Chen Yi-Chih, Chen Hung-Chih 2
Intellectual Property (IP) Management is a subject of recent focus in Taiwan . More than 1 million patents have been filed in Taiwan and each year, Taiwan dedicates NT $80 3 trillion in research and development. The estimated cost for IP prosecution, maintenance, litigation, conciliation, compensation and authorization amounts to NT $200 trillion (U.S.$6.5 trillion) 4. Even though many enterprises have gradually recognized the importance of intellectual property, the situation has not significantly improved based on the statistics stated above. Observation shows that only few enterprises in Taiwan have taken active steps to manage their IP and it was only after facing infringement lawsuits and tremendous amount of loyalty payments, most companies started to realize the important of IP management.
Two main causes are believed to have negative impact on the lacking and ineffectiveness of most Taiwanese enterprises' IP management:
Therefore, it is critical to assist these enterprises to develop and implement an effective IP management strategy under which the full potential of their IP can be utilized and the maximum value of the enterprises' IP can be realized.
The Intellectual Property Office of the Ministry of Economic Affairs recognized the importance of governmental role to address this issue. Since 2003, it has collaborated with the Institute of Information Industry to work on a project for developing a standardized IP management system. In 2005, the project was handed over to the Industrial Development Bureau which then carried on the development and promotion of the Taiwanese Intellectual Property Management System (TIPS). Taiwanese enterprises 5 are able to use TIPS as a basis to establish their own comprehensive IP management systems. Based on our experiences in promoting TIPS and the feedbacks from those enterprises which have followed the TIPS's guidance to establish their IP management systems, we are pleased to find that TIPS is capable of assisting enterprises to develop a comprehensive IP management system. The system no only meets an enterprise's operational needs but also can be continuously improved owing to its adoption of the PDCA management cycle 6.
On December 9, 2004, The Ministry of Economics, in recognition of the needs to assist Taiwanese enterprises to better manage and more fully utilize their intellectual property, organized a “Taiwanese IP Management Standardization and Promotion Summit”. In order to establish a consensus on IP management among Taiwanese enterprises and to encourage the enterprises to implement an internal IP management system, the Taiwanese government positioned TIPS as an industry standard.
In 2006, The Industrial Development Bureau (IDB) of the Ministry of Economic Affairs (MEA) established a TIPS promotion program and revised the 2004 draft of the Intellectual Property Management System Standard to become the Taiwan Intellectual Property Management System (TIPS). The industrial experts' opinions and comments were gathered and used to amend the draft, TIPS was then formally announced 7 on March 23, 2007 and consequently promoted. In hopes to protect Taiwanese enterprises and to improve their market competitiveness, IDB initiated extensive promotion program, encouraging Taiwanese enterprises and organizations to establish a convenient, efficient, and low-cost IP management system by following the TIPS's guidance
The main characteristic of TIPS is the incorporation of the PDCA (Plan-Do-Check-Action) model from the ISO 9001:2000 Quality Management System. By adopting this model, not only the challenges of IP management can be resolved, but the whole system can also be continuously improved.
Since TIPS shares the ISO's characteristics of being credible, comprehensive, and easily adaptable, TIPS and be easily integrated into the ISO standards within an enterprise such that the conflicts between these two systems will be minimized and it will only require minimum organizational structural changes and implementation costs. If an enterprise has already implemented ISO, implementing TIPS becomes more easily and efficient.
In addition, TIPS emphasizes the concepts of using “process-oriented approach” and “systematic management” 8. Enterprises can merge their existing infrastructures and TIPS to establish a convenient, effective and efficient IP management system to reduce losses caused by IP infringement. Enterprises may also strengthen their market competitiveness and increase profits through royalty income.
TIPS includes nine chapters. The first four chapters cover Summary, which describes the background of TIPS; Scope of Application and Terminologies. Clause 0.3.1 9 of TIPS states that the purpose of TIPS is to promote the utilization of IP management as one of the means to maximize an enterprise's profits. Rather than an individual or a specific department, protecting IP assets is the responsibility of all employees within the enterprises. In addition, the establishment of an IP management system is essential regardless of the scale, product or service provided by an enterprise. Clause 1.2 of TIPS clearly provides that TIPS is applicable to all enterprises, despite their types, scales, products and services provided. Therefore, TIPS is not designed solely for large enterprises. It can be applied to all kinds of organizations which include but not limit to a company, a specific department/division within a company, a laboratory or a project team.
Before establishing TIPS, the government recognized that an enormous amount of resources is required to establish an IP management system. Therefore, the ISO9001:2000 quality management framework was adopted and TIPS was developed based upon the ISO's management principles. By incorporating IP managing strategies into an enterprise's operation goals and internal activities, the IP management system is no longer just a risk management system but a system that is closely aligning to the overall operations of an enterprise.
Since it was found that many domestic companies implemented ISO9001:2000 Quality Management System solely for compliance purposes, people are skeptical about its effectiveness. In fact, if one understands the rigorous formulation processes behind the quality management system and its principles, one would recognize that an enterprise's IP management system can be significantly improved by adopting the management characteristics of ISO Quality Management System.
The main characteristics shared between TIPS and ISO are outlined as follows:
Since TIPS shares the above mentioned characteristics of the ISO Quality Management System, it not only can reduce the risks of infringing the IP rights of the others, but also can assist an organization to achieve its operational goals provided that the organization has designed relevant processes pursuant to the requirements of TIPS and has thoroughly implemented the designed processes. Using TIPS's external evaluation mechanism 10, enterprises implementing with TIPS can prove to their customers and external stakeholders that they have the capability to manage and maintain their IP.
If an enterprise follows TIPS to establish its IP management system, its expected benefits include the followings:
The Taiwanese government hopes that enterprises can systematically manage their IP through the implementation of TIPS. In other words, following TIPS's guidance, the Taiwanese enterprises should establish an IP management system that incorporates the usage of the PDCA management cycle (Plan-Do-Check-Action) and process management approach and such system must be built by taking into account the enterprise's business operation strategies and objectives. Enterprises should have clear processes and related rules for handling all IP related issues. For example, prior to filing a patent application, there should be a plan for the ways to acquire the targeted IP and prior art research shall be conducted. Based on the search results, enterprises can then decide whether they would like to internally develop the targeted IP or to seek licensing opportunities. Effective IP management processes shall be able to answer the following questions:
The following section aims to explain how Taiwanese enterprises can establish or modify their current IP management system to achieve its full potential:
All employees within an organization shall participate in order to realize the most benefits out of the IP management system. Leadership responsibilities, roles and responsibilities allocation, training and education programs and the subsequent auditing processes on the performance of operation shall be clearly defined and planned. Establishing a successful IP management system shall not be the sole responsibility of the legal department. During the implementation stage, the following personnel should participate and complete the related tasks:
Establishing a systematic IP management system requires the participation of all employees and it requires reengineering of the existing processes. It is not an easy task to be established and planned solely by the legal department. All other departments within an enterprise shall participate and offer their suggestions. The followings are the recommended stages for implementing an IP management system:
Stage | Tasks | Description | Responsibility | Remark |
1. Preparation | 1). Review of current status | Understand resources available and the status of operation | Data collection; define roles and responsibilities | |
2). Establish implementation team | Identify team members and team leader | Confirm organizational structure for implementation | ||
3). Set goals and establish all management programs | Evaluate current situation to formulate IP management policy, and define measurable goals. Processes planning shall be made by taking into account the management responsibility, resource management, product development, and performance analysis and improvement. This helps to identify the position of a process within the overall IP management system and its inter-relationships between the processes themselves. | Provide evaluation report; organize IP management deployment document | Documentation: IP Management Manual à Procedures à Guidelines à Records | |
2. Training and Education & System Integration | 4). Relevant training and education | Understand the direction, method, and spirit of standardization. | Participated by the implementation team and management representatives. | |
5).Drafting documentation | Decide documentation framework, format, table of contents, numbering principles, and appoint editors and the completion date. | Management team assigns tasks | ||
6). Establishing documentation | Drafting and revising procedural documentation | Internal discussion and review | IP management principles (refer to prior text) | |
Define the scope and content of standard format. Appoint editors and the completion date. | Establish standard format as an example before documenting | |||
Prepared IP management manual to aid employees and customers to understand the organization's IP management system | Implementation team and management team | |||
3.Implementation | 7). Provide training & education specifically for the internal audit personnel | Explain the purpose of auditing and execution details | Participated by Internal audit committee | Prepare checklist for auditing to be used by auditing personnel |
8). Conduct system implementation and internal audits | Execute documentation processes for the management system and conduct internal audits and review the performance | Implementation, review, correction and prevention. Participated by all employees | ||
9). Conduct overall examination of the intellectual property management system | Implement IP management system | Participated by all members of the implementation team |
Chapter five through chapter eight of TIPS define the core of the guidelines which cover the basic requirements of IP management requirements; top management's responsibilities; resource management; the acquisition, protection, maintenance and exploitation of IP, as well as performance evaluation and improvement. To facilitate Taiwanese enterprises' understanding of TIPS and how to use it to establish a comprehensive IP Management system, we provide the following main steps of establishing an IP management system based on the TIPS's requirements:
In the era of knowledge economy, the abilities of most domestic enterprises to manage tangible assets have gradually matured (ex. ERP system). However, the abilities to manage intangible assets which include intellectual property have yet to be developed. Management systems in most domestic enterprises are fragmented. For example, legal departments are only responsible for contract reviewing tasks; R&D staff has limited IP knowledge. The importance of IP is often overlooked and most enterprises do not see that intellectual property management is the responsibility of every employee. As a consequence, the Taiwanese government establishes and promotes TIPS to encourage domestic enterprises to adopt a systemic approach of managing their intellectual property and TIPS is also provided as a tool to assist enterprises to establish a sound intellectual property management system.
The purpose of implementing TIPS is not to request enterprises to establish a separate management system. In order to maintain efficiency and competitiveness, an enterprise has to have an integrated management system to support its core operations and also to meet the requirements of different management system standards. Eliminating overlaps of the requirements between different quality management systems is an inevitable trend. TIPS incorporates IP management with the ISO 9000 quality management system, which is capable of simplifying the complicated IP management tasks into an effective and standardized IP management system.
