework development standards such as federated machine leaning models provide personalized AI and safeguard data privacy, data protection, data security and data access rights. Using Privacy by Design as a starting point, with build in public values, the federated learning model is consistent with Human-Centered AI and the European Trustworthy AI paradigm. As technology shapes society, society shapes technology.
 Mauritz Kop, Stanford Law School TTLF Fellow, Stanford University; Managing Partner at AIRecht, Amsterdam, The Netherlands. Correspondence: email@example.com. The author would like to thank Mark Lemley, Begoña Gonzalez Otero, Teresa Quintel, Suzan Slijpen and Nathalie Smuha for valuable remarks on an earlier draft of this article, and Christophe Geiger for his lecture on Big Data, Artificial Intelligence, Freedom of Information and the TDM exception, organized by IViR, 10 March 2020.
 Data and information are not always interchangeable terms. From a European trade secrets perspective, it is not clear whether data or datasets fulfill the requirements of Article 2(1) of the EU Trade Secrets Directive (TSD). When data is mentioned in the TSD, the terms seems to be not understood as “datasets” but rather in the context of customer/supplier lists – “commercial data” in recital 2 or “personal data” in Article 9(4). The TSD was not developed with the data-driven economy in mind, but rather on the information society (recitals 1 and 4).
 See for international commercial law aspects: Kristina Irion & Josephine Williams (2019). ‘Prospective Policy Study on Artificial Intelligence and EU Trade Policy’. Amsterdam: The Institute for information Law (IViR) 2019. See for consumer protection: Gabriele Accardo and Maria Rosaria Miserendino, ‘Big Data: Italian Authorities Published Guidelines and Policy Recommendation on Competition, Consumer Protection, and Data Privacy Issues’, TTLF Newsletter on Transatlantic Antitrust and IPR Developments Stanford-Vienna Transatlantic Technology Law Forum, Stanford University, 2019 Volume 3-4. https://ttlfnews.wordpress.com/2019/11/29/big-data-italian-authorities-published-guidelines-and-policy-recommendation-on-competition-consumer-protection-and-data-privacy-issues/. See for unfair competition law, data sharing and social media platforms: Catalina Goanta, ‘Facebook’s Data Sharing Practices under Unfair Competition Law’, TTLF Newsletter on Transatlantic Antitrust and IPR Developments Stanford-Vienna Transatlantic Technology Law Forum, Stanford University, 2018 Volume 2. https://ttlfnews.wordpress.com/2018/06/08/facebooks-data-sharing-practices-under-unfair-competition-law/ See for competition law as a driver for digital innovation and its relationship with IP law: Josef Drexl, ‘Politics, digital innovation, intellectual property and the future of competition law’, Concurrences Review 4 (2019), 2-5. https://www.concurrences.com/en/review/issues/no-4-2019/foreword/politics-digital-innovation-intellectual-property-and-the-future-of-competition
 All European Member States have civil law systems. Great Britain, as the USA, has a common law system.
 WIPO Conversation on Intellectual Property (IP) and Artificial Intelligence (AI), Second Session,
Draft Issues Paper on Intellectual Property Policy and Artificial Intelligence, prepared by the WIPO Secretariat, December 13, 2019 https://www.wipo.int/about-ip/en/artificial_intelligence/policy.html
 Ibid. See also: https://www.wipo.int/meetings/en/doc_details.jsp?doc_id=470053
 WIPO is planning to launch a digital time stamping service that will help innovators and creators prove that a certain digital file was in their possession or under their control at a specific date and time. See: ‘Intellectual property in a data-driven world’, WIPO Magazine October 2019 https://www.wipo.int/wipo_magazine/en/2019/05/article_0001.html The time stamping initiative is a digital notary service that resembles the BOIP i-Depot, see https://www.boip.int/en/entrepreneurs/ideas
 Directive 96/9/EC of the European Parliament and of the Council of 11 March 1996 on the legal protection of databases (Database Directive): https://eurlex.europa.eu/LexUriServ/LexUriServ.do?uri=CELEX:31996L0009:EN:HTML For an analysis of the rules on authorship and joint authorship of both databases and database makers’ sui generis rights, and how to overcome potential problems contractually see: Michal Koščík & Matěj Myška (2017), ‘Database authorship and ownership of sui generis database rights in data-driven research’, International Review of Law, Computers & Technology, 31:1, 43-67, DOI: 10.1080/13600869.2017.1275119
 See also CJEU, Case C-490/14 Verlag Esterbauer, The CJEU notes that the term “database” is to be given a wide interpretation. In the case of hand-labelled data for supervised machine learning, application of the Database Directive is not really straight forward. The Database Directive does not distinguish between hand and machine coding in what it protects, only between digital and analogue databases. It has been evaluated for the second time in 2018, see https://ec.europa.eu/digital-single-market/en/protection-databases
 Mezei, Péter, Digital First Sale Doctrine Ante Portas — Exhaustion in the Online Environment (June 7, 2015). JIPITEC – Journal of Intellectual Property, Information Technology and E-Commerce Law, Vol. 6., Issue 1., p. 23-71, 2015. Available at SSRN: https://ssrn.com/abstract=2615552. This rule has two exceptions: online transmission of the database and lending or rental of databases do not result in exhaustion.
 Bernt Hugenholtz, ‘Something Completely Different: Europe’s Sui Generis Database Right’, in: Susy Frankel & Daniel Gervais (eds.), The Internet and the Emerging Importance of New Forms of Intellectual Property (2016), 205-222. See also SCOTUS landmark decision Feist: Feist Publications, Inc., v. Rural Telephone Service Company, Inc., 499 U.S. 340 (111 S.Ct. 1282, 113 L.Ed.2d 358), No. 89-1909. https://www.law.cornell.edu/supremecourt/text/499/340
 See also James Grimmelmann, ‘Copyright for Literate Robots’ (101 Iowa Law Review 657 (2016), U of Maryland Legal Studies Research Paper No. 2015-16) 678, https://scholarship.law.cornell.edu/facpub/1481/. Access to out-of-commerce works held by cultural heritage institutions also requires clearance. In Europe, this license can be obtained from collective rights organisations (Article 8 CDSM Directive).
 The non-technologically neutral definition of ‘text and data mining’ in the CDSM Directive is ‘any automated analytical technique aimed at analysing text and data in digital form in order to generate information which includes but is not limited to patterns, trends and correlations’.
 Whether for research purposes or for commercial product development purposes.
 Bernt Hugenholtz, The New Copyright Directive: Text and Data Mining (Articles 3 and 4), Kluwer Copyright Blog (July 24, 2019), http://copyrightblog.kluweriplaw.com/2019/07/24/the-newcopyright-directive-textand-data-mining-articles-3-and-4/?print=print Article 4 CDSM allows right holders to opt out of the TDM exemption.
