Patent protection for ai-assisted and ai-generated inventions: a patent examination guidelines approach
Rodrigo Guerra M. e Silva – State University of Campinas – Institute of Geosciences – Department of Scientific and Technological Policy – Brazil
Alexandre Guimarães Vasconcellos – Division of Graduate Studies and Research of the Academy of Intellectual Property, Innovation and Development of the Instituto Nacional da Propriedade Intelectual – INPI/BR – Brazil
Rafaela Guerrante – Coordinator of the Strategic Committee for Gender, Diversity and Inclusion – CEGDI do Instituto Nacional da Propriedade Intelectual – INPI/BR -Brazil
Eduardo Fonseca – Partner, Patent Agent at Moffat & Co – Canada.
Sérgio Luiz Monteiro Salles Filho – State University of Campinas – Institute of Geosciences – Department of Scientific and Technological Policy – Brazil
keywords: Intellectual property, patent, artificial intelligence and machine learning.
Abstract
The current industrial property system responsible for the protection of several intellectual assets, including utility patents, is facing scrutiny regarding its ability to meet the global demand for patent exam (nearly 3.5 million new patent applications annually) and to effectively protect the growing number of inventions generated or assisted by machine learning. The system has shown signs of saturation (5.6 million patents pending examination) and a potential collapse soon. This scenario is worsened by the fourth industrial revolution, specifically due to the challenge of examining patents after the advent of autonomous inventions equipped with intelligent systems, as well as those generated by robots that have the ability to learn. This study focuses on analyzing the suitability of examination guidelines of different jurisdictions (Brazil, USA, China, and the European Union) related to computer-implemented inventions addressing machine learning, considering their autonomous and dynamic nature, aiming to contribute to discussions about the future of the patent system. The study finds that machine learning is highly diverse, encompassing various models. It also observes that some models do not alter the status quo of the current patent system, while others introduce a new profile to inventions, suggesting a continuous improvement characteristic not fully addressed by existing patent examination guidelines. This scenario reinforces questions about the adequacy of the system and the need for more in-depth debates regarding regulations and operationalization of the patent rights system.
Introduction
Worldwide, inventors submit approximately 3.5 million patent applications annually to the International Patent System (IPS). While this number may vary based on the analyzed time frame, it consistently shows an upward trend. According to WIPO reports (2022), long-term trend indicates a continuous increase in patent applications. In 1995, the milestone of one million global filings in a single year was surpassed. In 2012, it reached two million, and by 2016, it grew to three million. The growth is steady but not entirely linear. During the SARS-CoV-2019 pandemic (2020-2022), the IPS experienced a modest slowdown, with just a 3.6% increase (3,281,900 – 3,401,100). However, despite this temporary deceleration, the global backlog of pending applications continued to expand.
To address this demand and attempt to avoid delays, patent offices worldwide rely on examination guidelines and a large workforce of examiners responsible for analyzing legal requirements and conducting technical comparison between existing inventions and those claimed by the inventor. This process, dependent on highly qualified human resources, is time-consuming and expensive, constituting the primary cause of delays in patent decision-making.
The United States Patent and Trademark Office (USPTO) receives around 600,000 annual applications, employing over 10,000 professionals directly or indirectly dedicated to this process (over 8,000 being patent examiners). Nevertheless, they accumulate around one million pending decision requests (WIPO, 2022). China’s patent office (CNIPA) leads the world in patent applications, representing half of the global filings. In 2020, they had the same number of pending applications as the USPTO (~1 million). However, in 2021, they counted 1.3 million, indicating an inability to keep up with current demand. Nations with constrained resources for examiner recruitment face even greater delays, surpassing 10 years, as recently seen in Brazil. However, this inefficiency is not a local problem. The global backlog was 5.2 million in 2021 and had already surpassed 5.6 million in 2022 (WIPO, 2022), indicating a worsening trend and a potential collapse of the system unless measures are taken.