TIPS helps an enterprise to establish a systematic process for managing its IP. Through competitive analysis, market trend analysis, and periodic IP management operations review, a company can revise and amend its IP management policies and goals and continually improve its IP management system. For example, sales departments shall collect market trends, competitive information and shall also consciously avoid acquiring materials that may raise infringement concerns. Human resource departments shall focus their efforts in providing IP education and training. Finance departments shall evaluate the costs required for maintaining the existing IP rights and inform the R&D departments to conduct relevant review at the appropriate time. R&D departments shall conduct prior art search before a new research project is commenced.
TIPS offers a simple, efficient, and low-cost management system which assists an enterprise to establish an IP management system that aligns to its business goals and operation activities. We hope that by promoting and encouraging domestic enterprises to adopt and implement TIPS, Taiwan can strengthen its international competitiveness and sustain the growth of its economy and the whole society.
1.Taiwan Intellectual Property Management System (TIPS). The Ministry of Economics Affairs combined the IP management principles and the PDCA (Plan-Do-Check-Action) model used in ISO9001:2000 quality management system to create TIPS. The adoption of PDCA model helps organizations to establish a systematic and effective IP management system which can be continuously improved.
2. Chen Yi-Chih is a Section Manager at the Science and Technology Law Center ; Chen Hung-Chih is a legal Researcher at the Science and Technology Law Center .
3. Data Source:
http://www.atmt.org.tw/html/modules/news/article.php?storyid=135&PHPSESSID=cab6428078a0435c5af1b2e7bbe2b121 (last visited: 08/11/2007 )
4. Data Source:
http://www.cyberone.com.tw/ItemDetailPage/PDAFormat/PDAFContent.asp?MMContentNoID=36372(last visited: 08/11/2007 )
5. “Enterprise” as defined in TIPS includes company, corporate, school, research institute, a specific department or a project team is also included.
6. TIPS was developed based on the PDCA (Plan-Do-Check Action) model, a typical ISO management process which requires continuously monitoring, evaluating, analyzing and improving the whole system.
7. The TIPS guidelines can be found at: http://www.tips.org.tw/public/public.asp?selno=236&relno=236
8. Refer to article: New Philosophy of Intellectual Property – Use ISO Quality Management to establish a systematic IP management in Intellectual Property Journal, issue 74, 02/2005.
9. http://www.tips.org.tw/public/public.asp?selno=236&relno=236 (last visited: 08/12/2007 )
10. The guidelines of TIPS also serve as the requirements for certification purpose. The Industrial Development Bureau of the Ministry of Economic Affairs will issue a certificate to an organization if such organization has implemented an IP management system satisfying the requirements of TIPS.
The Introduction to the Trade Secret Management Guidelines 2024/09/09 Due to an open, collaborative culture and the need to balance knowledge sharing with protection, research academic institutions always face unique challenges in managing confidential information. However, trade secret protection is still essential for research academic institutions in order to safeguard their competitive advantages and valuable research results. Accordingly, the “Trade Secret Management Guidelines”, released by the Taiwan Intellectual Property Office (TIPO) on May 27, 2024, is specifically tailored for the trade secret protection in the research and academic circumstance. Taking into account the essential differences between research academic institutions and enterprises, these guidelines use a phased and scalable approach to implement trade secret protection measures. With these guidelines, each research academic institution can evaluate its own size, research field, available resources, etc., and establish an appropriate trade secret management system to effectively identify, protect and manage its trade secrets. The Trade Secret Management guidelines outline 13 measures for managing trade secrets, covering the entire life cycle of trade secret protection. In addition, these guidelines recommends that research academic institutions adopt a phased implementation strategy, starting from the "entry-level" stage focusing on basic measures, and gradually entering the "basic" and "enhanced" stages to improve each management measure. The following is an overview of each measure: 1.Distinguishing Trade Secrets In order to facilitate the protection of trade secrets, institutions should distinguish what is trade secret information at the entry stage, whether it is self-developed or obtained from others. As progressing to the basic stage, institutions should define and provide examples of trade secrets they produce or acquire. When entering the enhanced stage, institutions should develop a process for identifying whether an information is a trade secret. 2.Access Control To prevent unauthorized disclosure, institutions should control access to trade secrets. At the entry stage, institutions must set access permissions for trade secrets. As progressing to the basic stage, access should be granted based on the need for the information. When entering the enhanced stage, institutions need to adjust access permissions in response to job changes and personnel turnover. 3.Identification Identifying trade secrets helps ensure that those accessing the information are aware of its confidentiality. At the entry stage, institutions need to identify information considered to be trade secrets, but there are no restrictions on the identification method. As progressing to the basic stage, institutions need to clearly define how trade secrets will be identified. When entering the enhanced stage, the key is to ensure that all contacts know that the information they come into contact with is a trade secret. 4.External Disclosure Review In order not to affect subsequent research or applications, institutions should review the information that will be disclosed to the public. At the entry stage, institutions should ensure that information is reviewed by responsible personnel before it is disclosed to the public. As progressing to the basic stage, institutions need to identify which items should be reviewed. When entering the enhanced stage, institutions shall distinguish what should be reviewed based on the nature of the information disclosed to the public. 5.Circulation Control Controlling the circulation of trade secrets can prevent them from being arbitrarily disclosed. At the entry stage, institutions should ensure that responsible personnel have consent to the circulation of trade secrets. As progressing to the basic stage, the key is whether the behavior of circulating trade secrets is recorded. When entering the enhanced stage, institutions should take countermeasures to prevent trade secrets from being leaked during the circulation process. 6.Reproduction Control Controlling the reproduction of trade secrets can ensure that the use of trade secrets is limited to a controllable scope. At the entry stage, institutions should limit the reproduction of trade secrets. As progressing to the basic stage, the key is whether the behavior of reproducing trade secrets is recorded. When entering the enhanced stage, institutions should take measures to avoid the increased risk of leakage of trade secrets after the reproduction behavior. 7.Destruction At the entry stage, institutions should ensure that the consent of the responsible personnel is obtained when destroying trade secret. As progressing to the basic stage, institutions should consider the impact of destruction before destroying trade secrets and ensure that records of destruction are retained. When entering the enhanced stage, the key is whether the trade secrets are destroyed in an irrecoverable way. 8.Usage record retention Keeping the usage record of trade secrets can help provide evidence in litigation. At the entry stage, institutions should keep records of access and use of trade secrets. As progressing to the basic stage, institutions should further specify the items that need to be retained in the records. When entering the enhanced stage, institutions should ensure that the records retained are authentic and will not be arbitrarily tampered with. Therefore, if necessary, a third party agency can be entrusted with the preservation of evidence. 9.Designating Responsibility for Implementation Setting up responsible personnel can help ensure that the trade secret management mechanism is effectively implemented. At the entry stage, institutions simply need to ensure there is someone responsible for driving trade secret management. As progressing to the basic stage, institutions shall assign dedicated personnel to be responsible for trade secret management. When entering the enhanced stage, institutions must establish a dedicated unit to coordinate the protection of trade secrets through a clear division of powers and responsibilities. 10.Confidentiality and Ownership Arrangements Signing confidentiality and ownership agreements can ensure that internal personnel who may have access to trade secrets understand their confidentiality obligations. At the entry stage, institutions only need to sign a written confidentiality agreement with those who may have access to trade secrets. As progressing to the basic stage, institutions shall clearly define the items that must be included in the confidentiality and ownership agreement. When entering the enhanced stage, institutions need to further evaluate whether to adjust the contents of confidentiality and ownership agreements in response to job changes and personnel turnover. 11.Promotion and Training Through promotion and training, institutions can gradually improve personnel's awareness of confidentiality and help them understand the key points of trade secret management. At the entry stage, institutions can promote the importance of trade secret management and provide appropriate training to all personnel. As progressing to the basic stage, institutions should establish promotional materials, explain the management objectives and specific practices in the training, and conduct an evaluation of effectiveness. When entering the enhanced stage, institutions need to adjust the content required for training based on differences in units, objects, trade secret characteristics, etc. 12.Departure Management At the entry stage, institutions should remind departing personnel of their confidentiality obligations. As progressing to the basic stage, institutions shall further require the departing personnel to hand over the trade secrets they held during their tenure. When entering the enhanced stage, institutions need to conduct exit interviews with important departing personnel to ensure that they clearly understand their confidentiality obligations. 13.Confidentiality and Ownership Arrangements with External Parties Signing confidentiality and ownership agreements with external parties can help prevent institutions mired in unnecessary controversies. At the entry stage, institutions only need to sign confidentiality agreement with external parties before providing trade secrets to them. As progressing to the basic stage, institutions shall clearly define the items that must be included in the confidentiality and ownership agreement. When entering the enhanced stage, institutions needs to confirm with external parties and reach a consensus on the management measures that both parties need to take. The Trade Secret Management guidelines provide a comprehensive framework for research academic institutions in Taiwan to protect their trade secrets. Regardless of the size of the research academic institutions or the field it focuses on, as long as they follow the above-mentioned 13 measures and adjusts according to their current management situations, they can gradually establish a trade secret management system that meets their own needs.
