作者 | 张鹏 牟雨菲 孙杰
Zhong Lun Law Firm
table of contents
1. The English court's judgment on the eligibility of the object of protection of the AI invention involved in the case
2. Rules for the examination of the eligibility of the subject matter of protection of artificial intelligence inventions under English law
3. Comparison of the progress of the judgment of the object of AI protection in the UK with China, the European Union, Japan, and the United States
(1) China's rules for the examination of the eligibility of AI patents
(2) The examination rules of the European Patent Office on the eligibility of artificial intelligence patents
(3) Japan's examination rules on the eligibility of AI patents
(4) U.S. rules on the examination of the eligibility of AI patents
summary
On November 21, 2023, the High Court of England and Wales in the United Kingdom ruled that patent applications related to AI neural networks do not apply to the statutory exclusions of patentable technology under the UK Patents Act 1977.
We present the judgment and analyze the rules for examining the eligibility of the subject matter of protection for AI inventions under English law. On this basis, this paper compares the latest progress of patent protection for AI innovation in the UK with the legal application of relevant rules in the mainland, the European Union, Japan and the United States, in order to make a comprehensive analysis of the judgment of the object of patent protection for AI innovation in various jurisdictions.
On 21 November 2023, the High Court of England and Wales (the "High Court of England") published its judgment in the case of Emotional Perception AI Ltd v Comptroller-General ("Emotional Perception AI")[1], overturning the UK Intellectual Property Office ("UKIPO") BL O/ The 542/22 examination decision held that a patent application related to artificial neural networks (hereinafter referred to as "ANN") in question did not apply to the statutory exclusions of patentable technology in the UK Patents Act 1977[2] and was patentable.
Starting from this case, this article will explore the review rules for the patentability of AI technology in the United Kingdom, the European Union, Japan, the United States and mainland China, and make a comparative analysis and brief commentary on the relevant precedent paths in each jurisdiction.
1. The English court's judgment on the eligibility of the object of protection of the AI invention involved in the case
The patent application filed by Emotional Perception AI relates to a system and method for providing users with media file recommendations, involving the training and use of ANNs to identify semantically similar files. It is mainly used to train artificial intelligence to reflect the subjective response of humans to stimuli such as audio, music, images, videos or text, using artificial neural networks to identify similarities between the contents of two files.
For audio files, measurable signal quality, such as rhythm, tone, timbre, and texture, is extracted from both files to identify musical characteristics (404). The ANN outputs attribute vectors (408) for each musical attribute, and these attribute vectors are combined into multidimensional vectors for each file. The weight or bias value of the ANN is adjusted by backpropagation (422) based on the distance between the two multidimensional vectors (410) and any difference between that distance and the quantized semantic dissimilarity distance between the two files in semantic space (412), such as the textual description of the trajectory or other semantic associations with events, feelings, themes, or environments. Therefore, the ANN is trained to align the distance between files in the attribute space, which corresponds to the distance between files in the semantic space. ANN can then identify music files that are subjectively semantically similar to the object based only on the objective musical characteristics of the object.
A typical application area of this patent application is the recommendation of music repertoire. The patent application is able to recommend semantically similar music based on human perception and emotion. Specifically, the ANN related to the invention is trained on a series of paired music files that generate vectors or coordinates in the semantic space and the property space, respectively. The distance between vectors or coordinates in semantic space (hereinafter referred to as "semantic distance") is used to measure the degree of similarity in the semantic descriptions of pairs of music files (e.g., happy, sad), and the distance between vectors or coordinates in attribute space (hereinafter referred to as "attribute distance") is used to measure the similarity of physical properties (such as tone, timbre, etc.) of paired music files. Through the back-propagation training process, ANN continuously adjusts the weighting and bias in its nodes and evaluations, and repeatedly corrects the error of the attribute distance, so that the attribute distance tends to the corresponding semantic distance. After the ANN training is completed, its node parameters are fixed, and it is intended to be applied and implemented in the music recommendation system to achieve its expected function. In the execution and application phase, ANN will receive a specific music track provided by the remote user, analyze and determine its physical properties, generate an attribute vector or coordinates, and determine the semantically similar music by finding music files with similar properties in the reference database, so as to recommend music semantically similar to the given music track to the user. The system will eventually send users messages and files recommending semantically similar music.