 Ducato, Rossana and Strowel, Alain M., ‘Limitations to Text and Data Mining and Consumer Empowerment: Making the Case for a Right to Machine Legibility’ (October 31, 2018). CRIDES Working Paper Series, 2018. Available at SSRN: https://ssrn.com/abstract=3278901
 Geiger, Christophe and Frosio, Giancarlo and Bulayenko, Oleksandr, ‘The Exception for Text and Data Mining (TDM) in the Proposed Directive on Copyright in the Digital Single Market – Legal Aspects’ (March 2, 2018). Centre for International Intellectual Property Studies (CEIPI) Research Paper No. 2018-02.
 Ibid. (supra note 19)
 See also WIPO (supra note 6)
 Such as in smart cities, smart energy meters, Wi-Fi lamps and user gadgets including smart wearables, televisions, smart cameras, smartphones, game controllers and music players.
 Countries with more room in their legal frameworks i.e. less legal barriers to train machine learning models are Switzerland, Canada, Israel, Japan and China.
 Ducato and Strowel (supra note 17)
 Such an innovation friendly reform directly impacts the Digital Single Market. It is to be hoped that the necessary policy space to realize these much needed revisions exists in Brussels.
 For the latest scientific breakthrough in machine learning methods see: Matthew Vollrath, ‘New machine learning method from Stanford, with Toyota researchers, could supercharge battery development for electric vehicles’, February 19, 2020 https://news.stanford.edu/2020/02/19/machine-learning-speed-arrival-ultra-fast-charging-electric-car/ According to Stanford professors Stefano Ermon and William Chueh the machine isn’t biased by human intuition. The researcher’s ultimate goal is to optimize the process of scientific discovery itself.
 An example of such an AI system is a generative adversarial network, which consists of two different neural networks competing in a game.
 Drexl, Josef and Hilty, Reto and Beneke, Francisco and Desaunettes, Luc and Finck, Michèle and Globocnik, Jure and Gonzalez Otero, Begoña and Hoffmann, Jörg and Hollander, Leonard and Kim, Daria and Richter, Heiko and Scheuerer, Stefan and Slowinski, Peter R. and Thonemann, Jannick, Technical Aspects of Artificial Intelligence: An Understanding from an Intellectual Property Law Perspective (October 8, 2019). Max Planck Institute for Innovation & Competition Research Paper No. 19-13. Available at SSRN: https://ssrn.com/abstract=3465577
 For example in NASA Antenna. See: Hornby, Greg & Globus, Al & Linden, Derek & Lohn, Jason. (2006), ‘Automated Antenna Design with Evolutionary Algorithms’, Collection of Technical Papers – Space 2006 Conference. 1. 10.2514/6.2006-7242. https://ti.arc.nasa.gov/m/pub-archive/1244h/1244%20(Hornby).pdf
 Kairouz, Peter & McMahan, H. & Avent, Brendan & Bellet, Aurélien & Bennis, Mehdi & Bhagoji, Arjun & Bonawitz, Keith & Charles, Zachary & Cormode, Graham & Cummings, Rachel & D’Oliveira, Rafael & El Rouayheb, Salim & Evans, David & Gardner, Josh & Garrett, Zachary & Gascón, Adrià & Ghazi, Badih & Gibbons, Phillip & Gruteser, Marco & Zhao, Sen. (2019). ‘Advances and Open Problems in Federated Learning’, https://arxiv.org/pdf/1912.04977.pdf
 Bonawitz, Keith & Eichner, Hubert & Grieskamp, Wolfgang & Huba, Dzmitry & Ingerman, Alex & Ivanov, Vladimir & Kiddon, Chloe & Konečný, Jakub & Mazzocchi, Stefano & McMahan, H. & Overveldt, Timon & Petrou, David & Ramage, Daniel & Roselander, Jason. (2019), ‘Towards Federated Learning at Scale: System Design’, https://arxiv.org/pdf/1902.01046.pdf
 Ibid. (supra note 30)
 Ibid. (supra note 31)
 Tjong Tjin Tai, Eric, ‘Een goederenrechtelijke benadering van databestanden’, Nederlands Juristenblad, 93(25), 1799 – 1804. Wolters Kluwer, ISSN 0165-0483. The author contends that data files should be treated analogous to property of tangible objects within the meaning of Book 3 and 5 of the Dutch Civil Code, as this solves several issues regarding data files.
 Until new European legislation creates clarity, gaps and uncertainties will have to be filled by the courts.
 Unfortunately, licensing large datasets commercially almost never works out in practice.
 For further reading about IP and property rights vested in private data see Begonia Otero, ‘Evaluating the EC Private Data Sharing Principles: Setting a Mantra for Artificial Intelligence Nirvana?’, 10 (2019) JIPITEC 87 para 1. https://www.jipitec.eu/issues/jipitec-10-1-2019/4878. For non-personal machine generated data see P. Bernd Hugenholtz, ‘Data Property: Unwelcome Guest in the House of IP (25 August 2017), http://copyrightblog.kluweriplaw.com/2017/08/25/data-producers-right-unwelcome-guest-house-ip/ and Ana Ramalho, ‘Data Producer’s Right: Power, Perils & Pitfalls’ (Paper presented at Better Regulation for Copyright, Brussels, Belgium 2017)
 Kerber, Wolfgang, ‘A New (Intellectual) Property Right for Non-Personal Data? An Economic Analysis‘ (October 24, 2016). Gewerblicher Rechtsschutz und Urheberrecht, Internationaler Teil (GRUR Int), 11/2016, 989-999. See also Landes, William M., and Richard A. Posner. “An Economic Analysis of Copyright Law.” The Journal of Legal Studies, vol. 18, no. 2, 1989, pp. 325–363. JSTOR, www.jstor.org/stable/3085624
 James Boyle, The Public Domain: Enclosing the Commons of the Mind, (Orange Grove Books 2008) 236
 Kop, Mauritz, AI & Intellectual Property: Towards an Articulated Public Domain (June 12, 2019). Forthcoming Texas Intellectual Property Law Journal 2020, Vol. 28. Available at SSRN: https://ssrn.com/abstract=3409715 The legal concept of Res Publicae ex Machina is a catch-all solution.