The Dispute on WTO TRIPS IP Waiver Proposal and the Impact on TaiwanThe Dispute on WTO TRIPS IP Waiver Proposal and the Impact on Taiwan 1. IP Waiver proposal On October 2, 2020, South Africa and India summit a joint proposal (IP/C/W/669) (hereinafter as “first proposal”) for TRIPS council of the World Trade Organization(WTO), titled “Waiver from Certain Provisions of the Trips Agreement for the Prevention, Containment and Treatment of Covid-19”, called for temporary IP waiver of intellectual property in response for Covid-19 pandemic. In first proposal, it supported a waiver from the implementation or application of Sections 1, 4, 5, and 7 of Part II of the TRIPS Agreement in relation to prevention, containment or treatment of COVID-19, which directs to copyright and related rights, industrial designs, patents and protection of undisclosed information. All enforcement measures under part III of the TRIPS agreement such as civil and administrative procedures and remedies, border measures and criminal procedures for protecting aforesaid intellectual property shall also be waived until widespread vaccination is in place globally, and the majority of the world's population has developed immunity[1]. On May 25, 2021, the first proposal was revised (IP/C/W/669/Rev.1, hereinafter as “second proposal”) and resubmitted for WTO by the African Group, The Plurinational State Of Bolivia, Egypt, Eswatini, Fiji, India, Indonesia, Kenya, The Ldc Group, Maldives, Mozambique, Mongolia, Namibia, Pakistan, South Africa, Vanuatu, The Bolivarian Republic Of Venezuela and Zimbabwe[2]. In the second proposal, the scope of IP waiver was revised to be limited to "health products and technologies" used for the prevention, treatment or containment of COVID-19, and the minimum period for IP waiver was 3 years from the date of decision. 2. The Pros and Cons of IP Waiver proposal The IP waiver proposal is currently supported by over 100 WTO members. However, in order to grant the waiver, the unanimous agreement of the WTO's 159 members would be needed[3], but if no consensus is reached, the waiver might be adopted by the support of three-fourths of the WTO members[4]. The reason for IP waiver mainly focus on the increase of production and accessibility of the vaccines and treatments, since allowing multiple actors to start production sooner would enlarge the manufacturing capacity than concentrate the manufacturing facilities in the hands of a small number of patent holders[5]. Médecins Sans Frontières (MSF) also support IP waiver proposal to prevent the chilling effect of patents as hindrances of the introduction of affordable vaccines and treatment in developing countries[6], and urges wealthy countries not to block IP waiver to save lives of billions of people[7]. Most opponents against IP waiver proposal are rich countries such as European Union (EU), UK, Japan, Switzerland, Brazil, Norway, Canada, Australia[8]. On May 5, 2021, United States Trade Representative (USTR) announced its support the IP waiver, but only limited into vaccine[9]. EU was the main opponent against IP waiver proposal at the WTO[10]. On June 4, 2021, EU offered an alternative plan to replace IP waiver proposal. Specifically, EU proposed that WTO members should take multilateral trade actions to expand the production of COVID-19 vaccines and treatments, and ensure universal and fair access thereof. EU calls for WTO members to ensure that COVID-19 vaccines, treatments and their components can cross borders freely, and encourage producers to expand their production and provide vaccines with an affordable price. As to IP issues, EU encourages to facilitate the exploitation of existing compulsory licensing systems on TRIPS, especially for vaccine producers without the consent of the patent holder[11]. Many pharmaceutical companies also express dissent opinions against the IP waiver proposal. The International Federation of Pharmaceutical Manufacturers and Associations (IFPMA) indicated that the proposal would let unexperienced manufacturers, which are devoid of essential know-how, join into vaccine supply chains and crowd out the established contractors[12]. The chief patent attorney for Johnson & Johnson pointed out that since the existing of IP rights not only promote the development of safe and effective vaccines at record-breaking speed, but also allow the IP owner to enter into agreements with appropriate partners to ensure the production and distribution of qualitied vaccines, the problem resides in infrastructure rather than IP. Thus, instead of IP waiver, boosting adequate health care infrastructure, vaccine education and medical personnel might be more essential for COVID-19 vaccines equitably and rapidly distributed[13]. Pfizer CEO warned that since the production of Pfizer’s vaccine would require 280 different materials and components that are sourced from 19 countries around the world, the loss of patent protection may trigger global competition for these vaccine raw materials, and thus threaten vaccine production efficiency and affect vaccine safety[14]. Moderna CEO said that he would not worry about the IP waiver proposal since Moderna had invested heavily in its mRNA supply chain, which did not exist before the pandemic, manufacturers who want to produce similar mRNA vaccines will need to conduct clinical trials, apply for authorization, and expand the scale of production, which may take up to 12 to 18 months[15]. 3. Conclusion The grant of the IP waiver proposal might need the consensus of all WTO members. However, since the proposal might not be supported by several wealth countries, which might reflect the interest of big pharmaceutical companies, reach the unanimously agreement between all WTO members might be difficult. Besides, the main purpose for IP waiver is to increase the production of vaccines and treatments. However, when patent protection was lifted, a large number of new pharmaceutical companies lacking necessary knowhow and experience would join the production, which might not only result in snapping up the already tight raw materials, but also producing uneven quality of vaccines and drugs. Since patent right is only one of the many conditions required for the production of vaccines and drugs, IP waiver might not help increase the production immediately. Thus, other possible plans, such as the alternative plan proposed by EU, might also be considered to reduce disputes and achieving the goal of increasing production. As to the impact of the IP waiver proposal for Taiwan, it can be analyzed from two aspects: 1. Whether Taiwan need IP waiver to produce COVID-19 vaccine and drugs in need Since there is an established patent compulsory licensing system in Taiwan, the manufacture and use of COVID-19 vaccine and drugs might be legally permissible. To be specific, Article 87 of Taiwan Patent Act stipulates: “In response to national emergency or other circumstances of extreme urgency, the Specific Patent Agency shall, in accordance with an emergency order or upon notice from the central government authorities in charge of the business, grant compulsory licensing of a patent needed, and notify the patentee as soon as reasonably practicable.” Thus, in response to national emergency such as COVID-19 pandemic, Taiwan Intellectual Property Office (TIPO) could grant compulsory licensing of patents needed for prevention, containment or treatment of COVID-19, in accordance with emergency order or upon notice from the central government authorities. In fact, in 2005, in response to the avian flu outbreaks, TIPO had grant a compulsory licensing for Taiwan patent No.129988, the Tamiflu patent owned by Roche. 2. Whether IP Waiver would affect Taiwan’s pharmaceutical or medical device industry In fact, there are many COVID-19 related IP open resources for innovators to exploit, such as Open COVID Pledge[16], which provides free of charge IPs for use. Even for vaccines, Modena had promised not to enforce their COVID-19 related patents against those making vaccines during COVID-19 pandemic[17]. Therefore, currently innovators in Taiwan could still obtain COVID-19 related IPs freely without overall IP Waiver. Needless to say, since many companies in Taiwan still work for the research and development of COVID-19-related medical device and drugs, sufficient IP protection could guarantee their profit and stimulate future innovation. Accordingly, since Taiwan could produce COVID-19 vaccines and drugs in need domestically by existing patent compulsory licensing system, and could obtain other COVID-19 related IPs via global open IP resources, in the meantime IP protection would secure Taiwan innovator’s profit, IP waiver proposal might not result in huge impact on Taiwan. [1]Waiver From Certain Provisions Of The Trips Agreement For The Prevention, Containment And Treatment Of Covid-19, WTO, Oct 2, 2020, https://docs.wto.org/dol2fe/Pages/SS/directdoc.aspx?filename=q:/IP/C/W669.pdf&Open=True (last visited July 5, 2021) [2]Waiver From Certain Provisions Of The Trips Agreement For The Prevention, Containment And Treatment Of Covid-19 Revised Decision Text, WTO, May 25, 2021, https://docs.wto.org/dol2fe/Pages/SS/directdoc.aspx?filename=q:/IP/C/W669R1.pdf&Open=True (last visited July 5, 2021) [3]COVID-19 IP Waiver Supporters Splinter On What To Cover, Law360, June 30, 2021, https://www.law360.com/articles/1399245/covid-19-ip-waiver-supporters-splinter-on-what-to-cover- (last visited July 5, 2021) [4]The Legal Framework for Waiving World Trade Organization (WTO) Obligations, Congressional Research Service, May 17, 2021, https://crsreports.congress.gov/product/pdf/LSB/LSB10599 (last visited July 5, 2021) [5]South Africa and India push for COVID-19 patents ban, The Lancet, December 5, 2020, https://www.thelancet.com/journals/lancet/article/PIIS0140-6736(20)32581-2/fulltext (last visited July 5, 2021) [6]MSF supports India and South Africa ask to waive COVID-19 patent rights, MSF, Oct 7, 2020, https://www.msf.org/msf-supports-india-and-south-africa-ask-waive-coronavirus-drug-patent-rights (last visited July 5, 2021) [7]MSF urges wealthy countries not to block COVID-19 patent waiver, MSF, Feb. 3, https://www.msf.org/msf-urges-wealthy-countries-not-block-covid-19-patent-waiver (last visited July 5, 2021) [8]Rich countries are refusing to waive the rights on Covid vaccines as global cases hit record levels, CNBC, Apr. 22, 2021,https://www.cnbc.com/2021/04/22/covid-rich-countries-are-refusing-to-waive-ip-rights-on-vaccines.html (last visited July 5, 2021) [9]Statement from Ambassador Katherine Tai on the Covid-19 Trips Waiver, May 5, 2021, https://ustr.gov/about-us/policy-offices/press-office/press-releases/2021/may/statement-ambassador-katherine-tai-covid-19-trips-waiver (last visited July 5, 2021) [10]TRIPS waiver: EU Council and European Commission must support equitable access to COVID-19 vaccines for all, Education International, June 9, 2021, https://www.ei-ie.org/en/item/24916:trips-waiver-eu-council-and-european-commission-must-support-equitable-access-to-covid-19-vaccines-for-all (last visited July 5, 2021) [11]EU proposes a strong multilateral trade response to the COVID-19 pandemic, European Commission, June 21, 2021, https://trade.ec.europa.eu/doclib/press/index.cfm?id=2272 (last visited July 5, 2021) [12]Drugmakers say Biden misguided over vaccine patent waiver, Reuters, May 6, 2021, https://www.reuters.com/business/healthcare-pharmaceuticals/pharmaceutical-association-says-biden-move-covid-19-vaccine-patent-wrong-answer-2021-05-05/ (last visited July 5, 2021) [13]J&J's Chief Patent Atty Says COVID IP Waiver Won't Work, Law360, Apr. 22, 2021, https://www.law360.com/ip/articles/1375715?utm_source=rss&utm_medium=rss&utm_campaign=section (last visited July 5, 2021) [14]Pfizer CEO opposes U.S. call to waive Covid vaccine patents, cites manufacturing and safety issues, CNBC, May 7, 2021, https://www.cnbc.com/2021/05/07/pfizer-ceo-biden-backed-covid-vaccine-patent-waiver-will-cause-problems.html (last visited July 5, 2021) [15]Moderna CEO says he's not losing any sleep over Biden's support for COVID-19 vaccine waiver, Fierce Pharma, May 6, 2021, https://www.fiercepharma.com/pharma/moderna-ceo-says-he-s-not-losing-any-sleep-over-biden-s-endorsement-for-covid-19-ip-waiver (last visited July 5, 2021) [16]Open Covid Pledge. https://opencovidpledge.org/ (last visited July 7, 2021) [17]Statement by Moderna on Intellectual Property Matters during the COVID-19 Pandemic, Moderna, Oct. 8, 2020, https://investors.modernatx.com/news-releases/news-release-details/statement-moderna-intellectual-property-matters-during-covid-19 (last visited July 7, 2021)