The content of claim 1 of the patent application in question is, "a training system (300) artificial neural network' ANN'(NNR310, NNTO312, NNTI314, NNTX318) is configured to identify the similarity or dissimilarity of the contents of the first data file (302) with respect to the contents in different data files (304), which comprises: extracting measurable signal qualities from each of the first data files and different data files for a selected pair of different data files, and identifying multiple attributes from those measurable signal qualities; to generate 10 corresponding attribute vectors (ORx, OTOx, OTIx and OTXx) in the attribute space for each attribute of the first data file and the different data files of the selected pairs; assemble the first multidimensional vectors (350) for the first data file and different second multidimensional vectors (352) for different data files, wherein each multidimensional vector is obtained by assembling the corresponding attribute vectors from the output of the ANN; determine the distance measure (330) between the first multidimensional vector (350) and the second multidimensional vector (352); and respond to the determined distance measurement, adjust the weights by backpropagation process and/ or bias ANN (NNR310, NNTO312, NNTI314, NNTX318), the process takes into account the identified differences between the distance measure in the determined attribute space and the quantized semantic dissimilarity distance measure of the first data file evaluated in the semantic space relative to the different data files, so that the system is trained by assigning the semantic perception reflected in the quantized semantic dissimilarity distance measure relative to the attribute evaluation reflected by the distance measure (330) between the first multidimensional vector (350) and the second multidimensional vector (352). 。 ”
The judgment of the High Court of the United Kingdom in the Emotional Perception AI case pointed out that there are two types of ANNs involved in the invention patent application, namely hardware ANNs and software ANNs. A hardware ANN refers to a physical box containing electronic components that contains nodes and layers in the form of hardware, nodes similar to neurons in the brain, and layers similar to neural network layers in the brain. Software ANN refers to the existence of ANN in the form of computer emulation through software, in which case the traditional computer runs a piece of software that enables the computer to emulate the hardware ANN. Based on this, two judgments are drawn:
First, the AI invention in question was not a computer program.
In its decision on the examination of the patent application, the UK Intellectual Property Office held that the patent application was a computer program which should be excluded from patent protection under section 1(2)(c) of the UK Patents Act 1977. The High Court of England has a different view on this.
On the one hand, the English High Court held that the hardware ANN clearly did not involve any computer program. In the case of hardware ANNs, if the invention in question was implemented solely through a hardware ANN, the invention should not be excluded from the scope of patent-eligible subject matter, which was also recognized by the UKIPO.
On the other hand, the English High Court held that instead of executing a series of instructions or codes pre-set by humans, the software ANN continuously adjusts node parameters through self-training to achieve the intended file recommendation function. The court held that software ANNs essentially operate on a different level than the underlying computer software, in the same way as hardware ANNs, and that if hardware ANNs are not computer programs, then software ANNs are not either. Accordingly, the English High Court concluded in its judgment that the patent application in question was not a computer program and that the exclusion in section 1(2)(c) of the Patents Act 1977 did not apply to the present case at all.
Second, assuming that the invention in question is a computer program, then if it solves the external technical problem of the computer and provides an external technical effect, it is not a claim to the computer program itself, but should also fall within the scope of the subject matter of patent protection.
Further, the English High Court held that if the invention solves an external technical problem of a computer and provides an external technical effect, it is not a claim to the computer program itself and should not be excluded from the scope of patent protection, which is similar to the Protecting Kids case,[3] in which the invention in this case has an external technical effect (contribution) by transferring data outside the computer system in the form of file transfer, has a substantial technical contribution, and does not belong to the computer program "itself" (as). such), which should not be excluded from the scope of patent eligibility subject matter.
2. Rules for the examination of the eligibility of the subject matter of protection of artificial intelligence inventions under English law
For a long time, judging from the jurisprudence in the European region, the difficulty in the patentability of AI technology lies in the analysis of the underlying algorithm and the "computer program" of the claimed claims. Article 52 of the EPC provides for "patentable inventions": paragraph 1 of the European Patent Convention provides that "a European patent shall be granted for any new invention that is inventive and capable of industrial application". Paragraph 2 provides that "in particular, the following shall not be considered as inventions within the meaning of paragraph 1: a) the discovery of scientific theories and mathematical methods; "If the technology in question is found to be a computer program, it should be excluded and denied patent eligibility under Section 52 of the European Patent Convention, which corresponds to Section 1 of the UK Patents Act 1977.
