Looking at the current patent laws of various countries, there are no laws recognising AI systems or machines as inventors. Major global intellectual property authorities such as China, the United States, Japan and Korea all regard inventors as natural persons, and Lithuania and Estonia, contracting states of the EPC, also explicitly limit inventors to natural persons that created the invention.
In October and November 2018, Dr Stephen Thaler filed two European patent applications, listing an AI machine named DABUS as the inventor. Mr Thaler designated the AI machine as the inventor for two reasons: 1) the European Patents Office (EPO) does not explicitly state that the inventor must be a natural person; and 2) the invention was created by an AI and the applicant is obliged to disclose the true inventor to the public.
The EPO rejected both applications on the basis that listing an AI as an inventor is not in compliance with Article 81 of the EPC and Rule 19(1) of its Implementing Regulations, which require the designation of an inventor. In its decision, the EPO held that a machine has no rights and legal personality to exercise rights, cannot be employed, nor can it have any legal title over its output that can be transferred by law or by agreement.
A review of the patent application process in the United Kingdom shows that the UK Intellectual Property Office (IPO) does not accept the designation of an AI machine as an inventor. The IPO believes that even if the AI machine is regarded as the inventor, it is unclear how the applicant would obtain the ownership of the invention from the machine because the AI machine is incapable of owning any rights.
The EPC supports and further expands the position held by the UKIPO. In Thaler v The Comptroller-General of Patents, Designs and Trade Marks, the EPC held that merely inventing something would not result in a patent being granted to an inventor. For an invention to be granted, a patent must be applied for, which can only be done by a person. As an AI machine is incapable of submitting patent applications for their inventions, it could not be regarded as an inventor under existing UK patent laws. The EPC also holds that an AI machine is incapable of holding and transferring property rights, which an invention, an application for the grant of a patent and the patent itself all are. The court did leave open the question of if the owner of the machine (as a person) could apply and then rely on the invention having been devised by the machine. Dr Thaler had expressly declined to make this argument and the judge left the matter by saying the argument was not an improper one.
Article 17 of the PRC Patent Law states that the inventor or designer has the right to indicate that he is the inventor or designer in the patent document. Rule 13 of the Implementing Rules stipulates that the ‘inventor or designer’ refers to a person who has made creative contributions to the substantive characteristics of the invention. The Patent Examination Guidelines specifically state that the ‘inventor should be an individual’.
Title 35 of the United States Code, Section 100(f) (35 USC §100(f)) stipulates that the ‘inventor’ means the individual or collective group of individuals who invented or discovered the subject matter of the invention. 35 USC §115 also stipulates that each inventor must also execute an oath or declaration in connection with the patent application. A machine will be incapable of making such an oath or declaration.
In Japanese patent law, though the inventor is not clearly defined, the interpretation of the relevant provisions of the Japanese Patent Act and judgement of the courts show that Japan adopts the principle that ‘inventors are limited to natural persons who have actually performed the creative activity’.
- David Brinck argues there is a pressing need for a change in the law to recognise AI entities as inventors
Summary of inventorship
In relation to inventorship, all the countries discussed have adhered to ‘inventor-centrism’ and clearly delineated between machines and humans. In fact, the five major offices of China, the United States, Europe, Japan and South Korea reached a consensus on AI in the IP5 expert roundtable held in October 2018 that although it may be difficult to determine whether a specific invention was made by a human or machine, the inventor must be a natural person.
Patent-eligible subject matter
Patent applications in the AI field are usually significantly different from those in other fields. For example, solutions usually include algorithms and other intellectual activities and methods that are considered non-technical features. Such intellectual activities and methods are generally excluded by patent laws of various countries. In order to provide clearer guidelines and rules, major global IP organisations that handle most AI patent applications have taken steps to modify and improve patent examination rules to cope with the rapid developments in technology.
In December 2019, The China National Intellectual Property Administration (CNIPA) issued a revision of the Patent Examination Guidelines, which came into effect on 1 February 2020. The revised Guidelines focus on the examination rules for patent applications involving algorithm features or business rules and methods. In order to determine subject-matter eligibility, the CNIPA refers to Articles 25.1(2) and 2.2 of the PRC Patent Law (in this order). Specifically, claims that do not contain any technical features but involve abstract algorithms that fall within rules or methods for intellectual activities under the scope of Article 25.1(2) will not be considered as patent-eligible subject matters. However, if the claim contains technical features in addition to the algorithm and is overall not a rule or method of intellectual activity, it should not automatically be denied subject-matter patent eligibility. Instead, it is then necessary to examine whether the claim falls within the ‘technical solution’ described in Article 2.2 of the PRC Patent Law. If the claim states that the technical problem to be solved uses technical methods utilising the laws of nature, and achieves a technical purpose in line with the laws of nature, then the solution described in the claim is within the scope of the ‘technical solution’ described in Article 2.2 of the PRC Patent Law.
In November 2018, the EPO revised its Guidelines for Examination, adding guidelines for AI and machine learning. According to the EPO, AI and machine learning are based on computational models and algorithms, and such models and algorithms are of an abstract mathematical nature. Since certain terms that refer to abstract models and algorithms do not necessarily imply the use of technical means, it can be considered when examining if the claimed subject matter has a technical character. In addition, the EPO considers that if a classification method serves a technical purpose, the steps of generating the training and training the classifier can contribute to the technical character of the invention.
The US Patent Act does not limit subject matter eligibility of patents but provides exceptions through judicial precedents, namely the exclusion of abstract ideas, laws of nature and natural phenomena. This is shown by the Mayo case in 2012 and the Alice case in 2014, which sets the standard for judging patent-eligibility subject matter. In order to make the application process more consistent, the US Patent and Trademark Office (USPTO) subsequently issued the 2019 Revised Patent Subject Matter Eligibility Guidance in January and the October 2019 Update: Subject Matter Eligibility. The revisions elaborate more on Step 2A of the Alice-Mayo test, which determines whether a claim recites a judicial exception.