Antitrust Liability to the Conduct of “Refusal to License” of the Standard Essential PatentAntitrust Liability to the Conduct of “Refusal to License” of the Standard Essential Patent 2022/07/19 The notion of Standard Essential Patent(SEP)emerges in the era when manufacturers seek ‘‘compatibility’’ and ‘‘interoperability’’ of their products. The concept of SEPs is proposed to help manufacturers ‘‘talk’’ to each other so the collective manufacturers enjoy the advantage of economies of scales. Meanwhile, the compatibility and interoperability derived from SEPs enhance the consumers’ valuation of the product which creates the ‘‘network effect’’ of the products. There is a long-debated issue in the field of SEP—to what extent shall the SEP holders license their patents in the various level of the supply chain. This issue has much to do with the ‘‘FRAND commitment’’, and is worthy of further analysis. I. SEP and FRAND Commitment The concept of SEP is—when any certain patented technology is selected by the ‘‘Standard Setting Organization’’(SSO)as the commonly used standard, such the patented technology is categorized as a SEP. The SEP holder therefore enjoys stronger ‘‘market power’’ because market participants have no choice but to use the SEP and are required to seek license from the SEP holders. Therefore, to prevent the SEP holders from abusing their market power, SSOs usually require SEP holders to make the FRAND commitment; that is, to license on ‘‘fair, reasonable and non-discriminatory’’ terms. Once the SEP holder breaches the commitment, the SSOs might exclude that technique from the standard. II. “License to all”or“Access to all”issues under FRAND Commitment The FRAND commitment, by textual reading incorporates the wording of ‘‘non-discriminatory’’, and can infer two co-related yet debatable concepts—the ‘‘License to all’’ or ‘‘Access to all’’ arguments. The ‘‘License to all’’ argument holds that all participants in the supply chain retain the access to the specified SEP, while the ‘‘Access to all’’ argument, on the contrary, contends that FRAND commitments don’t necessarily ask SEP holder to license to all practitioners, but when a SEP holder is going to license, he must license on FRAND terms. According to observations, there is a common phenomenon in the SEP licensing practice—most SEP holders tend to license only to the End-Product manufacturers rather than to the manufacturers of the ‘‘Smallest Saleable Patent Practicing Unit’’(SSPPU). What the SEP holders expect through ‘‘refusal to license’’ to the SSPPU manufacturers are to maximize the potential royalties. Cases inclusive of the Qualcomm case[1] and the Continental case[2] have shown such practical tendency, and only when the SSOs can well define the definitions of FRAND commitments might the issue be truly settled. There are some End-Product manufacturers that consider it ‘‘discriminatory’’ and against the FRAND commitments if the SEP holders refuse to negotiate with SSPPU manufacturers requesting to be the licensee. On the other hand, some consider it inappropriate for the End-Product manufacturers to refuse all negotiations when the SEP holder requests it to be the party to the licensing negotiations[3]. III. The ‘‘refusal to license’’ and the derived Anti-Trust Issue As generally admitted, a firm has no general duty to deal with others[4]; however, there are times when SEP holders’ ‘‘refusal to deal∕license’’ behaviors can constitute wrongful monopoly under Sherman Act section 2. The U.S. judicial practices have categorized three main ‘‘refusal to deal∕license’’ behaviors as wrongful monopoly under Sherman Act section 2; they are[5]: 1.dominant firm forces its customers not to do business with new competitors of that firm, or the dominant firm will terminate business with the customer[6]; 2.dominant firm tries to abandon or alter an existing relationship[7]; 3.dominant firm refuses to provide access to ‘‘essential facility’’ (the equipment or techniques that is indispensable when others would like to compete in the relevant market with the dominant firm). As SEP can be categorized as an ‘‘essential facility’’, this paper will only focus on the third category. The ‘‘Essential Facility Doctrine’’ is—when any monopolist withholds an essential facility and refuses to provide his competitors with the access to the said essential facility, a wrongful monopoly due to the Facility holders’ ‘‘refusal to deal∕license’’ is constituted. According to the leading case—the MCI case[8], four factors are to be proved by the plaintiff when seeking resort to ‘‘Essential Facility Doctrine’’; they are:(1)the monopolist’s control of an essential facility;(2)the inability of a competitor to duplicate that essential facility;(3)the monopolist’s denial of access to that essential facility to a competitor;(4)the feasibility of providing the essential facility to the competitor by the monopolist. As we can shortly conclude here, if a SEP holder constitute wrongful monopoly because of his ‘‘refusal to license’’ behavior, the perquisite is that the SEP holder would like to join in the ‘‘competition’’ in the relevant market himself. IV. Conclusion—the commonly seen ‘‘refusal to license’’ behavior of SEP holders doesn’t constitute wrongful monopoly As mentioned before, ‘‘competition’’ serves as the prerequisite for the ‘‘Essential Facility Doctrine’’; thus, some SEP holders’ refusal to license to SSPPU manufacturers behaviors—such as Qualcomm in the Qualcomm case and Nokia in the Continental case—are not in accordance with ‘‘Essential Facility Doctrine’’ and do not constitute wrongful monopoly. Qualcomm and Nokia chose not to license to SSPPU manufacturers merely because they want to earn more royalties by licensing to End-Product manufacturers; they didn’t make this choice because themselves would like to compete in the SSPPU markets. However, since there is no clear definition of FRAND yet, whether the SEP holders have truly breached the FRAND commitment remains unsolved puzzle and shall retain to SSO’s clearer definition and the Court’s further rulings. [1]FTC v. Qualcomm Inc., 969 F.3d 974 (9th Cir. 2020). SEP holder Qualcomm would only like to license to the cellphone OEM manufactures rather than to other chips manufacturers. [2]Continental Automotive Systems, Inc. v. Avanci, LLC, et al, No. 20-11032 (5th Cir. 2022). SEP holder Nokia and a licensing platform—Avanci (that Nokia had joined) would only like to license to car manufacturers rather than to Telematics Control Unit(TCU)manufacturers. [3]Japan Patent Office [JPO], GUIDE TO LICENSING NEGOTIATIONS INVOLVING STANDARD ESSENTIAL PATENTS (2018), https://www.jpo.go.jp/e/support/general/sep_portal/document/index/guide-seps-en.pdf(last visited July 19, 2022). [4]See United States v. Colgate & Co., 250 U.S. 300 (1919);Pacific Bell Telephone Co. v. linkLine Communications, Inc., 555 U.S. 438 (2009); Aerotec Int'l v. Honeywell Int'l, 836 F.3d 1171 (9th Cir. 2016) [5]ANDREW I. GAVIL, WILLIAM E. KOVACIC & JONATHAN B. BAKER, ANTITRUST LAW IN PERSPECTIVE: CASES, CONCEPTS AND PROBLEMS IN COMPETITION POLICY 630-654 (2002). [6]See Lorain Journal Co. v. United States, 342 U.S. 143 (1951) [7]See Image Technical Services, Inc. v. Eastman Kodak Co., 504 U.S. 451 (1992); Aspen Skiing Co. v. Aspen Highlands Skiing Corp., 472 U.S. 585 (1985) [8]MCI Communications Corp. v. American Tel. & Tel. Co., 708 F.3d 1081 (7th Cir. 1983)
Copyright Ownership for Outputs by Artificial IntelligenceCopyright Ownership for Outputs by Artificial Intelligence One. Introduction I. From Machine Learning to Deep Learning, AI is Thinking The famous philosopher, mathematician and physicist René Descartes from France in the 17th century said: “Cogito ergo sum”. This is considered a radical skepticism in the context of philosophy. When a philosopher raises the question that how one person can be sure of his/her existence, it is not about the feeling, cognition or experience with the world. Rather, it is about thinking. Artificial intelligence works like interconnected human neurons, with the logics and algorithms built with codes and processed with high speed. The nutrient it requires is the massive amount of data. In the past, artificial intelligence only works according to the logical setup and instructions from developers. In the era of machine learning today, humans have empowered machines with the capability of processing. This is achieved not by writing comprehensive and exhaustive rules. Rather, it is by making machines able to figure out rules on their own. In other words, all we need to do is to prepare data. Machines can be trained to think and judge. Artificial intelligence will eventually generate its outputs and start to create contents. Image recognition is a good illustration of how machine learning works, as part of the wider AI. The identification of cats is a classic example. A large number of pictures and photos of cats are provided, with descriptions of features to train machines. The purpose is to train machines into building their own criteria as to what cats are about. According to the Proceedings of the Seventh IEEE International Conference on Computer Vision in 1999, image recognition is processed with the technology similar with neurons for visual recognition by primates[1]. Twenty years on, machine learning (as part of artificial intelligence) has come a long way. The number of neural network models, built on neurons, has grown exponentially[2]. Deep learning has been developed with layers of neurons. There are links only between neighbouring layers to reduce the number of variables and enhance the speed of computing. In the context of machine learning, learning is about the selection of an optimal solution from multiple variables[3]. Big data is fed into the man-made neural networks constructed in the computers so that they are constantly trained and learning. Hung-yi Lee[4], a scholar specialized in artificial intelligence in Taiwan, provides a simple analogy for this technology. Machine learning is like a human brain with one layer of neurons; whilst deep learning works with many neurons and hence can learn on their own, make judgement and establish logics[5]. In other words, artificial intelligence is capable of analysing, identifying and decision-making on its own, and human is becoming less relevant in this process. Artificial intelligence is able to think. This is not only a factual description, but also a trigger to fundamentally change the legal institution of nations. II. Who Owns the Outputs Generated with Thinking? Over the long run, whether the legal institution and the society are ready to give artificial intelligence “quasi” right of personality is a topic worth exploring. In the immediate term, what normative models should be used to define the ownership of copyrights for the outputs and creations by artificial intelligence? The decision on copyright ownership has always been a hot topic in the field of intellectual property. The legal system in the U.S. describes the protected entity for copyright as “the fruits of the intellectual labor”. Article 798 of the Civil Code in Taiwan says, “Fruits that fall naturally on an adjacent land are deemed to belong to the owner of such land, except if it is a land for public use”. The fruit, i.e. outputs generated by artificial intelligence, also falls into the society of rules governed by rights and obligations. Of course, it is necessary to first define and regulate the entity that owns the rights. This begs many fundamental questions in the context of copyright laws. Who owns the rights? The developers (perhaps on a pro-rata basis), data owners, or the companies that provide infrastructure to