The English courts have stricter rules for the examination and determination of patents involving technology such as computer software or computer programs, and the English courts are more conservative and rigid in the application of Section 1 of the English Patent Law to the exclusionary provisions of Section 1 of the English Patent Law compared to the application of Section 52 of the European Patent Convention ("EPC"), which is homologous to the European Patent Office. The rules for determining computer programs under English law follow the European Patent Office's "technical effect approach" and have their origins in the VICOM case,[4] and the English courts' subsequent rules for determining the exclusionary provisions in section 1 of the Patents Act 1977 in Aerotel Ltd v. Macrossan's Application are as follows:
a. Claim interpretation;
b. identify the contributions that the claims may provide;
c. Analyze whether the contribution is based solely on the patent eligibility exclusion. If the result is positive, the patentability is denied;
d. Analyze whether the above contributions are technical.
The AT&T case involved a patent application for a "Content Agent Hosting Service System", which was essentially a software program that selects multimedia files to be delivered to customers based on compatibility with their devices connected to the multimedia provider. The Court refined the Aerotel Rules:
a. Whether the required technical effects have a technical effect on the physical world beyond the digital computer;
b. Whether the required technical effect works at the structural level of the computer, in other words, whatever the data processed or the program run, the intended technical effect will ultimately be produced. This claim shall not rely solely on or depend on the data, program or algorithm entered[5];
c. Whether the required technical effect causes the computer to operate in a completely new way. This item is an example of a "technical contribution or technical effect beyond a computer";
d. Whether there has been an increase in the speed or reliability of the computer. This item is an example of a "technical contribution or technical effect beyond a computer";
e. Whether the technical problem to which the patent at issue is addressed is overcome, rather than merely circumvented, by the claimed invention.
To sum up, the English courts' patentability review rules for technology related to computer programs are that a claim is patentable if it produces a technical effect or technical contribution in a way that goes beyond the computer, and the technical effect or technical contribution is not entirely dependent on or based on the computer data or computer program entered. In the Emotional Perception AI case, the technology in question was patentable because it automatically matched music for real-world users, and the matching process did not rely entirely on the input data or program, but was calculated and output through the ANN. In the Emotional Perception AI case, the UKIPO has temporarily cancelled the Guidelines for the Examination of Patent Applications Related to Artificial Intelligence Inventions[6], pending consideration of the decision. The Court's decision is very favourable to AI innovators and, if adopted, could change the way the UK examines the eligibility of AI invention patents.
3. Comparison of the progress of the judgment of the object of AI protection in the UK with China, the European Union, Japan, and the United States
(1) China's rules for the examination of the eligibility of AI patents
The essence of artificial intelligence lies in the combination of algorithm models and application scenarios, and the core is algorithm innovation. Generally speaking, AI technology is a comprehensive technology bundle based on technical algorithms and integrated into multiple technical fields and different functional dimensions driven by "big data" and "big computing", among which the basic algorithm as the basic layer is the core technological innovation of such technology, and it is also the claim itself asserted in the AI patent application. However, AI algorithms, as claims, may be identified as rules and methods of intellectual activity,[8] mathematical methods,[9] or computer programs, etc., and thus face controversies such as whether and how to apply the exclusion rules of the traditional patent legal systems in various jurisdictions, and whether the eligibility of relevant AI patents should be excluded.
On December 31, 2019, the State Intellectual Property Administration of China (CNIPA) issued the Announcement on <专利审查指南>Amendments (hereinafter referred to as "Announcement No. 343")[10], which in Chapter 9 of Part II, "Relevant Provisions on the Examination of Invention Patent Applications Containing Algorithmic Features or Business Rules and Method Features", the patent eligibility review of invention patent applications containing artificial intelligence algorithms shall be conducted in accordance with the following steps and rules: (1) in accordance with the Patent Law Article 25, paragraph 1 (b), examines whether the claims are "rules and methods of intellectual activity". Specifically, if a claim contains technical features in addition to abstract algorithmic features, the claim is not a rule or method of intellectual activity as a whole. If the claims as a whole do not belong to the rules and methods of intellectual activity, they will proceed to the second step of examination. (2) Examine whether the claims are "technical solutions" in accordance with Article 2, Paragraph 2 of the Patent Law. Specifically, if the claim records that a technical means of using the laws of nature are used for the technical problem to be solved, and a technical effect that conforms to the laws of nature is obtained, then the solution limited by the claim is a technical solution, for example, the data processed by the artificial intelligence algorithm is data with exact technical meaning in the technical field, and the execution of the algorithm involves the use of the laws of nature to solve the technical problem. If the claim is a technical solution within the meaning of the patent law, it is patentable.