 Exhaustion of certain IP rights may apply, see note 11. See also Shubha Ghosh and Irene Calbol, ‘Exhausting Intellectual Property Rights: A Comparative Law and Policy Analysis’, (CUP 2018), 101
 Ibid. Kop (supra note 40). See also Deltorn, Jean-Marc and Macrez, Franck, Authorship in the Age of Machine learning and Artificial Intelligence (August 1, 2018). In: Sean M. O’Connor (ed.), The Oxford Handbook of Music Law and Policy, Oxford University Press, 2019 (Forthcoming) ; Centre for International Intellectual Property Studies (CEIPI) Research Paper No. 2018-10. Available at SSRN: https://ssrn.com/abstract=3261329
 This means that there should be no sui generis database right vested in such datasets in Europe. No contract or license will be required for the consent of the right holders for analysis, use or processing of the data.
 Directive (EU) 2019/790 of the European Parliament and of the Council of 17 April 2019 on copyright and related rights in the Digital Single Market and amending Directives 96/9/EC and 2001/29/EC (CDSM Directive), https://eur-lex.europa.eu/eli/dir/2019/790/oj
 Kop (supra note 40). The legal concept of Res Publicae ex Machina is a catch-all solution.
 Autonomously generated non personal datasets should be public domain.
 Hilty, Reto and Hoffmann, Jörg and Scheuerer, Stefan, Intellectual Property Justification for Artificial Intelligence (February 11, 2020). Draft chapter. Forthcoming in: J.-A. Lee, K.-C. Liu, R. M. Hilty (eds.), Artificial Intelligence & Intellectual Property, Oxford, Oxford University Press, 2020, Forthcoming; Max Planck Institute for Innovation & Competition Research Paper No. 20-02. Available at SSRN: https://ssrn.com/abstract=3539406 The article debates the question of justification of IP rights for both AI as a tool and AI-generated output in light of the theoretical foundations of IP protection, from both legal embedded deontological and utilitarian economic positions.
 Kop (supra note 40). Not opting for the patent route poses the risk of (bona fide) independent invention by someone else who does opt for the patent route instead of the trade secret strategy.
 Wachter, Sandra and Mittelstadt, Brent, ‘A Right to Reasonable Inferences: Re-Thinking Data Protection Law in the Age of Big Data and AI’ (October 05, 2018). Columbia Business Law Review, 2019(1). https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3248829
 Kop (supra note 40). Besides that, uncertainty about the scope of the TDM exceptions leads to litigation.
 For certain AI systems, open data should be required for safety reasons.
 European Commission, ‘A European strategy for data’, Brussels, 19.2.2020 COM(2020) 66 final, https://ec.europa.eu/info/sites/info/files/communication-european-strategy-data-19feb2020_en.pdf & https://ec.europa.eu/info/strategy/priorities-2019-2024/europe-fit-digital-age/european-data-strategy_en#documents
 Towards a European strategy on business-to-government data sharing for the public interest. Final report prepared by the High-Level Expert Group on Business-to-Government Data Sharing, Brussels, European Union, February 2020, doi:10.2759/731415 https://www.euractiv.com/wp-content/uploads/sites/2/2020/02/B2GDataSharingExpertGroupReport-1.pdf The report provides a detailed overview of B2G data sharing barriers and proposes a comprehensive framework of policy, legal and funding recommendations to enable scalable, responsible and sustainable B2G data sharing for the public interest.
 High-Level Expert Group on Artificial Intelligence, ‘Policy and Investment Recommendations for Trustworthy AI’ (European Commission, 26 June 2019). https://ec.europa.eu/digital-single-market/en/news/policy-and-investment-recommendations-trustworthy-artificial-intelligence
 Ibid. (supra note 6)
 Johan van Soest, Chang Sun, Ole Mussmann, Marco Puts, Bob van den Berg, Alexander Malic, Claudia van Oppen, David Towend, Andre Dekker, Michel Dumontier, ‘Using the Personal Health Train for Automated and Privacy-Preserving Analytics on Vertically Partitioned Data’, Studies in Health Technology and Informatics 2018, 247: 581-585
 Regulation (EU) 2016/679 of the European Parliament and of the Council of 27 April 2016 on the protection of natural persons with regard to the processing of personal data and on the free movement of such data, and repealing Directive 95/46/EC (General Data Protection Regulation). A new European ePrivacy Regulation is currently under negotiation. Data protection and privacy are two different things.
 Regulation (EU) 2018/1807 of the European Parliament and of the Council of 14 November 2018 on a framework for the free flow of non-personal data in the European Union (FFD Regulation).
 Practical guidance for businesses on how to process mixed datasets: https://ec.europa.eu/digital-single-market/en/news/practical-guidance-businesses-how-process-mixed-datasets
 Besides the GDPR, the Law Enforcement Directive (LED) regulates requirements aimed at ensuring that privacy and personal data are adequately protected during the use of AI-enabled products and services. LED: Directive (EU) 2016/680 of the European Parliament and of the Council of 27 April 2016 on the protection of natural persons with regard to the processing of personal data by competent authorities for the purposes of the prevention, investigation, detection or prosecution of criminal offences or the execution of criminal penalties, and on the free movement of such data, and repealing Council Framework Decision 2008/977/JHA.
 I speak from personal experience in our law firm. This concerns especially European AI-startups who often do not have the necessary budget to be properly advised on how to navigate data protection and data sharing regulation. See for a first report that confirms this claim: OECD Report ‘Enhancing Access to and Sharing of Data – Reconciling Risks and Benefits for Data Re-use across Societies’, November 26, 2019, Chapter 4. https://www.oecd.org/sti/enhancing-access-to-and-sharing-of-data-276aaca8-en.htm
 A solution that takes away legal roadblocks and encourages market entry of early-stage AI-startups could be targeted government funding in the form of knowledge vouchers.
 From this point of view, innovation remains at odds with privacy.
 In certain domains, performing independent audits and conformity assessments by notified bodies might be a better option. Especially in a civil law legal tradition, where lawmakers draft concise statutes that are meant to be exhaustive.
 With 500 million consumers, Europe is the largest single market in the world.
 For a close comparison of the GDPR and California’s privacy law, see Chander, Anupam and Kaminski, Margot E. and McGeveran, William, ‘Catalyzing Privacy Law’ (August 7, 2019). U of Colorado Law Legal Studies Research Paper No. 19-25. Available at SSRN: https://ssrn.com/abstract=3433922 The article contends that California has emerged as an alternate contender in the race to set the new standard for privacy (which, as mentioned in note 3, is not always the same as data protection).
 Such as China, India, Japan, South Korea and Taiwan.
 Autonomous AI agents that utilize data and deep learning techniques to continuously perform and improve at its tasks already exist. AI agents that autonomously invent novel technologies and create original art. These AI systems need data to mature.
 The Council of Europe, located in Strasbourg, France is not the same governing body as the European Commission. The Council of Europe is not part of the European Union. The European Court of Human Rights, which enforces the ECHR, is part of the Counsel of Europe.