developers? Once the boundary of imagination and reality is pushed further, the ownership of rights is no longer limited to human creators and may be extended to artificial intelligence. Moreover, it is possible for governments to insist that copyrights are only for human creations and the intellectual property created by artificial intelligence may fall into the public domain and hence fall unprotected legally, given the significance of public interest involved. This paper explores the copyright ownership for the outputs generated by artificial intelligence by systemically observing the real-life cases in the industry. This is followed with an analysis on the perspectives from the European Union, the United Kingdom and the United States. The purpose is to examine the contexts and normative models of artificial intelligence and copyrights and finally develop a preliminary framework for the regulation of artificial intelligence now and the future. Two. Creativity Capability of Artificial Intelligence Is a Reality With artificial intelligence and Big Data driving the development of industries, the exploration with the construction and normative models of the legal system should start with the reflection of social values, so as to achieve the purpose of social order with laws and regulations. The construction of the legal system for technology should be anchored on the observation of facts, given the rapid advancement and evolution of emerging technologies. The fact today is that artificial intelligence is being used for art creations such as musical composition, poetry and painting. Developers train artificial intelligence with massive data and enable deep learning to grasp the essence of artworks in order to generate outputs. Whether the ultimate purpose is commercial profitability or not, most of these outputs have reached a certain level of quality. Below is a brief introduction of creative techniques and new business models of artificial intelligence in music composition, poetry writing, painting and news writing. I. Original Music Generated with Deep Learning: Fast and User-friendly The vibrant development of the Internet has created an online celebrity economy. Youtubers, Internet personalities, cyberstars, Wanghong (or internet fame in Mandarin) produce films or release podcasts to attract the audience for direct/indirect and commercial/non-profit-seeking purposes. The production of such films and live broadcasting, or the creation of original online or PC games creates the demand for background music or sound effects. Ed Newton-Rex, who earned a bachelor of arts degree in music from University of Cambridge, founded JukeDeck[6] after he went to a computer science class in Harvard University. JukeDeck is an online music generator, developed with deep learning(as part of artificial intelligence). This paper believes that JukeDeck meets the industry demand with two offerings[7]: (I) JukeDeck Rapid generation of pleasant and unique music with deep learning The algorithm design by Ed Newton-Rex with artificial intelligence is different from the generation of background music and other music by the websites that use loop audio files. JukeDeck generates music pleasing to the ears with one tone at a time and avoids repetitions by analyzing musical forms, harmonies and tones with deep learning, so that the users in pursuit of originality and unique can acquire the musical materials within approximately 30 seconds, without worrying that they sound similar with others[8]. Greater flexibility in length to create bespoke styles and feelings JukeDeck offers flexibility in the length of music, up to five minutes depending on the preference of users. An extension is possible by mixing up different fragments. It is also possible to define musical styles and formats, e.g. piano, folksongs, electro and ambient music[9], as well as the feelings to be aroused, such as uplifting and melancholic. The music generated by deep learning is different from the free or paid music databases which use the so-called canned music and suffer the problems of mismatches between the film length and music length[10]. (II) Amper Music Amper Music was founded by the Hollywood songwriter Drew Silverstein (founder/CEO), Sam Estes and Michael Hobe[11] with the ambition to take a step further from music generation by artificial intelligence. In the spring of 2018, the company raised another $4 million for the development of music composition with artificial intelligence, the expansion of international markets and the recruitment of more talents. In the press release, Drew Silverstein said, “Amper’s rapid growth is a testament to how the massive growth of media requires a technological solution for music creation. Amper’s value stems not only from the means to collaborate and create music through AI, but also from its ability to help power media at a global scale.”[12] Similar with JukeDeck’s appeal to the public, Amper Music’s artificial intelligence allows users with no musical experience to create real-time and order original music[13]. It supports all the media formats. All is required is the choice for rhythms, styles and musical instruments desired[14]. Meanwhile, Amper Music posits that its music is royalty free, and comes with a global, perpetual license when synced to the outputs. In other words, users do not have to worry about legal procedures or financial costs[15]. II. Writing Pens Take Flight: A Challenge to the Fundamental of Literary Creation and Trigger for Labor Transformation Neuhumanismus (or Neohumanism) is about the achievement of self-mastery and humanity ideals through the study of classics. Compared with humanism, neohumanism places a greater focus on emotional expression and artistic creation. It also emphasizes the importance of language learning to self-realization of individuals.[16] After studying the works of 519 contemporary poets in the Chinese society, artificial intelligence has published modern poetry and made successful inroads to the world of literature traditionally driven by emotions and imaginations. In fact, it has posed a credible challenge to the human-centric humanism where only humans are endowed with the gift of artistic creativity. Artificial intelligence has been nominated for literary awards, evidenced of the quality of outputs generated by deep learning. With the support of massive data and analytics, it is only a matter of time for artificial intelligence to possess the literary creativity comparable to humans. However, the concern for originality in literature and the issues surrounding plagiarism and copyrights are the key determinants that influence of literary creation by artificial intelligence. This begs the questions about the ethics of literary creation. It is necessary to start with an understanding of how artificial intelligence creates, before the analysis of ethical and regulatory frameworks. (I) Xiaoice’s Collection “Sunshine Misses Windows” Xiaoice is the chatbot launched by Microsoft’s Software Technology Center Asia (STCA) in China in 2014. In 2017, Xiaoice published her collection of poems “Sunshine Misses Windows”[17], written by looking at pictures. The deep learning algorithms behind were co- developed by Wu Zhao-Zhong and Cheng Wen-Feng, two students in the Graduate Institute of Networking and Multimedia, National Taiwan University. The artificial intelligence writes poetry with the following methodology[18]: Use of image recognition technology to identify the keywords in the pictures: The adoption of image recognition technology developed by Microsoft’s Software Technology Center Asia (STCA) to identify the nouns in the pictures such as the bridge, skies and trees and the adjectives that express feelings such as beautiful or annoying. Matching of keywords from the training database: The training data for the matching of keywords and poetry database was the works of a total of 519 contemporary poets since the 1920s. The purpose was to fill in the gap between keywords and training data. Generation of poems: deep learning trained in the language model with keywords to create poems Improvement of poems: literary professionals and readers invited to give ratings. Submission of writings as an anonymous author to improve Xiaoice’s capability. The above is a summary of Xiaoice’s creative journey. Microsoft claims that the collection of poems was 100% written by Xiaoice, and it is the first collection of poems 100% written by artificial intelligence in history. The poems were not edited by humans and wrong characters were maintained as they were. The title “Sunshine Misses Windows” was also named by Xiaoice herself[19]. Despite all these, the originality and even the most fundamental “literality” of these poems are still questioned. At the end of 2018, the Research Institute for Humanities and Social Sciences, Ministry of Science & Technology and National Taiwan University organized the forum “Culture and Technology II: AI’s Literature Dream — Sunshine Misses Windows. Does Humanity Have a Boundary?” The professor in the Department of Chinese Literature, National Taiwan University and the poet Tang Juan discussed Xiaoice’s works[20] and commented as a critic of contemporary poetry. Xiaoice uses extensively the same vocabulary (such as the beach). Unable to use punctures, she can only break sentences and lines. Most importantly, her writings do not reflect our times and real experience. In other words, Xiaoice’s poems do not possess the unique perspective and soul of poets and literary characters. This may be the outcome of her reading of works from 519 poets from the 1920s. As a result, she is not able to connect with our times and real life and finds it difficult to resonate the shared emotions of people today. Tang Juan’s comment is more than just about literature. It is also about the selection and sourcing of training data, a prerequisite for the development of artificial intelligence, as well as the cost and consideration for copyright licensing. The research and development by corporates in artificial intelligence requires the corresponding and suitable training materials, particularly in the domain of literature. As commented by the poet Tang Juan, it requires extensive sources of contemporary works. It means the increasing difficulty to circumvent the works still protected by copyrights. If this cost consideration remains a hurdle, it is impossible to make improvements in further research. Put differently, the composition of training data is potentially a cost concern for copyright licensing. Before the legal system becomes well-developed and the establishment of consensus on the issues concerning training data, the possible infringement is an absolutely necessary balancing act for any robust developers and companies involved in artificial intelligence. (II) Yuurei Raita’s “The Day A Computer Writes A Novel” In 2013, Nikkei started to offer the Nikkei Hoshi Shinichi Literary Award to outstanding short Si-Fi novels, as a tribute to the late science fiction writer Hoshi Shinichi[21]. Three years later, Yuurei Raita’s “The Day A Computer Writes A Novel” appeared on Nikkei’s list of acceptance for competition. Miss Yoko is the leading character in this 2000-character short sci-fi novel[22]. Raita-kun is in fact an artificial intelligence team “Wagamama artificial intelligence as a writer” led by Hitoshi Matsubara, President of the Japanese Society of Artificial Intelligence and a professor in Future University[23]. Below is a description of their deep learning techniques[24]: Analysis of writing styles from training data: The team provides training data as the learning basis for artificial intelligence. (For this competition, the data is approximately 1,000 short stories written by Hoshi Shinichi.) The purpose is to analyze the frequently used words, novel structures and characters. Resource integration by the team: The team integrates the analyzed data with online information, storyline programs, human emotions and settings, and decides on characters, contents and plots[25]. Researchers provide three instructions, i.e., when, the weather, doing what so that artificial intelligence