In the decision of the State Intellectual Property Office of China No. 10720 on the request for review, the State Intellectual Property Office held that "claim 1 requires the protection of an information notification method, which cannot avoid the use of manpower in the process of transmission of the report details for the existence of background art, that is, the transmittor must enter the other party's email address through the computer or enter the number on the fax machine, so it is time-consuming and laborious. There is no problem of efficiency, the problem to be solved by the method of claim 1 is to automatically transmit the notification information to the information receiving device of the unit that needs to be received after the report is edited in detail through the network system, so it belongs to the technical problem in the information transmission; and in order to complete the automatic transmission of the notification information including the report, the method adopts the basic database of the receiving unit, the information notification system judges whether a notification information is received, reads the notification information and takes out the relevant data of the receiving unit, finds out the corresponding contact information according to the relevant data of the receiving unit, and the information notification system transmits the notification information to the information receiving device and other meansThrough the control of the network system, the information receiving device, the information notification system, etc., the notification information is automatically transmitted to the receiving device, so the technical means that follow the natural law are utilized; Since the problems solved, the means used and the effects obtained by the method are all technical, claim 1 belongs to the subject matter of patent protection. "From the perspective of examination practice, the overall scheme of claimed protection is regarded as a whole to judge whether it has technical characteristics, and the judgment of whether it has technical characteristics is carried out without considering the prior art situation.
Therefore, for an invention patent application containing an artificial intelligence algorithm, China first examines whether it falls within the scope of patent eligibility exclusion, i.e., the rules and methods of intellectual activities, from a reverse perspective, and if the application does not fall within the scope of patent eligibility exclusion, it further examines whether it constitutes a technical solution within the meaning of the Patent Law from a positive perspective. Only by passing the two-step examination from both the positive and negative perspectives can the patent application for an AI algorithm invention patent be patentable. In addition, when examining an invention patent application containing an artificial intelligence algorithm, the principle of overall examination should be adhered to, that is, the technical features and the algorithm features should not be simply separated, but all the contents of the claim should be taken as a whole, and the technical means involved, the technical problems solved and the technical effects obtained should be analyzed.
(2) The examination rules of the European Patent Office on the eligibility of artificial intelligence patents
The European Patent Office is more flexible in the application of the exclusions under Section 52 of the EPC, and the relevant jurisprudence and official guidance are more mature. In the HITACHI/Auction Method case, the Board of Appeal ("the Board") of the European Patent Office ("the Board") determined the steps to determine whether such technology falls within the statutory exclusionary rule:
a. The technology in question must be an "invention", i.e., it must be inventive and industrially applicable;
b. "Invention" means "technical subject matter having a technical character";
c. Confirmation of whether the technical subject matter of the claims falls within Article 52(1) of the EPC (i.e., the determination that it does not fall within the exclusions of Section 52) must be carried out before the novelty, inventive step and industrial applicability tests are carried out.
In Clipboard Formats I,[11] the Windows Clipboard function was found to be patentable, and the Examining Board of the European Patent Office found that any technical means were "inventions" and that "computer systems" were technical approaches in the present case. A "method of utilizing a computer system" means a series of steps that are actually executed to achieve a certain technical effect, rather than a string of computer execution code that has the possibility of achieving a technical effect only when the computer is running internally.
Therefore, based on the examination rules of the European Patent Office, a method of the appropriate technical route is an "invention", and any "invention" that is inventive and industrially applicable is patentable, even if it involves exclusions such as computer programs, mathematical methods or business methods, and the exclusions in Section 52 of the EPC are therefore no longer a major obstacle to the examination of AI patent applications before the EPC.