 European Commission, White Paper on Artificial Intelligence – A European approach to excellence and trust, Brussels, 19.2.2020 COM(2020) 65 final, https://ec.europa.eu/info/sites/info/files/commission-white-paper-artificial-intelligence-feb2020_en.pdf
 Alternative Regulatory Instruments (ARIs) such as the AI Impact Assessment, see: https://airecht.nl/blog/2018/ai-impact-assessment-netherlands See also: Carl Vander Maelen, ‘From opt-in to obligation? Examining the regulation of globally operating tech companies through alternative regulatory instruments from a material and territorial viewpoint’, International Review of Law, Computers & Technology, 2020, DOI: 10.1080/13600869.2020.1733754
 See also WIPO (supra note 8). WIPO is comparing the main government instruments and strategies concerning AI and IP regulation and will create a dedicated website that collects these resources for the purpose of information sharing.
 Hilty (supra note 47)
 Kop (supra note 40)
 Smuha, Nathalie A., From a ‘Race to AI’ to a ‘Race to AI Regulation’ – Regulatory Competition for Artificial Intelligence (November 10, 2019). Available at SSRN: https://ssrn.com/abstract=3501410. The author contends that AI applications will necessitate tailored policies on the one hand, and a holistic regulatory approach on the other, with due attention to the interaction of various legal domains that govern AI.
 Lawrence Lessig, The Law of the Horse: What Cyberlaw Might Teach, 113 Harvard Law Review 501-549 (1999)
 Otero (supra note 37). For user generated data see: Shkabatur, Jennifer, ‘The Global Commons of Data’ (October 9, 2018). Stanford Technology Law Review, Vol. 22, 2019; GigaNet: Global Internet Governance Academic Network, Annual Symposium 2018. Available at SSRN: https://ssrn.com/abstract=3263466
 For an example of interconnectivity and interoperability of databases in line with the fundamental rights standards enshrined in the EU Charter: Quintel, Teresa, Connecting Personal Data of Third Country Nationals: Interoperability of EU Databases in the Light of the CJEU’s Case Law on Data Retention (March 1, 2018). University of Luxembourg Law Working Paper No. 002-2018. Available at SSRN: https://ssrn.com/abstract=3132506
 John Wilbanks; & Stephen H Friend, ‘First, design for data sharing’, (Nature, 2016)
 Drexl, (supra note 2). The Fourth Industrial Revolution may even require a complete redesign of our current IP regime.
 Kop (supra note 40). For non-IP policy tools that incentivize innovation, see: Hemel, Daniel Jacob and Ouellette, Lisa Larrimore, ‘Innovation Policy Pluralism’ (February 18, 2018). Yale Law Journal, Vol. 128, p. 544 (2019); Stanford Public Law Working Paper; Stanford Law and Economics Olin Working Paper No. 516; U of Chicago, Public Law Working Paper No. 664; University of Chicago Coase-Sandor Institute for Law & Economics Research Paper No. 849. Available at SSRN: https://ssrn.com/abstract=3125784. See also: Mauritz Kop, ‘Beyond AI & Intellectual Property: Regulating Disruptive Innovation in Europe and the United States – A Comparative Analysis’ (December 5 2019) https://law.stanford.edu/projects/beyond-ai-intellectual-property-regulating-disruptive-innovation-in-europe-and-the-united-states-a-comparative-analysis/
 Kop (supra note 40)
 Combination is the key. Examples of potential unethical use of AI are facial recognition and predictive policing.
 See note 30 and 31.
 High-Level Expert Group on Artificial Intelligence, ‘Ethics Guidelines for Trustworthy AI’ (European Commission, 8 April 2019). See https://ec.europa.eu/newsroom/dae/document.cfm?doc_id=60419. See also Paul Opitz, ‘European Commission Working on Ethical Standards for Artificial Intelligence (AI)’,
TTLF Newsletter on Transatlantic Antitrust and IPR Developments Stanford-Vienna Transatlantic Technology Law Forum, Stanford University, 2018 Volume 3-4, https://ttlfnews.wordpress.com/2018/06/08/european-commission-working-on-ethical-standards-for-artificial-intelligence-ai/
By Maria E. Sturm
On July 18, 2019, the EU Commission, the European antitrust supervisory body, fined Qualcomm Inc. for using predatory prices between 2009 and 2011. The EU Commission argued as follows:
- It explained that Qualcomm had a dominant position in the world market. This dominance is based on:
- Its market share of 60% and
- The high entry barriers: competitors are confronted with significant initial investments for research and development in this sector, as well as with obstacles regarding intellectual property rights.
- While it is not illegal to hold a dominant position, the EU Commission accuses Qualcomm of abusing such a position by using predatory prices for UMTS-chip sets. To prove this, the EU Commission used a price-cost test for the chipsets in question. In addition, it claimed to have qualitative evidence for anti-competitive behavior. Qualcomm used this problematic pricing policy when its strongest competitor, Icera, tried to expand on the market in an effort to hinder Icera from building up its market presence.
- Finally, the EU Commission argues that Qualcomm’s behavior could not be justified by potential efficiency gains.
This Commission decision is the latest, but presumably not the final point of a long case history. Already in 2015, the Commission has initiated a formal investigation and issued Statements of Objections against Qualcomm. In 2017, the Commission requested further information from Qualcomm. As Qualcomm did not respond to this request, the Commission made a formal decision which Qualcomm claimed to be void. Qualcomm filed for annulment of the decision and an application for interim measures. The latter was dismissed on July 12, 2017. Art. 278 TFEU establishes the principle that EU actions do not have a suspensory effect, because acts of EU institutions are presumed to be lawful. Thus, an order of suspension or interim measures is only issued in exceptional circumstances where it is necessary to avoid serious and irreparable harm to the applicant’s interests in relation to the competing interests. The cumulative requirements are:
- The applicant must state the subject matter of the proceedings.
- The circumstances must give rise to urgency.
- The pleas of fact and law must establish a prima facie case for the interim measures.
Qualcomm mainly brought forward two arguments for the urgency of the matter:
- The amount of work and cost for answering the questions would be too burdensome. In response, the Court stated that the damage would be only of a pecuniary nature. Such damage normally is not irreparable, because it can be restored afterwards. Qualcomm did not argue that its financial viability would be jeopardized before the final judgment would be issued, that its market share would be affected substantially, or that it was impossible to seek compensation.
- The enormous amount of work for the employees of the financial department would make it impossible for them to perform their regular tasks. However, Qualcomm mentioned itself that the burden would weigh particularly heavily on a limited number of employees in the financial department only. The Court held that if only some employees of only one department would be affected, the argument is not strong enough to justify urgency.