The Points of Revision of the Examination Guidelines and Examination Handbook for Computer Software-Related Inventions issued by the Japan Patent Office (JPO) classifies AI-generated inventions as computer software-related inventions. The JPO sets out two main considerations in examining subject-matter eligibility in computer software-related inventions: (1) the general rule, which is to consider the invention as a ‘creation of a technical idea utilising the laws of nature’ or as those that utilise the laws of nature as a whole; and (2) the special criteria, which is to consider the invention from a computer software viewpoint.
Sufficiency of disclosure
The particularities of AI patent technologies, such as support vector machines and artificial neural networks that are commonly used in AI algorithms, are black box models, difficult to explain and naturally contradict the basic principle that an invention must be made public to be entitled to be in the patent system. Therefore, the sufficiency of disclosure of AI patent applications is of great interest.
Article 26.3 of the PRC Patent Law states that the specification should provide a description of the invention or utility model sufficiently clear and complete that a person skilled in the relevant technical field can carry it out. According to the latest revision of the Patent Examination Guidelines, the specifications of a patent application of an invention containing algorithm features shall clearly and completely describe the solution adopted by the invention to solve its technical problems. When describing the specifications of a solution including technical features, algorithm features that support and interact with technical features of the solution can be included. The description should state how the technical features and the algorithm features work together to produce beneficial effects.
Article 83 of the European Patent Convention states that a European patent application shall disclose the invention in a manner sufficiently clear and complete for it to be carried out by a person skilled in the art. In addition, Section 1, Chapter III of Part F of the Guidelines for Examination points out that since the application is addressed to the person skilled in the art, it is not necessary nor desirable to give details of well-known ancillary features, but the description must disclose any features essential for carrying out the invention in sufficient detail to render it apparent to the skilled person how to put the invention into practice.
35 USC §112 states that the specification should meet three general requirements: written description, clear directions for implementation and best mode. The Computer-Implemented Functional Claim Limitations Guidance, which came into effect in January 2019, refines and explains the general requirements.
Article A36 (4) (I) of the Japanese Patent Act states that the description shall be clear and sufficient as to enable any person ordinarily skilled in the art to which the invention pertains to work the invention. According to the JPO, training data with multiple types of data for machine learning is used in AI-related inventions. Conditions for determining sufficiency of disclosure include the condition where it can be recognised that there is a certain association between the various types of data, such as a correlation or where it can be presumed that there is a certain correlation between the multiple types of data in view of a common general technical knowledge.
Inventiveness is the most important patentability requirement and plays an important role in the patent system. The judgment of inventiveness is often relatively subjective. In addition, due to the differences in legislative history and societal values among the countries, there are differences in judging inventiveness, particularly for AI patents.
Article 22.3 of the PRC Patent Law sets out the general criteria to determine inventive step. To show inventive step, the invention or utility model must have prominent substantive features and notable progress compared to the prior art. For AI-related patent applications, examination of inventive step still follows the general approach for assessing inventive step; namely, the ‘three-step method’ used to determine whether the technical features of the invention is obvious to a person skilled in the art or not. The latest revision of the Patent Examination Guidelines further provide that when examining the inventive step of a patent application that contains both technical features and algorithm features, both the technical features and algorithm features must mutually interact and support each other in the functioning of the invention, and should be considered as a whole.
Article 56 of the European Patent Convention states that an invention shall be considered to have inventive step if, having regard to the state of the art, it is not obvious to a person skilled in the art. Section 5, Chapter VII of Part G of the Guidelines for Examination establishes the ‘problem-solution approach’, which sets out three main stages to assess inventive step in an objective and predictable manner. In particular, when assessing the inventive step of a claim containing technical and non-technical features (often the case with AI-related inventions), the problem-solution approach is applied in such a way as to ensure that all the features which contribute to the technical character of the invention are taken into account, while features which do not contribute to the technical character of the invention should not be considered.
Article 29(2) of the Japanese Patent Act states that if a person ordinarily skilled in the art of the invention could easily make the invention based on a prior art or an invention that is publicly known, then the invention lacks inventive step. According to its Guidelines for Examination, all claims described in the invention (whether technical or non-technical features) will be considered in determining inventive step.
35 USC §103 sets out the requirement of non-obviousness but does not clearly define it. Guidance can, however, be found in the 1962 case Graham v John Deere and in the 2007 case KSR International v Teleflex, where the US Supreme Court set out four elements in resolving the question of non-obviousness and refined the determination of non-obviousness to increase the flexibility in judging it, respectively.
Summary of inventive step/non-obviousness
It can be seen that there is a difference between the EPO and the JPO and CNIPA in treating non-technical features when evaluating inventive step in AI-generated inventions. The EPO explicitly does not consider non-technical features without technical contribution, while the JPO and CNIPA explicitly consider both technical and non-technical features described in the claims. Though the USPTO does not make clear the distinction between technical and non-technical characteristics, in practice, it makes a judgement on non-obviousness after examining all the evidence.
As AI-generated inventions become more common in the future, the existing patent legal framework and system may be unable to fully meet the demands of such inventions. As a consequence, major intellectual property authorities have responded to the emergence of AI-generated inventions by revising and clarifying existing patent laws and guidelines. Though standards of protection and examination rules between the countries may differ, there is an open and positive attitude towards patent protection of AI-generated inventions.
Wensen An is a patent lawyer and director of the Lusheng law firm, a member of the Rouse Network