Judging from recent cases, in practice, the European Patent Office mostly focuses on patent novelty and inventive step, and applications for such technologies are often found to be incompatible with the requirements of the inventive step because they do not have a specific field of application, the claims lack a description of the model, or the training process for training the model. [12] With regard to the "inventive step" of an AI, the focus of the examination is that the core technical features (mostly computer programs or basic algorithms) must contribute to the technical characteristics of the invention in order to make the technology in question technically effective. In response to this requirement, the Examining Board of the European Patent Office confirmed:[13]
a. the technology in question achieves a technical purpose by applying it to a technical field;
b. Achieving technical effects by applying them to a specific technical application.
In addition, it is often difficult to grasp the level of detail of AI technology claims as to whether the technical features are adequately disclosed,[14] because most of the information is sensitive and complex because it usually involves the selection of network architecture, input representation, and/or training data. This also makes it difficult for general-purpose artificial intelligence, as well as related technologies that make breakthroughs purely at the level of basic algorithms, to pass the patent examination before the European Patent Office.
It can be seen from this that the Reexamination Board of the European Patent Office held that AI itself is not technical, and AI that only provides a general purpose for AI or only improves the efficiency of algorithms is not considered technical. It is generally more difficult for general AI technologies to meet the examination rules established by the current precedents, as they are different from field-specific AI technologies that are customarily patented, such as controlling specific technical systems or processes (e.g., X-ray equipment or cooling processes of steel); Digital audio, image or video enhancement or analysis, medical diagnostics, etc. [15] Overall, however, the EPO has a positive attitude towards AI patents that enable technical contributions to generalized AI or underlying algorithms, and "there is no doubt that ANNs can provide useful technical tools for automating human tasks or solving technical problems." However, in most cases, this needs to be fully specified, especially with regard to the training data and the technical tasks being processed. ”[16]
(3) Japan's examination rules on the eligibility of AI patents
Japan adopts a "technical attribute testing method" similar to that of the European Union in determining the subject matter of patent protection, requiring it to constitute a technical solution in order to obtain patent law protection, and at the same time adopts very loose applicable standards for AI technology. Schemes that combine the basic algorithms of artificial intelligence with application scenarios are generally considered to be "technical solutions" that can be protected by patent law.
As mentioned above, in March 2018, the JPO issued the "Examples of Examination Guidelines for AI-related Technologies", which provides the patentability examination criteria for solutions that combine basic AI algorithms and application scenarios based on specific cases. Among them, the invention and creation of the combination of artificial intelligence algorithms and application scenarios are clearly the object of protection of the patent right. An example given in the "Examples of Review Guidelines for Artificial Intelligence Related Technologies" is "a training model based on dormitory reputation based text data to prompt computer equipment to output qualified dormitory reputation values", which uses neural networks to process text information, analyze the frequency of occurrence of special words that reflect dormitory reputation in the text data, extract fields about dormitory conditions, comprehensively analyze all fields, and use the training model to obtain a reasonable dormitory reputation evaluation value. According to the "Examples of Examination Guidelines for AI-related Technologies", the above-mentioned scheme uses hardware resources to realize the information processing of software, and is a "technical solution" that can be protected by patent law.
(4) U.S. rules on the examination of the eligibility of AI patents
The U.S. adopts the Simulated Prior Art Exclusion Test for the determination of patentability of AI technology, and the part involving abstract concepts is prescripted as prior art that does not contribute any to the novelty and inventiveness of the patent, and is excluded from the judgment of novelty and inventive step.
Section 101 of the U.S. Patent Act provides that any invention or discovery of any new and useful process, machine, product, synthesis of matter, or any new and practical improvement thereof, may be granted a patent under the conditions and requirements set forth in this Act. Case law defines each of the four categories of subject matter that can be protected by patent law: process, machine, product, and synthesis of matter. That is to say, method refers to the way of dealing with certain substances so that they produce a certain result, which is a certain act or series of actions that act on the object substance, causing it to change and produce different states or things[17], machine, refers to the machine of the whole, one or several parts of the whole machine, the merging of one or several parts, and the merging of the original parts to form a machine[18], and the synthesis of matter, which refers to all the articles synthesized by two or more substances, and ...... All synthetic items, whether they are the result of chemical synthesis or mechanical physical synthesis, whether they are gases, liquids, powders, or solids[19][20].