The latest decision in this case was issued on April 9, 2019, about Qualcomm’s claim for annulment. Qualcomm brought forward six pleas against the EU Commission, but all of them were dismissed:
- Infringement of the obligation to state reasons.
According to relevant case law, statements of reasons must be appropriate to the measure at issue and must disclose the reasons clearly and unequivocally. The Commission must show that the request is justified. The undertakings concerned must be able to assess the scope of their duty to cooperate and their right to defense must be safeguarded. However, the EU Commission is not obliged to communicate all information at its disposal, as long as it clearly indicates the suspicions. Since the Commission clearly indicated the products and the customers involved, as well as the suspicions of infringement, it fulfilled its obligation to state reasons.
- Infringement of the principle of necessity.
There must be a correlation between the request for information and the presumed infringement and the Commission must presume reasonably that the information will help determine whether the alleged infringement has taken place. Qualcomm accused the Commission of having expanded the scope of the investigation and contended that the required information was not necessary. Concerning the first point, Qualcomm mainly referred to the Statement of Objections and said the Commission must terminate its preliminary investigations before issuing such a document. However, according to the Court, the Statement of Objections is only procedural and preliminary, the Commission is thus free to continue with fact-finding afterwards. Concerning the second point, the Court grants the Commission broad powers of investigation, including the assessment of whether information is necessary or not. In particular, the Commission needed the information required from Qualcomm to avoid factual errors in calculating the price-cost test. The Commission had to request the information to keep up with its obligation to examine carefully and impartially.
- Infringement of the principle of proportionality.
Qualcomm argued that the information requested was disproportionate, because it was not legally obliged to keep documents for longer than three and a half years and that they did not organize their documents systematically. Here, the Court said that at least from the moment when Qualcomm learned about the Commission’s investigation, it should have been more careful and should have kept documents. Moreover, undertakings who keep their documents in an organized and systematic order cannot be penalized for that. Therefore, the unsystematic organization falls under Qualcomm’s responsibility. Furthermore, a significant workload is not per se disproportionate. One must consider it in relation to the investigated infringement. An alleged predatory pricing requires complex analyses of a large amount of data and is not disproportionate because of its nature.
- Reversal of the burden of proof.
This plea is, according to the Court, based on a misreading. Qualcomm was asked to neither audit financial accounts nor prove that they have conducted their business in accordance with the law. Rather, it was only asked to issue information for the Commission to conduct the price-cost test and internal documents pertaining to the relevant period.
- Infringement of the right to avoid self-incrimination.
Qualcomm claims that the Commission’s requests were beyond the scope of simply providing information. According to Regulation No 1/2003, undertakings cannot be forced to admit that they have committed an infringement, but they are obliged to answer factual questions and to provide documents, even if this information may be used to establish against them. Therefore, they do not have an absolute right of silence. On the contrary, in order to ensure the effectiveness of Regulation No 1/2003, the Commission is entitled to request all necessary information concerning relevant facts to prove the existence of anticompetitive conduct.
- Infringement of the principle of good administration.
Qualcomm claimed the required information was excessive and that the Commission abused its investigative power by prolonging a flawed investigation. According to the relevant case law, the principle of good administration requires EU institutions to observe the good guarantees afforded by the legal order. Those guarantees include the duty to examine carefully and impartially all the relevant aspects of the individual case. Here, the Court held that the Commission requested the information in question precisely to comply with its duty to examine carefully and impartially the arguments put forward by Qualcomm regarding the Statement of Objections.
On June 18, 2019, Qualcomm lodged an appeal for the annulment of the General Court’s judgment. The case is to be continued.
By Marie-Andrée Weiss
The U.S. Copyright Office published on 23 April 2019 a report on moral rights entitled Authors, Attribution, and Integrity: Examining Moral Rights in the United States. Karyn Temple, the Office Director, wrote in her introduction that the report focuses “on the personal rights of individual authors and artists, who have often been excluded in broader conversations about copyright legal reforms.”
This concern echoes the philosophy behind laws in European countries which are called “author’s rights” (droit d’auteur) and protect a work as being the imprint of the author’s personality. As it has been explained, for instance, by the European Court of Justice in Infopaq, a literary work is composed by words “which, considered in isolation, are not as such an intellectual creation of the author who employs them. It is only through the choice, sequence and combination of those words that the author may express his creativity in an original manner and achieve a result which is an intellectual creation.”
Since a work expresses the personality of the author, he or she must then be provided moral rights to protect the integrity of the work, as well as his or her right to be presented as the author. Moral rights are often presented as the main difference between copyright and author’s rights.
Are there moral rights in the U.S.?
Moral rights can be provided by contracts and licenses, which “have been at the forefront of protecting moral rights in the United States for many years and are commonly used in creative industries for that purpose” (p. 39 and p. 127). But what about the law?
The U.S. only acceded to the Berne Convention for the Protection of Literary and Artistic Works in 1988, in which article 6bis provides for moral rights, independently of the author’s economic rights. The author has “the right to claim authorship of the work and to object to any distortion, mutilation or other modification of, or other derogatory action in relation to, the said work, which would be prejudicial to his honor or reputation.”
As this right seems to provide authors a way to prevent fair use of their work, including the creation of derivative works, which are protected by the First Amendment, it is not surprising that the U.S. has not embraced the doctrine of moral rights. The report deals with the tensions between the First Amendment and moral rights (p.28), and calls the fair use doctrine “a vital First Amendment safeguard” (p. 30).
Indeed, no seminal moral right law was enacted after the U.S. joined the Berne Convention, as Congress determined, maybe a little bit hastily, that the United States already provided sufficient protection for the rights of attribution and integrity “through an existing patchwork of laws,” including the Lanham Act and some provisions of the Copyright Act (p. 7, p. 24 and p. 36).
However, in 1990 Congress enacted the Visual Artists Rights Act (“VARA”), section 106A of the Copyright Act, which provides authors of narrowly defined “work[s]of visual art” the right “to claim or disclaim authorship in the work, as well as a limited right to prevent distortion, mutilation, or modification of a work that is of recognized stature.”
In 1996 and 1997, the U.S. ratified the WIPO Performances and Phonograms Treaty, article 5 of which provides a moral right to performers who interpret works of art. Congress also considered in this instance that this right was already protected in the U.S. by the existing patchwork of laws, and that there was thus no need to enact a specific law (p. 26).
Congress however did enact in 1998 the Digital Millennium Copyright Act, which added section 1202 to the Copyright Act. Section 1202 prohibits, in some instances, removing, altering, or providing false copyright management information (“CMI”).