automatically generates detailed and tangible contents. Automatic generation of new works: Artificial intelligence refines the details and polishes the texts, to generate the new story by Hoshi Shinichi with fragments such as: “The same temperature and humidity in the room is maintained as usual. Yoko sits idly on the sofa, dishevelled and playing a dull game uninterested.” The procedures of novel contents generation described above indicate that artificial intelligence still relies on humans for setups and assistance. In contrast with the claim by the Microsoft team that Xiaoice is 100% artificial intelligence, the team in Japan confessed that artificial intelligence writing is still in a nascent stage. At least in literature types such as novels, artificial intelligence still needs appropriate guidance from humans for necessary writing elements, in order to generate and connect fragments to establish the finalized pieces. In general, artificial intelligence can only be held responsible for 20% of work[26]. However, the development of technology continues at its pace. When it is no longer easy to differentiate a piece of creative writing is by humans or by machines, the limitation of copyright protection to human’s creative works will be an obsolete approach. (III) Tencent: Robot “Dreamwriter” The above two AI writing teams focus on creative literature. In China, Tencent has developed Dreamwriter to rapidly generate news products. In the 2018 International Media Conference in Singapore[27] hosted by the East West Center, a think tank in the U.S. at the end of June 2018, Tencent demonstrated its translation engine. Speakers spoke in Chinese and the engine did simultaneous translation into English shown on the projector screen[28]. Tencent’s artificial intelligence “Dreamwriter” project started as a push engine for news flashes such as sports events. It later extended into financial and economic data and reporting, a field with extensive data and conducive to AI development and ML acceleration[29]. Dreamwriter only takes half to one second to generate a piece of news. It can generate approximately 5,000 articles per day, equivalent to the output of 208 journalists. This implies a transformation of labor requirements in journalism. Human reporters will be involved in in-depth coverage that requires creativity, industry knowledge and judgement[30], whilst basic and factual reporting will be completed by artificial intelligence. III. Brave New Work for Paintings: Rights Ownership in the Presence of Sophisticated Deep Learning In the autumn/winter of 2018, the Paris-based AI team Obvious presented “Portrait of Edmond Belamy”[31] in Prints & Multiples auction in New York. This painting was sold for a surprising high price of[32] $432,000 (or over NT$13 million)[33], as the first AI-generated painting being auctioned. The Obvious team focuses on Generative Adversarial Network (GAN)[34], a hot topic for the development of deep learning. (I) Technique to Improve Deep Learning: Generative Adversarial Network (GAN) The GAN technique was developed by Ian Goodfellow[35] in 2014 to promote and enhance deep learning by massively reducing the amount of training data required and cutting down on human intervention, assistance and involvement[36]. The GAN method can be illustrated in a high level by referring to the classical example of the image recognition for cats previously mentioned. The neural network model (as a deep learning technique) enables artificial intelligence to learn how to identify cats from a massive volume of pictures of cats. However, it is necessary for humans to train the machine by providing signs and feature descriptions for each picture. In contrast, the GAN technique is about the training of two competing networks,[37] i.e., a generative network and a discriminant network[38]. The generative network is responsible for generating the pictures that resemble real cats (i.e. made-believe cats) and the discriminant network reviews and determine whether the pictures are authentic. The two networks enhance capabilities by competing with each other. The idea is to improve the learning and competence of deep learning[39]. (II) Application in the Art of Paintings The GAN method can be used to generate paintings such as “Portrait of Edmond Belamy”. It can also identify fake paintings. Founder/CEO Jensen Huang of Nvidia, a leading artificial intelligence company, said in a forum that the GAN technique allows one neural network to paint the pictures in the Picasso style and the other network to identify images and paintings with unprecedented discriminant capabilities[40]. The seventh year of the Lumen Prize gave the biggest award to a nude portrait generated with the GAN technique[41]. The GAN applications have been mushrooming – turning a scribble into an art, a low-definition picture into a high-definition one, an aerial graph into a photo[42]. Below is a brief description of the concepts and procedures for the Obvious research team’s completion of “Portrait of Edmond Belamy”[43]: Analysis of portraits from training data: A total of 15,000 portraits from the 14th century to the 20th century as the training data Generative network vs. discriminant network: The generative network generates paintings on the basis of training data. The discriminant network seeks to identity the difference from human-created paintings in order to improve the capability of the generative network. This process continues until the discriminant network is no longer to tell a machine-created painting from a human-created painting. (III) Ownership of Rights to High Economic Value of Artworks The winning of the Lumen Prize in the UK by the nude portrait generated by artificial intelligence and the surprisingly high auction price paid for Portrait of Edmond Belamy are the testimony of the artificial intelligence’s creative capability. The ownership of the right to the monetary value of these artworks is a topic worthy of exploration. “The development team ‘Obvious’ for ‘Portrait of Edmond Belamy’ posits that if the author is the person who paints the painting, it is artificial intelligence. If the author is the person who seeks to convey a message, it is us[44]. The human’s role is being undermined as deep learning technology becomes increasingly sophisticated. Going forward, can artificial intelligence become the owner of rights? What should be the regulatory framework for now? At this juncture, this paper conducts an international comparison by examining how different governments consider the emerging legal issues. Three. Copyright Ownership of Works Created by Artificial Intelligence The explanatory ruling by the Copyright Division, Intellectual Property Office, Ministry of Economic Affairs issued in 2018[45] has expressed the Taiwan government’s stance on the issue of whether the outputs generated by artificial intelligence can enjoy copyrights. Below is the summary: Presumption: Article 10 and Article 33 of the Copyright Law[46] stipulates that only natural persons or legal persons can be the owner of rights and obligations pertaining to creative works and enjoy the protection of copyrights. Positioning and logics: The outputs generated by artificial intelligence are the intellectual results expressed by machines created by humans. Machines are neither natural persons or legal persons and hence do not attract copyrights. Proviso: If the results are created with participation of natural/legal persons and the machines are being operated for analytics, the copyright of the results expressed should belong to the natural/legal persons concerned. The above explanatory ruling seems to position artificial intelligence completely as a tool. However, the above example suggests an obvious trajectory for the creative journey for deep learning as an artificial intelligence technique. In the current stage and the foreseeable future, the description that robot analytics are straight mechanical operations is completely obsolete given that artificial intelligence is being applied in industry with dramatically reduced (or even completely without) human intervention and participation. It is a worthwhile exercise to explore the international thinking regarding how the legal framework should address the ownership of rights for outputs generated with deep learning as an artificial intelligence technique and the derived services/products by either opening up new legal structures or simply extending on the existing system. I. European Union (I) European Parliament: Establishment of Electronic Personhood? The European Parliament's Committee on Legal Affairs (JURI) passed a report on January 12, 2018 to provide suggestions to the Civil Law Rules on Robotics and urge the European Commission to set up laws and regulations governing robots and artificial intelligence by defining electronic personhood, similar with legal personhood for corporates as litigation entities for any issues associated with rights and obligations of artificial intelligence[47]. (II) Court of Justice of the European Union: Only Works Accomplished by Humans Eligible for Protection The Court of Justice of the European Union’s landmark case Infopaq International A/S v. Danske Dagbaldes Forening[48]suggests that copyrights are only applicable for original works, with originality reflecting the “author’s own intellectual creation.” The general interpretation is that such works should reflect the author’s personality. Hence, only human authors meet this criterion[49]. The third paragraph of Article 1 of the Directive 2009/24/EC also clearly states that only works that are the authors’ own intellectual creation enjoys eligibility for protection[50]. (III) Data Protection: GDPR and Declaration of cooperation on Artificial Intelligence The General Data Protection Regulation (GDPR) in European Union attracted significant attention among the companies active in the EU market in 2018. In fact, the GDPR provides comprehensive and representative regulations that have direct influence on technological development of artificial intelligence training, as well as legal protection and right construction on data, the crude oil for deep learning. Below are a few examples: Article 20 on data portability: The data subject has the right to receive his/her personal data from the data controller in a structured, commonly used and machine-readable format. This helps the industry to establish metadata and forms the basis of the database for artificial intelligence training. The consistency of metadata will enhance the training. Article 22 on automated individual decision-making The data subject has the right not to be subject to a decision based solely on automated processing. The data controller must lay down suitable measures to safeguard the data subject’s rights. Article 35 on data protection by design and by default This article provides the legal protection of large-scale and systematic monitoring of public and open areas with artificial intelligence and strikes a balance between the use of personal data and the interest of data subjects. On top of the GDPR, the 24 member states of the European Union signed the Declaration of Cooperation on Artificial Intelligence in 2018, in order to enhance access to public sector data for the digital single market. II. United Kingdom (I) Copyright Law: Source of Laws for Program Developers to Obtain Copyrights The copyright laws are stipulated in the Copyright, Designs and Patent Act (CDPA) 9 (3)[51]. It forms the source of the laws that grant copyrights to the developers of computer-generated works. Article 178 of the CDPA defines computer-generated works as the outputs generated by machines without human authors[52]. In contrast with the Court of Justice of the European Union’s decision that only human authors are eligible for copyright protection, the UK government opens up another door by specifying that program designers can obtain copyrights even if creative sparks come from machines[53]. This system is considered the most efficient because it enhances incentives for investments[54]. (II) Public Sector: Open up Government Data The UK government also opens up its data by