At the same time, through judicial practice, the U.S. Supreme Court has clarified that the subject matter for which no patent is granted includes natural laws, physical phenomena, and abstract concepts[21]. The 2014 Alice case of the U.S. Supreme Court formed the basic logic of the "Simulated Prior Art Exclusion Test", which presumpposes the above-mentioned natural laws, physical phenomena, and abstract concepts as prior art that does not contribute any to the novelty and inventiveness of the patent, excludes them from the judgment of novelty and inventive step, and requires the novelty and inventive step of other parts of the claims. [22] In the case of AI technology invention patents, it is particularly necessary to determine whether they are "abstract ideas", that is, how to distinguish between invention patents that contain algorithmic features or features of business rules and methods protected by patent law and innovative creations that are abstract concepts that are not protected by patent law. This is due to the fact that the algorithm itself is more similar to mathematical logic, and there are certain differences from the technical means of solving technical problems.
However, some scholars believe that the traditional criterion of patent law to distinguish between abstract ideas and concrete technologies does not deny the object attribute of computer program algorithms, as many scholars imagine. Program algorithms are concrete method steps that run a physical system (computer) that is independent of the human brain, not abstract rules of thought. The execution of the program algorithm will lead to a "change in the state of matter" in the sense of traditional patent law. Therefore, the program algorithm meets the aforementioned traditional standards and can successfully pass the subject review. [23] The legal practice of the United States also confirms this view.
In summary, for English law, the Emotional Perception AI case marked a breakthrough in the patentability of computer programs, mathematical methods, and artificial intelligence technologies, and Section 1 of the Patents Act 1977 was therefore the focus of controversy in the judgment of this case. It is highly likely that this case will face further appeal, and the relevant parties will continue to monitor the further determination of the English Court of Appeal in this case. For other major intellectual property powers in the world, AI technologies still face the problem of how to have the patent feature of inventiveness after breaking through the relevant statutory exclusion restrictions, and the patentability of general-purpose AI or AI technologies that embody technological contributions with basic algorithms still needs to be gradually expanded and confirmed by the above-mentioned jurisdictions.
Annotations (scroll up and down to view)
【1】[2023] EWHC 2948 (Ch).
【2】Section 1(2). Patents Act 1977.
[3] [2012] RPC 13.: A communication analytics engine that involves sniffing and alerting children about inappropriate information when they browse a computer.
【4】T 0208/84.
【5】e.g. Fujitsu Ltd’s Application.
【6】https://www.gov.uk/government/publications/examining-patent-applications-relating-to-artificial-intelligence-ai-inventions. (Accessed: 5th December 2023)
[7] Liu Xin and Qin Chuxiang, "Patent Law in the Era of Artificial Intelligence: Problems, Challenges and Responses" [J], Electronic Intellectual Property, No. 1, 2021.
[8] Article 25 of the Patent Law.
【9】ibid. [n.1].
[10] Announcement No. 343 of the State Intellectual Property Office.
【11】T 0424/03.
【12】T 0702/20.
【13】T 2330/13.
【14】T0509/18.
[15] European Patent Examination Guidelines, Part G Chapter II, Section 3.3.1.
【16】Ibid, 20.
【17】Cochrane v. Deener, 94 U.S. 780(1877).
【18】Union Sugar Refinery v. Matthesson, 24 F.Case 686(C.C. Mass,1865).
【19】Diamond v. Chakrabarty, 444 U.S. 303, 206 U.S.P.Q. 193.
[20] Li Mingde, "American Intellectual Property Law (Second Edition)" [M], Beijing: Law Press, April 2014, p.37.
【21】ibid. [n.19]
[22] Di Xiaofei, "Analysis of the Patentability of Artificial Intelligence Algorithms: Distinguishing Abstract Concepts from the Perspective of Knowledge Production" [J], Intellectual Property, No. 6, 2020.
[23] Cui Guobin, "Abstract Ideas and Concrete Technologies in Patent Law: Analysis of the Object Attributes of Computer Program Algorithms", Journal of Tsinghua University (Philosophy and Social Science Edition), No. 3, 2005.