The deliberations held during the April 2023 event at the USPTO (USPTO AI Inventorship, 2023) encompassed significant discussions among Artificial Intelligence (AI) experts, including the concern on the potential increase in patent applications generated by AI itself. It is anticipated that the growing use of AI tools to assist in patent search and drafting, as well as in proposing new inventions, will accelerate the generation of new applications. This scenario aligns with Kevin Kelly’s concept (2016) of the increasing trend to incorporate intelligence and cognitive capabilities into previously non-intelligent objects, systems, and processes. This could lead to “redesign” and new patent applications for existing inventions by simply adding AI. Hence, there is a need to reassess current practices and even implement modern tools to assist operational functions and expedite the patent examination process (Guerrante, 2023; USPTO, 2023).
This study intends to assess the adequacy of the current IPS in different jurisdictions, evaluating their capacity to handle current and future patent application demands in the context of the fourth industrial revolution, especially in AI. The focus is on determining whether the examination guidelines of selected patent offices are suitable for the new characteristics presented by inventions generated or assisted by AI models and whether these guidelines do guide examiners to perform their functions efficiently and promptly, preventing an increase in the backlog of pending applications.
Despite being longstanding and robust, having a reach over the last three industrial revolutions, this study demonstrates that the IPS can be questioned regarding its adequacy to AI developments, as it does not keep pace with AI advancements, posing new challenges to the IPS. Additionally, it does not sufficiently address AI models in its guidelines, which produce non-deterministic and dynamic inventions, as well as inventions generated by AI, as it can be seen below.
New Challenge for IPS
In addition to the Covid-19 pandemic, the current cognitive revolution has demonstrated that several factors, such as economic, technological, and social aspects, can influence the pace of patent filings. The world is going through an industrial revolution (Schwab, 2015) and is undergoing a disruptive period. Society is in transition, where analog systems have been completely overtaken by the digital revolution (WEF, 2016; MARR, 2018), coupled with the recent advent of intelligent computational systems. This period of mutation is marked by one of the most significant technological paradigms shifts in history: the machine learning – ML (Elkhova, 2017), also known as artificial intelligence – AI. At the core of Industry 4.0 or the Fourth Industrial Revolution – FIR, represented by the integration of physical and virtual systems, geared towards more dynamic, flexible, and automated global production, accelerating the pace of creations and innovations even more than in previous industrial revolutions (WEF, 2016) and leading to more radical transformations in work patterns (EUROPEAN PARLIAMENT, 2015).
The replacement of human labor by machines has intensified, gradually replacing functions that were previously exclusive to the human mind, resulting in job losses (Frey, 2017), initially affecting the most repetitive and uninteresting tasks but still significantly impacting society (Brynjolfsson, 2014). At the same time, David H. Autor (2015) sees AI as an opportunity, not a risk, as it enhances and complements human work, as seen in the case of generative AI. AI has the capacity to create (Elkhova 2017; Boden 2020), potentially being the “last invention of man” (Bostrom, 2014), and its learning power is incorporated into inventions, requiring a distinct treatment by the IPS.
As predicted and debated by experts, the patent system is being impacted, and demands for adjustments are being made (WEF, 2018; Discher, 2019; JPO, 2019; TECHPATS, 2020; WIPO, 2020a; WIPO, 2020b; WIPO, 2021; Slowinski, 2021a). Tim Dornis (2021) even foresees the end of the patent system due to the advent of inventions generated by AI itself, which he believes will initially be denied in supreme courts for lacking “human inventors”. However, over time, the dominance of such autonomous sources of creation will become so significant that it cannot be prevented, breaking various patent doctrines, policies, and practices (Barbosa, 2010).
AI-assisted and AI-generated inventions
A crucial approach to assess the responsiveness of the IPS to these new pressures is through the investigation of patent guidelines, ensuring their adequacy for the new inventions that are already being filed by hundreds of thousands worldwide (WIPO Technology Trends, 2019; Discher 2019; TECHPATS, 2020) and those yet to come. This is not only a problem of the volume of new applications but also of the nature of the inventions, as defined by Drexl et al. (2019; 2021), inventions assisted, implemented, or generated by AI (Hilty, 2020).
AI-assisted inventions (assisted + implemented) may exhibit characteristics not found in current inventions, such as autonomy, adaptability, lack of transparency (black-box), randomness, dependence on chance, continuous improvement, non-reproducibility, and non-deterministic nature. This challenges the traditional conception of patents (JHA, 2020) as there is a continuous capacity for learning (dynamic), conflicting with the rules of the IPS (Kim, 2020; Nurton, 2020), based on the exact/corresponding copy of what was claimed.