Article 5 of the 2012 WIPO Beijing Treaty on Audiovisual Performances gives performers a moral right in their live performances. While the U.S. signed the Treaty in 2012, it has not ratified it yet, and neither have any of the G6’s members.
Finally, some states have their own moral right statutes, for instance, the California Art Preservation Act of 1979 (p.120).
No need for a blanket moral right statute
In its just-released report, the Copyright Office found no need to introduce a blanket moral rights statute at this time (p.9). Instead, it suggested amending the Lanham Act and the Copyright Act, as it “believes that updates to individual pieces of the patchwork may be advisable to account for the evolution of technology and the corresponding changes within certain business practices” (p. 39).
The Office also suggested that Congress could amend VARA. The federal statute only applies to “works of visual art,” which are narrowly defined by Section 101 of the Copyright Act as works existing in a single copy or a limited edition. The report noted several cases denied VARA protection because “the work was considered promotional or advertising material” (p.66). The Office recommended, however, that only commercial art created pursuant to a contract and intended for commercial use be excluded from VARA’s scope (p.68).
The Office also suggested that Congress consider narrowly amending section 43(a) of the Lanham Act so that its unfair competition protections would include false representations of the authorship of expressive works. Section 43(a) applies to “false designation[s] of origin, false or misleading description[s] of fact, or false or misleading representation[s] of fact.” The Supreme Court had put a stop in 2003 to the use of Section 43(a) as a substitute for moral right, finding in Dastar Corp. v. 20th Century Fox Films that the section should not be recognized as a “cause of action for misrepresentation of authorship of noncopyrighted works.”
This Supreme Court decision “resulted in the fraying of one square of the moral rights patchwork as originally envisioned by Congress” (p. 54). At stake in this case was the right of attribution of a work in the public domain, which had been commercialized by a third party without indicating the original author. Right of attribution is one of the standard moral rights.
The Office also suggested that Congress add a new cause of action in a new section 1202A of Title 17, so that the author of a work could recover civil damages if he or she can prove that the defendant knowingly removed or altered CMI with the intent to conceal the author’s attribution information. Indeed, it “is common practice in the digital world for CMI to be stripped from works, disconnecting a work from its authorship and ownership information” (p.86).
Moral Rights and Right of Publicity
The Office recommended that Congress adopt a federal right of publicity law in order to reduce the uncertainty and ambiguity created by the diversity of state right of publicity laws. Almost all of the U.S. states have a right of publicity, whether at via common law, statutory law, or both, but they differ in the length and scope of protection. “As a result, there is significant variability among the protections available to an author depending upon where he or she chooses to live, and the specter of federal copyright preemption looms over many right of publicity claims” (p.117).
The report noted that “the right of publicity had provided authors with causes of action for misattribution of authorship, material alterations to the author’s work, and distribution of the author’s work in connection with inferior packaging and artwork” (p.111). However, as this right protects the name and likeness of the author or performer, it “cannot address situations where the author’s name or likeness is absent. Thus, the right of publicity can stand as a proxy for the right of attribution against violations resulting from misattribution, but has little to say in cases where the author is not credited at all” (p.113). It cannot protect the integrity of the work either.
The report briefly noted that “the increasingly accessible video editing technology behind “deepfake” software can not only fundamentally alter the content of an author’s work, but can also lead to social and moral harm for the artists and the subject of the video through malicious use” (p.8). This new technology is likely to trigger new right of publicity laws. For example, New York tried unsuccessfully to enact a new right of publicity statute that specifically addressed the issue of deep fakes.
It remains to be seen if Congress will heed the report’s suggestions. Whether it does or not, the debate on moral rights is likely to continue.
By Marie-Andrée Weiss
The EU Directive 2019/790 of the European Parliament and of the Council on Copyright and Related Rights in the Digital Single Market was approved by the EU Parliament on 17 April 2019 and was published on 17 May 2019. It concludes a long and hard-fought lobbying campaign where authors, internet companies, and the general public fiercely debated the most controversial issues of the Directive, the new related rights of press publishers (Article 15) and the new responsibility regime for online platforms (Article 17).
The Directive also addressed how works in the public domain or out-of-commerce could be used by “cultural heritage institutions,” that is, a library or a museum, and how research organizations could reproduce protected works for scientific research.
Facilitating use of content in the public domain: Article 14
Not all of the provisions of the Directive are controversial. For instance, Article 14 provides that reproductions of works in the public domain cannot be protected by copyright, unless this reproduction is original enough to be itself protected by copyright.
This means that museums and other institutions will no longer be able to claim a copyright on reproductions of works in the public domain which are in their collections. It remains to be seen if some of them will claim that the reproductions are original enough to be protected. Museums may change the way they photograph their works, although it would be difficult to claim that a mere reproduction of a painting is original enough to be protected. It could be, however, possible to claim so for the reproduction of a sculpture, a building, or a garment (clothes can be protected by copyright in the EU).
Cultural heritage institutions are, however, granted by Article 6 the right “to make copies of any works or other subject matter that are permanently in their collections, in any format of medium, for purposes of preservation…or other subject matter.” They are thus given the fair use right to entirely reproduce a work, for preservation purposes only, and even for profit. The museum stores will be well stocked.
“Out-of-commerce works” Article 8
Collective management organizations which are “sufficiently representative of [relevant] rightholders” will have the right to conclude with cultural heritage institutions a non-exclusive non-commercial license for the use of “out-of-commerce works.” This will, for instance, allow books which are no longer published to be copied and distributed by libraries, and orphan works to be featured in museums. Authors will, however, have the right at any time to exclude their works from this scheme.
Articles 3 to 5 provide for a copyright exception “for reproductions and extractions made by research organizations and cultural heritage institutions in order to carry out, for the purposes of scientific research, text and data mining of works or other subject matter to which they have lawful access.”
The organizations will have to implement “an appropriate level of security” when storing the works. The rightholder will be able to expressly reserve their rights “in an appropriate manner, such as machine-readable means,” if the work is made available online. It is thus not an opt-in scheme, but an opt-out one, and an author failing to constrain such use by digital marking, or any other method, may not have much recourse.
Article 5 of the Directive provides for a copyright exception for works used for teaching, when provided by an educational establishment, either on-site or online, through “a secure electronic environment accessible only by the educational establishment’s pupils or students and teaching staff.” This definition encompasses MOOCs, but not blogs, even if the sole purpose of the blogger is to provide information about a particular topic.
The two most controversial articles in the Directive are Article 15, which provides a related right to press publishers, and Article 17, which makes platforms liable for content protected by copyright which are illegally shared online.