posting all the official statistics on the website www.data.gov.uk. The Digital Economy Bill provides the legal framework for government agencies to use each other’s data for the benefit of the public, so as to effectively address the issues surrounding frauds and debts and improve the real-timeliness and accuracy of national statistics. As part of the Brexit preparation, the UK government has created its own GDPR (2018) to ensure the continued smooth cross-border operations of companies after Brexit. As it offers higher protection of consumers’ data and information, it is worthwhile to refer to the UK GDPR as a template for legal systems and rights frameworks. III. United States (I) U.S. Copyright Office: Only Intellectual Achievements of the Human Mind Eligible for Protection The case law originated in 1991——Feist Publications v. Rural Telephone Service Company[55]confirms that copyrights protect the creative powers of the mind. In the Naruto v. Slater (2016)[56] case, the court determines that the photos taken by a monkey are not eligible for copyright protection. Article 313.2 of the implementation guidelines of the Copyright Act issued by the U.S. Copyright Office specify that the works created without human authors are not protected by the Copyright Act. The amendment to Article 313.2 in 2017 states clearly that the U.S. Copyright Act only protects the intellectual achievements of the human mind[57]. The U.S. Copyright Act 503.03(a), titled “Works-not originated by a human author” also states that only works created by a human author can register for copyrights[58]. (II) Employment Principle: Enhanced Incentives and Investment Willingness The above court judgements and the implementation guidelines of the U.S. Copyright Act indicate that the U.S. Copyright Office does not confer non-human copyright[59]. However, the U.S. judicial rulings have allowed “the work made for hire provision” as exception to the creative authors, in order to encourage corporate investments. The 1909 amendment to the U.S. Copyright Act included the hired employees as authors. Unless otherwise agreed, “the author or proprietor of any work made the subject of copyright by this Act, or his executors, administrators, or assigns, shall have copyright for such work under the conditions and for the terms specified in this Act”. A typical example is the news agency’s employment of full-time journalists to produce editorials. The works by employees are a company’s key copyright assets[60]. (III)Employment/Sponsorship Principle if Realized in Taiwan: Companies Investing in Works to Obtain Copyright Protection Article 11 of the R.O.C. Copyright Act stipulates the ownership of the right to the works of employees on a case-by-case and factual basis. The decision is based on the nature of work, e.g., completion under the employer’s instructions or planning, the use of the employer’s budgets or resources. It is not necessarily related to the work hours or locations. In principle, the employee is the author of the works completed by him/her on the job. However, the employment contract supersedes if it specifies that the employer is the author. On the other hand, if the employee is the author, the intellectual property belongs to the employer. The contract supersedes if it specifies that the employee enjoys the intellectual property. Article 12 is about sponsorship and commissioning. Unless specified by the contract, the sponsored owns the intellectual property of his/her works and the sponsor has the right to use such intellectual property[61]. In sum, the ownership of the right to the outputs generated by artificial intelligence is similar with the employment/sponsorship principle. It is not set in the vacuum of legal contexts. Therefore, the scholar in Taiwan Lin Li-Chih suggests that the employment principle in the U.S. may be adopted. She posits that when certain conditions are met, artificial intelligence may be treated as the author, so that the outputs generated by artificial intelligence can be protected and the investing research institutes or corporates can own the works[62]. As both legal persons and natural persons can be authors in Taiwan, Lin Li-Chih proposes this approach to resolve disputes given the massive value to be created by artificial intelligence for different applications and the potential lengthy legislative process or laws disconnected from industry expectations. The idea is to avoid the human author requirement from hindering industry investments and innovations for works generated by artificial intelligence[63]. According to the employment/sponsorship principle, deep learning as an artificial intelligence method can be inferred to as the author and then teams and companies that develop the algorithms should own the intellectual property of the works. This will serve as the legal foundation for intellectual property protection. Four. Conclusion: Legal System and Policy Framework for Emerging Technologies I. Construction of Laws and Regulations on a Rolling Basis According to the Reality of Emerging Technologies Every law has its purpose, and the contents of laws depend on their regulatory objectives. However, such contents should be anchored on facts, in order to align the intended purposes. This is particularly the case for the laws and regulations governing emerging technologies because such laws and regulations should capture the fact of technological developments. The most straightforward and fundamental approach to relax the control of the existing legal mechanism is via communication, coordination and understanding. It can be initiated with more dialogues between the government agency responsible for the construction of the legal environment and the industries and the public as subjects of the laws and regulations. Regulators may wish to come up with dedicated laws for the comprehensive coverage of emerging technologies given the lack of understanding about the technology and the sweeping effects of the technology. However, not all technologies require special legislations. According to Frank H Easterbrook’s article “Cyberspace and the Law of the Horse” published by the University of Chicago’s legal journal, it is advised to properly categorize and analyze existing laws and regulations and apply the suitable ones to new technologies for issues surrounding intellectual property, contracts and torts, as if from the Law of the Horse to the Law of Cyberspace[64]. Similarly, the ownership of copyrights associated with artificial intelligence and the governance of emerging technologies such as autonomous driving and robots may be dealt in this way. The above analysis on the legal regimes in the European Union, the United Kingdom and the United States highlights two issues concerning the regulation of artificial intelligence and the development of legal environments. The growing sophistication of deep learning will enhance the capability of artificial intelligence in thinking, analysis and creation, with human intervention expected to be reduced to almost zero. The legal regime governing emerging technologies cannot stand in the way of technological and industry development or incentives for investment, as originally intended by the intellectual property laws. A balancing act is required. This paper thus suggests two models: Forward-looking approach to label rights ownership with legal articles This is the route taken by the UK government, by directly amending the intellectual property laws to specify that intellectual property of artificial intelligence belongs to program developers. It is the most efficient approach of paving the way for technological development by providing incentives to companies and developers. Adoption of the employment/sponsorship principle in conjunction with safe harbor clauses Another approach is without touching on the sensitive issue of law amendments. Judicial rulings or administrative interpretations by competent authorities are gradually released in the context of existing laws. A temporary solution is introduced with the adoption of the employment/sponsorship principle with corresponding templates and references for contract construction in the industry. This can work in conjunction with safe harbor clauses in the long run, by slowly converging the diversity of opinions and perspectives from corporates, government agencies and academic/research institutions. Adjustments by tightening or loosening on a rolling basis should be made in order to work out the optimal boundary and establish the basis for legislation in the next stage. II. Data as a Prerequisite for Artificial Intelligence Training In Taiwan where the legal environment is not yet ready or clear, the ownership of intellectual property for outputs generated by artificial intelligence also involves the potential licensing royalties for the sourcing of training data. It is worth noting that the use of data for artificial intelligence may affect the basic human rights due to discrimination or bias resultant from training data or algorithms. Therefore, it is necessary to enhance transparency and the protection of human rights conferred by the constitution with corresponding legal systems and ethical frameworks such as due process and fairness principle[65]. The other critical issue is the training database required for artificial intelligence applications. The government should provide more open data as a policy to support technology development in the corporate world or at research organizations. It is also necessary to make government information the structured metadata in order to enhance the efficiency and quality of research outputs. This is to facilitate added value by private sectors with data as an infrastructure provided by the government. Put differently, the government opens up structured data to empower the research and development of artificial intelligence; whilst the private sectors offer professional technology and development capabilities. In terms of promoting data openness and applications, the government assumes greater accountability in the balancing between data use and data protection, the two equally important public interests. As an island of technology, Taiwan should look beyond the horizon of skies and oceans in the era where information and data flows without borders. The Taiwan government should establish the capability in data openness, protection and control by joining international forums. For instance, the government can apply with the APEC to join the Cross-Border Privacy Rules System in order to encourage regional collaborations in data control and construct datasets with the resources of the country. It is important to focus on the process of data collection, processing, analysis and utilization and ensure policies are implemented with the protection of civil and human rights such as the Right to Know, the Right to Withdraw and Citizen Data Empowerment. [1] David G. Lowe, Object Recognition from Local Scale-Invariant Features, Proceedings of the Seventh IEEE International Conference on Computer Vision, https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=790410 (last visited Dec. 27, 2018) excerpt from “These features share similar properties with neurons in inferior temporal cortex that are used for object recognition in primate vision”. [2] AI Lesson 101: Illustration of 27 Neural Network Models, Tech Orange, January 24, 2018, https://buzzorange.com/techorange/2018/01/24/neural-networks-compare/ (last visited on December 27, 2018) [3] Chen Yi-Ting (Bachelor’s Degree from Department of Physics, National Taiwan University, currently a PhD candidate in Department of Applied Physics, University of Stanford), Artificial Intelligence Starts with Neurons, May 3, 2018, https://case.ntu.edu.tw/blog/?p=30715 (last visited on December 27, 2018) [4] Hung-yi Lee’s personal profile at http://speech.ee.ntu.edu.tw/~tlkagk/. Currently teaching in Department of Electric Engineering, National Taiwan University; previously a guest scientist in MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL); specialization in machine learning and deep learning [5] Chen Yan-Cheng, Who Is Likely to Lose Jobs in