These claims are based on the fact that ML is programmed to recognize patterns in databases, which can be subsequently repeated, modified, and enhanced. Despite being mathematical and statistical methods, their results closely resemble the randomness or chance found in human actions. Various hyperparameters govern this process, including the use of pseudo-random sequence generators, as originally advocated by the mathematician John von Neumann. With the increase in computational capacity, pseudo-randomness in practice has become equivalent to real randomness. Consequently, AI can empower technology with the autonomous and non-deterministic ability to infer the cause-and-effect relationship between input features and output data until achieving satisfactory generalization (Varshney, 2022) and can perform its function on never-before-seen data.
Regarding inventions generated by AI, despite being based on previously existing knowledge primarily derived from humans, the generalization power of AI can go beyond what already exists and what can be imagined (Boden, 2020). Currently, generative AI can create texts, images, audios, codes, generate ideas, make decisions, and even invent (Kim, 2020; Oliver, 2021). However, the exact manner in which it learns can be a mystery, even to its developers. This computational phenomenon was termed by David Castelvecchi (2016) as the “black box of AI”.
Depending on the type of model and AI created, there are tools capable of unraveling the causal inference made by the machine, such as counterfactual verification techniques (Rajkomar, 1990; Varshney, 2022). However, models using more advanced machine learning are increasingly autonomous and therefore more difficult to be deciphered. This distinguishes AI models from conventional computer programs that are pre-programmed by humans. Thus, an invention may be independently generated by the machine (WO2020079499A1_DABUS), raising questions about the authorship/ownership of the invention (Oliver 2021; Nurton, 2020). It may also pose difficulties in describing exactly what was invented due to the “black box of AI” (Castelvecchi, 2016; WEF, 2018; WIPO, 2019; WIPO, 2020b).
Fixed or dynamic invention
An invention containing AI models can be enhanced by stochastic optimization methods, performing loops or backpropagation in the learning process, manually or autonomously modifying the hyperparameters of neural networks (Russel, 1995), aiming to reduce errors. However, the model may worsen or improve with each new cycle (epoch), yet the aim of this feedback loop is to eliminate less efficient models. In this regard, the developer can choose to stabilize the model or leave it dynamic, making it different with each new cycle (epoch). For example, a technology containing a deep neural network architecture (Deep Learning – DL – Hinton, 2006) with reinforcement learning – RL will improve with each use and can never be copied, leading to the “reproducibility, reusability, and robustness crisis of AI” (Pineau, 2019; Jha 2020). Since, even though it is an algorithm, reproducing every step faithfully would be necessary to obtain an exact copy, which is practically impossible. Any modification will generate a completely different model. These non-deterministic inventions cannot be examined in the same way as static or fixed inventions.
Methodology
In order to assess the capability of patent-granting offices to meet the global demand for new applications, it is crucial to measure the fundamental parameters of merit examination. Examination guidelines play a pivotal role in this process, providing clear criteria and specific guidance for IPS users and examiners, promoting greater standardization and uniformity in patent examination criteria, ensuring speed and consistency.
Patent guidelines are, therefore, key elements in this study as they serve as a mechanism for transparency (broad publicity) and guidance for patent examiners, especially in complex cases involving artificial intelligence (AI).
It is important to emphasize that patent guidelines are official and periodically updated. This adaptability to new technologies is one of the objects of analysis for experts (Wang, 2020; Jiang, 2019) and also in the present study. The potential lack of specific guidelines for AI would result in negative impacts, such as non-standardized, arbitrary, and sometimes incorrect decisions, leading to uncertainties and legal insecurity, resulting in significant delays in the examination process, inequalities between technological areas, and missed opportunities for innovation.
Even acknowledging that examination guidelines are not the sole factor determining patent examination time and efficiency, and that each jurisdiction examines the same claimed subject matter differently, it is crucial to conduct documentary (Gil, 2002) and qualitative research to analyze them in more than one territory. For this, the official patent guidelines of four jurisdictions with different profiles were evaluated based on readily available information from primary sources. These regions are: 1) Brazil (INPI – www.inpi.gov.br); 2) China (CNIPA – www.cnipa.gov.cn); 3) Europe (EPO – www.epo.org); and 4) the United States of America (USPTO – www.uspto.gov). These guidelines encompass Western and Eastern countries, developed and developing nations, with different political-economic and legal systems: civil law vs. common law (Jacinto, 2022), and varying volumes of patent applications and examiners, generating an estimator of global diversity.