Article 15 (formerly Article 11): a related right for press publishers
Article 15 provides press publishers established in the EU the exclusive right, for two years, to reproduce the works they publish and to make them available to the public, a right which has been named by some of its detractors “ancillary copyright.” Authors retain, however, the right to independently exploit their works.
Recital 54 of the Directive explains that the wide availability of online news is a key element of the business models of news aggregators and media monitoring services, and a major source of profit for them. However, this makes licensing their publications more difficult for publishers, and thus it is “more difficult for them to recoup their investments.”
Not surprisingly, this proposal was fiercely debated, by news aggregators, of course, but also by non-profit organizations that viewed this new right as a threat to free exchange of information on the Web. The rights provided by Article 15 do not apply, however, “to private or non-commercial uses of press publications by individual users.”
Article 15 does not apply to either “very short extracts of a press publication” or to “individual words,” an exception which can hardly be described as a fair use exception. It is nice to know, though, that one has the right to reproduce a single word without having to pay a fee.
Article 17 (formerly article 13): Towards an EU “DMCA”?
“Online content-sharing service providers” are defined by article 2(6) of the Directive as “provider[s] of an information society service of which the main or one of the main purposes is to store and give the public access to a large amount of copyright-protected works or other protected subject matter uploaded by its users, which it organizes and promotes for profit-making purposes.”
This long definition refers to digital platforms, such as Google or Facebook. They will have to obtain the authorization of the rightholder, for instance, through a license, in order to have the right to share the protected work with the public.
If they do not have this authorization, that is, in almost all cases, they will be liable for unauthorized acts of communication to the public of works protected by copyright, unless they “acted expeditiously, upon receiving a sufficiently substantiated notice from the rightholders, to disable access to, or to remove from their websites, the… works…and made best efforts to prevent their future upload” (Article 14.4(c)).
The platforms will have to put in place “an effective and expeditious complaint and redress mechanism…available to users of their services in the event of disputes.” This requirement is similar to the one put in place in 1998 by the Digital Millennium Copyright Act (DMCA), which provided a safe harbor for online service providers if they “expeditiously” remove or disable access to the infringing material after receiving a DMCA takedown notice.
Several legal scholars, such as Professor Wendy Seltzer and Professor Daphne Keller, have argued that the DMCA is a threat to free speech. Indeed, platforms regularly delete, automatically and zealously, works which are protected by the fair use doctrine upon receiving a DMCA notice. It is likely that the EU scheme will lead to similar overreach.
The Directive is ambiguous as to the way platforms are required to fulfill their new duties. Article 17.8 expressly provides that “application of [Article 15] shall not lead to any general monitoring obligation,” but Article 17.4(b) provides that the platforms must be able to demonstrate that they “made, in accordance with high industry standards of professional diligence, best efforts to ensure the unavailability of [protected] works.” Platforms may be inclined to consider that monitoring content by algorithms is indeed the current “high industry standards of professional diligence.”
Next stop: implementation, on a bumpy road
Member States have up to 7 June 2021 to transpose the Directive into their legal systems, since Directives, unlike Regulations, are not directly applicable in the EU.
However, the road to implementation is likely to be a bumpy one. Poland filed in May a complaint to the Court of Justice of the European Union against the EU Parliament and the EU Council, claiming that Article 17 of the Directive would lead to online censorship. The debate over the Directive is likely to continue.
By Marie-Andrée Weiss
On 4 January 2019, the U.S. Supreme Court granted certiorari to rule on whether Section 1052(a) of the Trademark Act, which prohibits the registration of immoral and scandalous marks, is facially invalid under the First Amendment. The case is Iancu v. Brunetti, Docket No. 18-302.
In 2011, Erik Brunetti filed an application to register FUCT as a federal trademark, in connection with a clothing line. The U.S.P.T.O. examining attorney refused to register it, considering it to be the past tense of the verb “to fuck,” a vulgar term. The Trademark Trial and Appeal Board (TTAB) affirmed in 2014. Brunetti appealed. While the case was pending, the U.S. Supreme Court issued its Matal v. Tam decision, which found that that the disparagement clause of the Lanham Act violated the First Amendment, because it discriminates based on content.
On December 15, 2017, the U.S. Court of Appeals for the Federal Circuit reversed the TTAB’s holding, and held that Section 2(a) is an unconstitutional restriction on free speech. The court denied the request for a rehearing in April 2018 and in September 2018, Andrei Iancu, the Director of the U.S.P.T.O., filed a petition for a writ of certiorari, which was granted by the Supreme Court.
The Lanham Act prohibits registering immoral and scandalous marks
Section 2(a) of the Lanham Act prohibits registration of immoral and scandalous marks, a prohibition which was first codified by Section 5(a) of the Trademark Act of 1905. The U.S.P.T.O. considers that a mark is immoral or scandalous if a substantial composite of the general public would find it shocking to the sense of truth, decency or propriety, in the context of contemporary attitudes and the relevant marketplace (see for instance In re Mavety Media Group Ltd. at 1371).
The government argued unsuccessfully on appeal that Section 2(a) does not implicate the First Amendment because it is either a government subsidy program or a limited public forum and that, alternatively, if it is speech, it is merely commercial speech. Such speech was defined by the Supreme Court in Va. State Bd. Of Pharmacy, as speech which does “no more than propose a commercial transaction.” It warrants the use of the Central Hudson, four-part test, not the strict scrutiny test.
In Tam, the Supreme Court had used a “heightened scrutiny test.” The Federal Circuit applied the strict scrutiny test, and found that the government had failed to prove that Section 2(a) advances the interests it asserts and is narrowly tailored to achieve that objective.
Trademark as speech
The Federal Circuit found Section 2(a) regulates speech based on its expressive content, and as such, does not merely regulate commercial speech. Indeed, the Supreme Court had noted in Tam that trademarks “often have an expressive content.” The Federal Circuit Court gave several examples of trademark applications using “FUCK” to further a worthy cause, such as FUCK HEROIN, FUCK CANCER or FUCK RACISM.
Brunetti is using the FUCT mark in connection with clothing featuring, as described by the TTAB, “strong and often explicit, sexual imagery that objectifies woman and offers degrading examples of extreme misogyny.” The Federal Circuit Court judges wrote in conclusion that they found “the use of such marks in commerce discomforting, and are not eager to see a proliferation of such marks in the marketplace.” Yet, it is speech which must be protected by the First Amendment.