the Era of Artificial Intelligence? Experts Explains the Professional Skills in Demand for Deep Learning, December 26, 2018. https://www.managertoday.com.tw/articles/view/56859 (last visited on December 27, 2018) [6] Details available on JukeDeck’s official website at https://www.jukedeck.com/(last visited on January 11, 2019) [7] In addition to the leverage of two key features of artificial intelligence, JukeDeck is also very friendly to creative teams in need of musical materials in terms of royalties, fee structures, UI/UX design. The company offers free downloads to non-commercial users. An individual or a small group (of fewer than 10 people) can enjoy five free downloads each month and pay $6.99 per song for the sixth download and above. Large groups (of ten people or more) should pay $21.99 for each download. [8] DIGILOG Authors, “A Nightmare for Musicians? AI Online Music Composer System – JukeDeck, DIGILOG, June 2, 2016, https://digilog.tw/posts/668 (last visited on January 2, 2019) [9] Laird Studio, Let the Online Music Composer Jukedeck Produce Unique Background Music for Your Films or Games! March 8, 2016, https://www.laird.tw/2016/03/jukedeck-jukedeck-bgm.html (last visited on January 10, 2019) [10] As above. [11] Amper Music’s official website at https://www.ampermusic.com/(last visited on January 10, 2019) [12] GlobeNewswire, Amper Music Raises $4M to Fuel Growth of Artificial Intelligence Music Composition Technology, March 22, 2018, https://globenewswire.com/news-release/2018/03/22/1444796/0/zh-hant/Amper-Music%E7%B1%8C%E8%B3%87400%E8%90%AC%E7%BE%8E%E5%85%83%E4%BB%A5%E6%8E%A8%E5%8B%95%E4%BA%BA%E5%B7%A5%E6%99%BA%E8%83%BD%E7%B7%A8%E6%9B%B2%E6%8A%80%E8%A1%93%E7%9A%84%E7%99%BC%E5%B1%95.html (last visited on January 10, 2019). This round was led by Horizons Ventures, with Two Sigma Ventures, Advancit Capital, Foundry Group and Kiwi Venture Partners. This brings the company's total investment to $9 million. [13] GlobeNewswire, same as above [14] Smart Piece of Wood, Free Online Composer Enabled by AI, Amper Music, March 1, 2017, Modern Musician,https://modernmusician.com/forums/index.php?threads/%E5%85%8D%E8%B2%BB%E7%B7%9A%E4%B8%8A%E5%B9%AB%E4%BD%A0%E4%BD%9C-%E7%B7%A8%E6%9B%B2%E7%9A%84%E4%BA%BA%E5%B7%A5%E6%99%BA%E8%83%BD%EF%BC%9Aamper-music.225650/ (last visited on January 10, 2019) [15] GlobeNewswire, same as Note 12 [16] Fang Yung-Chuan, Neohumanism, National Academy for Educational Research, http://terms.naer.edu.tw/detail/1312151/(last visited on January 10, 2019). Neohumanism emerged in Europe in the 18th and 19th century, against rationalism and utilitarianism advocated by the enlightenment movement. Neohumanism argues that the value of things is not hinged on practicality. Rather, it stems from the things themselves. Humanity is precious not because of rationality, but resultant from emotional satisfaction in life. Cultures are originated by the spontaneous activities of humanity, on the basis of emotions and imaginations. [17] Synopsis by books.com.tw, who sells online Xiaoice’s “Sunshine Misses Windows”, the first collection of poems generated by artificial intelligence in history, August 1, 2017, China Times Publishing Co. https://www.books.com.tw/products/0010759209 (last visited on January 13, 2019) [18] Wong Shu-Ting, AI Talents in Taiwan Find Stage in China: NTU Students Participate in R&D That Empowers Microsoft’s Xiaoice to Write Poetry by Looking at Pictures, BusinessNext, June 6, 2017, https://www.bnext.com.tw/article/44784/ai-xiaoice-microsoft(last visited on January 10, 2019 [19] Synopsis by books.com.tw, same as Note 17 [20] The organizer did not provide handouts from the speakers. The summary was based on the author’s note. [21]Lin Ke-Hung, “More Than Playing Chess. AI Writes Novels Too. AI Novel Passes Preliminary Screening for a Novel Award! Reading at Frontline, https://news.readmoo.com/2016/03/25/ai-fictions/(last visited on January 10, 2019) [22] Ou Tzu-Jin, “2,3,5,7,11..?AI-written Novel in Japan Nominated for a Literary Award, April 7, 2016, The News Lens , https://www.thenewslens.com/article/38783(last visited on January 10, 2019) [23] TechBang, AI Team in Japan Develops Robots That Write Short Stories and Participates in Literary Competitions, TechNews, March 28, 2016, http://technews.tw/2016/03/28/ai-robot-novel-creation/(last visited on January 10, 2019) [24] Ou Tzu-Jin, same as Note 20 [25] TechBang, same as Note 21 [26] Lin Ke-Hung, same as Note 19 [27] The title of the forum was “What is News Now?”. It attracted over 300 journalists and media experts from the U.S. and Asia Pacific to discuss media phenomena today. Detailed agenda available at East West Centre’s official website at https://www.eastwestcenter.org/events/2018-international-media-conference-in-singapore(last visited on January 10, 2019) [28] Jason Liu, “Robot Writer, Transformation of South China Morning Post, State Monitoring, International Media Conference Day 1, China, Medium, June 25, 2018, https://medium.com/@chihhsin.liu/%E5%AF%AB%E7%A8%BF%E6%A9%9F%E5%99%A8%E4%BA%BA-%E5%8D%97%E8%8F%AF%E6%97%A9%E5%A0%B1%E8%BD%89%E5%9E%8B-%E5%9C%8B%E5%AE%B6%E7%9B%A3%E6%8E%A7-%E5%9C%8B%E9%9A%9B%E5%AA%92%E9%AB%94%E6%9C%83%E8%AD%B0day1-%E4%B8%AD%E5%9C%8B-c9c20bd00d75(last visited on January 10, 2019) [29] Jason Liu, same as above [30] Jason Liu, same as above [31] First Time Ever in the World!AI-Created Portrait, Sold at Christie's Auction for NT$13.34 Million, Liberty Times, October 26, 2018, http://news.ltn.com.tw/news/world/breakingnews/2592633(last visited on January 10, 2019) [32] The selling price is 40x higher than the expected price. The buyer’s identity is unknown. Chang Cheng-Yu, “First Time Ever! AI-Created Portrait Auctioned at Christie’s for NT$13.34 Million, October 26, 2018, LimitlessIQ,https://www.limitlessiq.com/news/post/view/id/7241/ (last visited on January 10, 2019) [33] Lin Pei-Yin, Does the NT$10m Worth AI Portrait Have Intellectual Property?” Apple Daily, Real-Time Forum, November 29, 2018, https://tw.appledaily.com/new/realtime/20181129/1475302/(last visited on January 10, 2019) [34] Jamie Beckett, What Are Generative and Discriminant Networks? Hear What Top Researchers Say, Nvidia, May 17, 2017, https://blogs.nvidia.com.tw/2017/05/generative-adversarial-network/(last visited on January 10, 2019) [35] Jamie Beckett, same as above. Ian Goodfellow is currently a Google research scientist. He was a PhD candidate in the Université de Montréal when he came up with the idea of generative adversarial networks (GAN). [36] Jamie Beckett, same as above [37] Jamie Beckett, same as above [38] Chang Cheng-Yu, same as Note 32 [39] Jamie Beckett, same as Note 34 [40] Video for the speech: GTC 2017: Big Bang of Modern AI (NVIDIA keynote part 4), link at https://www.youtube.com/watch?v=xQVWEmCvzoQ (last visited on January 10, 2019) [41] Wu Chia-Zhen, AI-Generated Nude Portrait Beats Real People’s Works by Claiming the UK Art Award and Prize of NT$120,000, LimitlessIQ, October 15, 2018 https://www.limitlessiq.com/news/post/view/id/7070/(last visited on January 10, 2019) [42] Jamie Beckett, same as Note 34 [43] Chang Cheng-Yu, same as Note 32 [44] Chang Cheng-Yu, same as Note 32 [45] The explanatory ruling by the Copyright Division, Intellectual Property Office, Ministry of Economic Affairs, Email 1070420, issued on April 20, 2018, https://www.tipo.gov.tw/ct.asp?ctNode=7448&mp=1&xItem=666643(last visited on January 2, 2019). The discussion was in response to the training outcome of voice recognition patterns based on analytics of the 1999 Citizen Hotline voice data. [46] According to Article 10 of the Copyright Law, authors enjoy copyright at the time of the work completion. Article 33 stipulates that copyright for legal-person authors lasts 50 years after the first publication of the work concerned. [47] Yeh Yun-Ching, Birth of New Type of Legal Right/Liability Entity ─ Possibility of Robots Owning Copyrights According to 2017 Proposal from European Parliament, IP Observer - Patent & Trademark News from NAIP Issue No. 190, July 26, 2017 http://www.naipo.com/Portals/1/web_tw/Knowledge_Center/Laws/IPNC_170726_0201.htm (last visited on January 2, 2019) [48] C-5/08 Infopaq International A/S v. Danske Dagbaldes Forening. [49] Andres Guadamuz, Artificial Intelligence and Copyright, WIPO MAGAZING, October 2017, https://www.wipo.int/wipo_magazine/en/2017/05/article_0003.html (last visited on January 19, 2019). [50] The article indicates that “A work should be protected in “the sense that is the authors’ own intellectual creation. No other criteria shall be applied to determine its eligibility for protection”. [51] Excerpt from the original legal article: in case of a literary, dramatic, musical or artistic work which is computer-generated, the author shall be taken to be the person by whom the arrangements necessary for the creation of the work are undertaken. [52] Excerpt from the original legal article: generated by computer in circumstances such that there is no human author of the work. [53] Andres Guadamuz, supra note 49. [54] Id. [55] Feist Publications v. Rural Telephone Service Company, Inc., 499 U.S. 340 (1991). “the fruits of intellectual labor that are founded in the creative powers of the mind.” [56] Naruto v. Slater, 2016 U.S. Dist. (N.D. Cal. Jan. 28, 2016). [57] The 2014 version of Article 313.2 provides a list of the examples not eligible for the U.S. Copyright Act protection. These include the works generated by the nature, animals or plants and the works purely generated by machines or machinery at random, without any creative inputs or intervention from humans. The examples given are photos taken by a monkey and murals painted by an elephant. The 2014 version establishes that works not created by humans are not eligible for copyright protection. The 2017 version takes a step further with more specific and straightforward wording. [58] Copyright Act 503.03(a): Works-not originated by a human author. In order to be entitled to copyright registration, a work must be the product of human authorship. Works produced by mechanical processes or random selection without any contribution by a human author are not registrable. Thus, a linoleum floor covering featuring a multicolored pebble design which was produced by a mechanical process in unrepeatable, random patterns, is not registrable. Similarly, a work owing its form to the forces of nature and lacking human authorship is not registrable; thus, for example, a piece of driftwood even if polished and mounted is not registrable. [59] Andres Guadamuz, supra note 49. [60] Lin Li-Chih, An Initial Examination of Copyright Disputes Concerning Artificial Intelligence —— Centered on the Author’s Identity, Intellectual Property Rights Journal, Volume 237, September 2019, pages 65-66 [61] The legislative rationale for Article 12 of the R.O.C. Copyright Act: The sponsor and the sponsored are typically in a more equal position for the works completed with sponsorship. It is different from the situation where the works are completed by an employee by using the hardware and software offered by the employer and receiving salaries from the employer. Therefore, the ownership of copyrights depends on the contract between the sponsor and the sponsored regarding the investment and sponsorship purposes. Unless otherwise specified by the contract, the sponsor typically provides funding because of his/her intention to use the works completed by the sponsored. Therefore, the intellectual property should belong to the sponsored. [62] Lin Li-Chih, same as Note 60, pages 75-76. Further reference of the principle used in the U.S. system: Annemarie Bridy (2016), The Evolution of Authorship: Work Made by Code, Columbia Journal of Law, https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2836568. Also the same author (2012), Coding Creativity: Copyright and the Artificially Intelligent Author, Stanford Technology Law Review, https://papers.ssrn.com/sol3/papers.cfm?abstract_id=1888622. [63] Lin Li-Chih, same as Note 60, page 76 [64] Frank H Easterbrook, Cyberspace and the Law of the Horse, 1996 U. CHI. LEGAL F. 207. [65] Please refer to State v. Loomis, 317 Wis. 2d 235 (2016).