Consequently, this study proposes a classification for the degree of adequacy of a jurisdiction’s office for the examination on the merits of AI inventions. It is divided into three levels: 1 – Lack of specific guidelines; 2 – Generic guidelines; and 3 – Specialized guidelines.
LEVEL 1: Lack of specific guidelines. Complete absence of guidelines for the examination of AI inventions. In these IP offices, examination guidelines do not include expressions such as artificial intelligence and machine learning. Even when present, examination guidance is quite superficial. The impact on examination delays in offices with this classification can be significant. Examiners may struggle to assess the legal criteria of the SIP.
LEVEL 2: Generic guidelines. There are examination guidelines for AI-assisted and/or generated inventions. They include expressions related to AI, but the guidance is still superficial or generic. The impact on examination delays can still be significant. Generic guidelines still lead to examination difficulties, resulting in uncertainties and challenges in interpreting and consistently applying patentability criteria and conditions.
LEVEL 3: Specific guidelines. Presence of specific guidelines for both AI-generated and assisted inventions. The examination time tends to be reduced. The guidelines are detailed regarding the identification of the type of invention, patentability criteria, technical requirements, and examination practices for AI, facilitating the work of examiners, enabling them to make more consistent and efficient decisions.
Results and discussion
From the analysis of information (primary source) available on the selected IP office websites (A – Brazil, B – China, C – USA, and D – Europe), it was possible to observe that there are official and properly published examination guidelines. All of them classify AI-endowed inventions as inventions implemented by computer programs. However, as mentioned, framing machine learning algorithms as conventional computer programs is a simplification that does not assist in the examination.
A – Brazil. Brazilian guidelines (IIC GUIDELINES, 2020; INPI RESOLUTION, 2016) are classified as LEVEL 1 – lack of specific guidelines for AI-generated or AI-endowed inventions. There is no mention of how to examine such inventions in the law (Law 9.279/96). The guidelines briefly refer in paragraph [013]: “Artificial intelligence (AI) techniques, including machine learning and deep learning tools, among others, when applied to solving technical problems, may be considered an invention.” Although the term AI is mentioned, there is no guidance, specifically stating that these are patentable. In paragraphs 006 to 008, 016, and 034, there is a definition of an algorithm (related to AI), stating that it is patentable if it does not fall under Article 10 (considered non-invention). Paragraph [008] also defines that a computer program itself is not patentable unless embedded in a process or device that is patentable. Paragraph [019] determines that language instructions and source code are not considered inventions, falling under another law (Law 9.609/98). In summary, it suggests that products and processes embedding machine learning (regardless of the model) are patentable.
B – China. Analogous to Brazilian guidelines, the Chinese guidelines should also be classified as LEVEL 1 (DECREE No. 3062001; Guidelines for Examination, 2006). There is no mention of expressions related to artificial intelligence in the Chinese guidelines. AI is considered a sequence of codes and a method of mental activity. Similar to Brazil, it would be eligible for protection only if embedded in another patentable invention.
C – USA. Preliminarily, the USA has a different legal format than the other countries analyzed in this study. Common law (USA), unlike Civil law (Europe, Brazil, and China), emphasizes jurisprudence. In general, laws exist, but jurisprudence prevails in legal decisions. On the other hand, for patent purposes, federal law is sovereign. Jurisprudence only complements legal decisions and can be used by the USPTO. There is a patent examination manual (MPEP, 2023), which also does not mention expressions related to artificial intelligence. However, the USA is well advanced in the discussion of authorship/ownership of robots in patents (inventorship) due to the “Thaler v. Vidal” case (2019), which includes the Dabus Robot as the sole inventor. In 2022, the USPTO decided that an inventor must be a natural person (Rule 35 U.S.C. §§ 100 – 105; MPEP 2109 and MPEP 2138.04), interpreting that the word “individual” referred exclusively to a human being. Later, an RFC (Request for Comments) was conducted in the USA (88 Fed. Reg. 9492), with the collaboration of experts, resuming the discussion, reconsidering the possibility of including machines as inventors or the possibility of including the person behind them as an inventor (USPTO AI REPORT, 2020) . The US guidelines lack debate on AI-assisted inventions but were classified as LEVEL 2 – Generic guidelines, due to the maturity level regarding AI-generated inventions. It cannot be classified at LEVEL 3 since there are no specific guidelines, and it still equates AI models to mathematical methods, abstract ideas, and conventional computer programs, discussing only eligibility issues (see example 39 of the Revised Patent Subject Matter Eligibility Guidance).