The Lanham Act does not define what is a scandalous or immoral mark
The Federal Circuit found that there is no “reasonable definition” of what is “scandalous” and “immoral” and thus Section 2(a) is not construed narrowly enough to be found constitutional. Obscenity is not protected by the First Amendment, and the Supreme Court has provided a definition of it in Roth v. United States, “material which deals with sex in a manner appealing to the prurient interest.” In a concurring opinion to Brunetti, Judge Dyk explained that rather than finding Section 2(a) unconstitutional, he would have limited the scope of the clause to obscene marks.
In his Respondent brief, Brunetti added another question for the Supreme Court, whether Section (2)(a) is unconstitutionally vague under both the First and the Fifth Amendment.
The Supreme Court is likely to offer a definition of what is an immoral or scandalous mark
The Director of the USPTO argued in its petition to the Supreme Court that Section 2(a) merely prohibits the registration of scandalous marks, such as those using vulgar terms and graphic sexual image. However, what is “vulgar,” or what is “graphic” is not easily agreed upon, especially in a country as big as the U.S., which is home to many different opinions and beliefs. For example, there are no “Do Not Cuss” signs in New York restaurants, but there are still some in the South.
Whether or not the Supreme Court finds Section 2(a) unconstitutional, the court is likely to provide a definition of what is “scandalous” or immoral. ”
By Marie-Andrée Weiss
Disney Enterprises has owned since March 2003 the HAKUNA MATATA trademark, registered in class 25 for t-shirts. Disney filed the application in August 1994, shortly after the U.S. release of its widely successful “The Lion King” animated movie, which has been adapted as a musical comedy. A live-action reimagining version of the animated movie, using CGI technology, will be released this year.
In the movie and the musical comedy, and, most certainly, in the upcoming film, a character is singing the “Hakuna Matata” song, urging the young Lion King not to worry. Indeed, “Hakuna Matata” means “no problems” or “no worries” in Swahili. An article written by Cathy Mputhia in November 2018 noted that “Hakuna Matata” is “widely used in East Africa.”
Disney now faces backlash over this trademark registration. An online petition published by Shelton Mpala is asking Disney to cancel the HAKUNA MATATA trademark arguing that its decision to register it was “predicated purely on greed and is an insult not only the spirit of the Swahili people but also, Africa as a whole.” The petition has gathered more than 180,000 signatures.
Can Hakuna Matata be registered as a trademark?
The online petition claims that “Disney can’t be allowed to trademark something that it didn’t invent.” If this would be true, thousands and thousands of trademarks would be cancelled.
Indeed, it is not necessary to create a term to able to register it as a trademark; unlike a work protected by copyright, a trademark can be protected even though it is not an original work. What matters is that it is distinctive enough, since the function of a trademark is to identify the source of a product or service. Trademarks which are arbitrary, fanciful, or suggestive can be protected without having to show secondary meaning. Generic trademarks cannot be protected.
Few people in the U.S. knew the “Hakuna Matata” expression before the release of The Lion King in 1994. Indeed, Google Ngram Viewer shows that the word was first used in the English language corpus in the Nineties. However, as noted by The Guardian, “[t]he phrase was popularised in 1982 by the Kenyan band Them Mushrooms, whose platinum-selling single Jambo Bwana (Hello, Mister) featured the phrase hakuna matata.”
When the expression was trademarked in 1994, the U.S. public associated it with the movie, which was a blockbuster, complete with derivative products, some using the “Hakuna Matata” phrase. As such, this combination of words was an efficient trademark, because the general public believed that the phrase had been invented by Disney and thus perceived the trademark as being arbitrary or fanciful.
Disney’s trademark application stated correctly, however, that the two words have a meaning in another language and were not invented. As such, HAKUNA MATATA is a suggestive trademark, a term which “requires imagination, thought and perception to reach a conclusion as to the nature of the goods,” as explained by the Second Circuit in Abercrombie & Fitch Co. v. Hinting World.
Nevertheless, the public in East Africa perceives this trademark as a mere common phrase, and generic terms cannot be registered as trademarks. Therefore, “Hakuna Matata” cannot be registered as a trademark in East Africa but can be in the U.S. or in other Western countries because the public there does not perceive it as a common expression. Research on WIPO’s Global Brand Database reveals that the term is registered as trademark almost only in Western countries, with a few exceptions such as Egypt, Thailand, and Korea.
Under trademark law, a term can only be registered if it is not commonly used in the country or geographical areas where it is registered. This practice, while legal, can be perceived by the country from which the term originated, however, as cultural appropriation.
Trademark and cultural appropriation
Even if it is legal to register “Hakuna Matata,” it may not be advisable. Cultural appropriation is now a well-known concept, but it had just started to be recognized when Disney registered the trademark in 1994, as shown by this Ngram Viewer.
Shelton Mpala told CNN that that he had started the petition “to draw attention to the appropriation of African culture and the importance of protecting our heritage, identity and culture from being exploited for financial gain by third parties.”
This issue is addressed by WIPO, which established in 2000 the Intergovernmental Committee on Intellectual Property and Genetic Resources, Traditional Knowledge and Folklore (IGC). The Committee’s mandate is to work towards reaching an agreement on one or more intellectual property international legal instruments which would protect traditional knowledge and traditional cultural expressions.
One of the IGC’s background briefs noted that “[t]he current international system for protecting intellectual property was fashioned during the age of enlightenment and industrialization and developed subsequently in line with the perceived needs of technologically advanced societies.”
Could a bar to register traditional knowledge and traditional cultural expressions be in the future?
Can this practice be regulated, and how? Cathy Mputhia suggested in her article that “relevant governments or communities [could] apply for expungement of already granted trademarks,” but noted that “there are certain thresholds that ought to be met for expungement of marks that contain heritage.”
These “expungement thresholds” will probably not be legal, but societal. If such expungement occurs for HAKUNA MATATA, it will be a business decision made by Disney designed to protect the company’s public image by acknowledging that the trademark hurts too many Africans. In this regard, Shelton Mpala chose a possible viable path towards his desired result.
One can imagine that a party could petition the TTAB to cancel the HAKUNA MATATA trademark registration under Section 2(a) of the Lanham Act which prohibits registration of immoral, scandalous, and disparaging marks. However, seems unlikely, as the U.S. Supreme Court held in 2017 in Matal v. Tam that Section 2(a)’s prohibition to register disparaging marks violates the First Amendment. The Court may soon rule similarly about scandalous and immoral marks, as it has accepted to review the In Re: Brunetti case, after the Federal Circuit held that the First Amendment also protects registration of such marks.
Section 2(b) of the Trademark Act bars the registration of a mark which consists of or comprises the flag or other insignia of the US, or any state, or municipality or foreign nation. Could the law be amended one day to add symbols of traditional knowledge and traditional cultural expressions? This would require a WIPO treaty protecting them, and the very hypothetical U.S. accession to the treaty.