D – Europe. European guidelines (EPO Guidelines, 2023) are the only ones that mention guidance on patenting inventions involving AI, classified as LEVEL 2. In Chapter II (Inventions), item 3.3.1 – Artificial intelligence and machine learning, it is stated that: the EPO considers inventions involving AI as computational models and algorithms capable of performing some functions, such as classification, clustering, regression, dimensional reduction; which can be neural networks, genetic algorithms; and represents AI models as support vector machines, k-means, kernel regression, and discriminant analysis. The examples given were concise but already indicate familiarity with different AI models. The guidelines are firm in determining that these neural networks and genetic algorithms alone are of an abstract mathematical nature, regardless of being trainable. Therefore, they will be examined as a conventional computational algorithm or model, falling under the prohibitions of mathematical methods in item G-II, 3.3. Thus, merely using terms like “support vector machine” or “neural network” does not necessarily imply the presence of technical character in the invention. It is not enough to mention such expressions but to demonstrate that there is a technical purpose (see Art. 52(1), (2), and (3)), which also cannot be merely linguistic processing, as in the example of document classification (T 1358/09). Instead, a technical contribution achieved by the use of algorithms must be demonstrated.
Conclusions
Indicators related to the efficiency of the current IP system (Intellectual Property), responsible for the legal protection of patents, show saturation of the system, primarily evidenced by the backlog. This has raised questions about its capacity to meet global demand for patents in face of the disruptive technologies of Industry 4.0. These technologies are expected to increase society’s inventive capacity, potentially enabling the generation of inventions autonomously and contributing to the phenomenon known as “cognification of everything” (Kelly, 2016), leading to “re-filing.” This would necessitate countries to have modern guidelines that keep pace with this technological evolution.
However, the research findings indicate that examination guidelines from the four selected jurisdictions (Brazil, China, USA, and Europe) lack well-defined guidance on AI-assisted inventions. Brazil and China were classified at LEVEL 1, while the USA and Europe showed more advanced discussions on machine inventors but were still at LEVEL 2. None reached LEVEL 3.
Despite being classified at different levels, all evaluated guidelines still need improvement to provide clearer and more precise guidance to examiners on minimum conditions for describing these inventions, determining if an invention belongs to the field of AI, comparing two inventions containing similar intelligent systems, and what would be considered novel or inventive. The analyzed material lacks discussions on the reproducibility of inventions, their industrial application, and the definition of an expert in the field. There is no mention of the non-deterministic nature of inventions and their potential for continuous improvement. Another crucial discussion not found in the analyzed guidelines is that of the “black box of AI” and its impact on the descriptive sufficiency required in all analyzed jurisdictions. There is also a need for a discussion on ethical and moral conflicts outlined in international treaties such as TRIPS, as well as ways to address them (Gebru, 2021).
It is concluded that countries consider AI-assisted inventions equivalent to inventions assisted by mathematical and abstract methods or conventional computer programs (without statistical learning). This is still far from a normative framework sufficient to expedite examination in this field.
The IP system is not yet normatively aligned with the 4th industrial revolution, and patents granted in this way may not translate into real incentives for innovation. Enforcement of these patents may be questioned in the market and legal instances. These inconsistencies hinder the strategic management of intellectual property and innovation at micro, meso, and macroeconomic levels, posing an obstacle to the formulation of public and private policies.
References
AI WATCH. (2020). Historical Evolution of Artificial Intelligence Analysis of the three main paradigm shifts in AI. Join Research Center Technical Reports. European Commission.
Autor, D. H. (2015). Why Are There Still So Many Jobs? The History and Future of Workplace Automation. Journal of economic perspectives. Vol.29, NO. 3 (pp. 3-30).
Barbosa, D. B. (2010). Tratado da Propriedade intelectual. Introdução à propriedade Intelectual. Lumen Juris.
Boden, M. A. (2020). Inteligência Artificial: Uma Brevíssima Introdução. UNESP.
Bostrom, N. (2014). Superintelligence: paths, dangers, strategies. Oxford University Press.
Brynjolfsson, E., & McAfee, A. (2014). The second machine age: Work, progress, and prosperity in a time of brilliant technologies. WW Norton & Company.
Castelvecchi, D. (2016). The back box of AI. Nature, vol 538.
DECREE No. 306. (2001). Implementing Regulations of the Patent Law of the People’s Republic of China. (2001, June 15). Promulgated by Decree No. 306 of the State Council of the People’s Republic of China, and effective as of July 1, 2001.
DIRETRIZES IIC. (2020). Diretrizes de Exame de Pedidos de Patente envolvendo Invenções Implementadas em Computador. Dezembro de 2020.
Discher, G. (2019). Artificial Intelligence and the Patent Landscape – Views from the USPTO AI: Intellectual Property Policy Considerations Conference.
Dornis, Tim. (2021). Artificial Intelligence and Innovation: The End of Patent Law As We Know It, 23 Yale Journal of Law & Technology (YALE J. L. & TECH.), pp. 97 et seq.
Drexl, Josef et. al. (2019). Technical Aspects of Artificial Intelligence: An Understanding from an Intellectual Property Law Perspective. Max Planck Institute for Innovation and Competition Research Paper No. 19-13.
Drexl, Josef et. al. (2021). Artificial Intelligence and Intellectual Property Law. Position Statement of the Max Planck Institute for Innovation and Competition of 9 April 2021 on the Current Debate. No. 21-10.
Elkova, O. I. & kudryashev, A. F. The creative ability of artificial intelligence. creativity studies. ISSN 2345-0479/eISSN 2345-0487. Volume 10(2): 135–144, 2017.
EPO Guidelines. (2023). Guidelines for Examination in the European Patent Office, Part G – Chapter II-6, item 3.3.1, March 2023.
EUROPEAN PARLIAMENT. (2015). Industry 4.0 Digitalisation for productivity and growth;
Frey, C. B & Michael A. Osborne. (2017). The Future of Employment: How Susceptible are Jobs to Computerisation? Technological Forecasting and Social Change. Vol 114, pg 254-280;
Gebru et. al. (2021). Datasheets for datasets. https://arxiv.org/abs/1803.09010.
Géron, A. (2021). Mãos à Obra: Aprendizado de Máquina com Scikit-Learn e TensorFlow. Rio de Janeiro. Alta books.
Gil, A. C. (2002). Como Elaborar Projetos de Pesquisa – 4. ed. – São Paulo: Atlas. ISBN 85-224-3169-8.
Guerrante, R. S. (2022). AI Solutions for the Efficiency of IPp Offices. WIPO Conversation on Intellectual Properties and Frontier Technologies: AI Technologies.
Guidelines for examination. (2006). State Intellectual Property Office of the People’s Republic of China. Chapter 9. Recuperado em maio de 2023, de http://www.chinadaily.com.cn/specials/guidelines2006(EN).pdf.
Hilty et. al. (2020). Comments on the Draft Issues Paper of the WIPO on Intellectual Property Policy and Artificial Intelligence of 11 February 2020, 9 pp..
Hinton, G. E. et. al. (2006). A Fast Learning Algorithm for Deep BeliefNets. Neural Computation 18, 1527-1554.
Jacinto Junior, F. G. & Souza, L. C. P. de. (2022). A Evolução do sistema jurídico chinês e a influência dos sistemas jurídicos ocidentais. Zi Yue, 2(1).
Jha, A. H. & AFEK, E. (2020). How to get started with ML Reproducibility Challenge 2020. Towards Data Science.
Jiang, T. & Hu, S. (2019). Intellectual Property Protection for AI-Related Inventions in Japan. 2019 International Conference on Virtual Reality and Intelligent Systems (ICVRIS), Jishou, China, pp. 286-289.
JPO. (2019). JAPAN PATENT OFFICE. Recent Trends in AI-related Inventions – Report;
Kelly, K. (2016). The Inevitable: Understanding the 12 Technological Forces That Will Shape Our Future. Penguin Books.
Kim, D. (2020). AI-Generated Inventions’: Time to Get the Record Straight? GRUR International, Volume 69, Issue 5, May 2020, Pages 443–456.
Lei 9.609 de 19 de fevereiro de 1998.
Lei 9.279 de 14 de maio de 1996.
Marr, Bernard. (2018). Why Everyone Must Get Ready For The 4th Industrial Revolution. Forbes. Retrieved 14 February 2018.
MPEP. (2023). Manual of Patent Examining Procedure. Ninth Edition, Revision 07.2022, Published February 2023.
Nedelkoska, L. & Quintini, G. (2018). Automation, skills use and training. OECD Social, Employment and Migration Working Papers, No. 202, OECD Publishing, Paris.
Nurton, J. (2020). EPO and URIPO Refuse AI-Invented Patent applications. IP Watch Dog.
Oliver, T. & Engmark, T. (2021). Who’s IP is it? The AI Inventor ou the AI’s inventor? Embedded Computing Design, 10/09/2021.
Pineau, J. et. al. (2020). Improving Reproducibility in Machine Learning Research. A Report from the NeurIPS 2019 Reproducibility Program.
Rajkomar, A. et. al. (2018). Ensuring Fairness in Machine Learning to Advance Health Equity. Ann Intern Med. 2018 December 18; 169(12): 866–872.
RESOLUÇÃO INPI/PR Nº 158, de 28 de novembro de 2016.
Russell, S. J. & Norvig, P. (1995). Artificial intelligence: a modern approach. Englewood Cliffs, N.J: Prentice Hall, 1995 (Prentice Hall series in artificial intelligence).
Schwab, K. (2015). The Fourth Industrial Revolution. World Economic Forum. ISBN-13: 978-1-944835-01-9. (12 December 2015).
Slowinski, P. (2021a). Artificial Intelligence, Novelty and Inventive Step: What Is the Impact of AI on Patent Law? Kluwer Law International, Alphen aan den Rijn.
Slowinski, P. (2021b). Rethink Software Protection. Artificial Intelligence & Intellectual Property, Oxford University Press, 2020. Max Planck Institute for Innovation & Competition Research Paper No. 20-17.
USPTO AI Inventorship (2023). Notice of Public AI Inventorship Listening Session-West Coast. FEDERAL REGISTER.
USPTO AI Report. (2020). Public Views on AI and Intellectual Property Policy.
TECHPATS. (2020). Artificial Intelligence and its Potential Implications on Patents.
Varshney, K. R. (2022). Trustworthy Machine Learning. Independently Published.
Villasenor, J. (2023). Reconceptualizing conception: making room for artificial intelligence inventions. Santa Clara High Tech. Law Journal 197.
WANG, L and Hu, S. Patent Protection for Artificial Intelligence in Europe. 2020 International Conference on Intelligent Transportation, Big Data & Smart City (ICITBS), Vientiane, Laos, 2020, pp. 591-594.
WEF (2016). Mastering the Fourth Industrial Revolution. World Economic Forum Annual Meeting. Davos-Klosters, Switzerland, 20-23 January 2016.
WEF (2018). Artificial Intelligence Collides with Patent Law. World Economic Forum. White Paper.
WIPO (2019). Technology Trends 2019: Artificial Intelligence. Geneva: World Intellectual Property Organization.
WIPO (2020a). WIPO Conversation on Intellectual Property (IP) and Artificial Intelligence (AI): Second Session. Revised Issues paper on intellectual property policy and artificial intelligence. WIPO Secretariat. DATE: MAY 21, 2020. WIPO/IP/AI/2/GE/20/1.
WIPO (2020b). WIPO’s Second Session of Conversation on IP and Artificial Intelligence Ends with Outline of Next Steps. Date: july, 2020.
WIPO (2021). IPCCAT – IPC Computer-Assisted Categorization.
WIPO (2022). World Intellectual Property Indicators 2022. Geneva. DOI:10.34667/tind.47082.
WO2020079499A1_DABUS (BR 11 2021 008931 4 A2).