Artificial intelligence inventions have become one of the emerging important patent issues. Per WIPO 2019 report, the number of AI-related patent disclosures increased significantly from 2012 to 2017 which reached more than fifty thousand cases. The patent disclosures relating to machine learning technologies accounted for nearly 90% of AI patent families over the past few years.
By referring to WIPO 2019 report, the Taiwan Intellectual Property Office (TIPO) has retrieved 1,567 AI-related invention patent applications of 2018 to provide an overview of artificial intelligence invention applications in Taiwan. According to TIPO analysis, the AI invention technologies for Taiwan patent cases include: neural network as the core technique (31 %), non-neural network machine learning (23%), hardware supporting neural network (11%), algorithms relating to neural network (3%). Their functional applications are mostly computer vision (49%), in line with international trends, followed by predictive analytics (29%). According to WIPO, predictive analysis mainly corresponds to the application fields of “Business,” “Life and Medical Sciences,” “Telecommunications,” and “Industrial and Manufacturing.” In comparison, Taiwan patent cases of predictive analysis are distributed in the application fields of “Industry and Manufacturing,” “Business,” and “Life and Medical Sciences.”
The TIPO also summarizes the following grounds of objection commonly seen in the examination opinion letters (office action) issued for AI-related patent applications:
| Grounds of objection | Description | Suggestion |
|---|---|---|
| The application does not meet the definition of invention | No technical ideas, mathematical methods by simple use of a computer to solve business problems, not utilizing the laws of nature | The invention must possess novel features and produce technical effects |
| The specification is not clearly and fully disclosed | Technical elements or details are not disclosed, and it is not able to understand as how the effects are achieved | The specification should be fully disclosed, enabling a person with ordinary knowledge to realize the effects contemplated by the invention |
| The claims are unclear | Vague meaning, the essential technical characteristics are not specified | It is difficulty to overcome the objection by filing a response without any amendments; the technical features of the claims could be recited and defined by incorporating the same from the specification |
| Lack of novelty or inventive step | It has been published in the articles, publicly implemented, or publicly known before the patent application;a person of ordinary skill belonged to the technical field can easily perform the claimed invention | According to the percentage of decisions rendered on AI patent applications, 10% were rejected for failing to file a response by the time limit, and among those applications responding with amendments, 25% were still rejected but 65 % were approved. Normally, the patent application may be allowed if the novelty and the inventive step could be resolved |
| Others | Description manner, patent subject matter, and unity of invention | Can be overcome by filing amendments |
In order to overcome the grounds for rejection, TIPO would suggest the Applicant to make amendments based on the examination opinions when filing a response, which will help to increase the chance of patent grant. TIPO has analyzed the amendments filed by the Applicants for patent applications received objection on lacking no novelty or inventive step. The amendments made by incorporating the technical features of the dependent claims that have been deemed as no rejection grounds for non-patentability into independent claims, or by deleting the groups of unpatentable claims, would achieved the highest success rate. For amendments filed by incorporating the technical features that were not included in the scope of the original application in accordance with the specifications, the success rate would be 80%. However, if the applicant did not make any amendments but only filed a response to refute the examination opinion, the success rate would be only about 50%.
Under the current patent system, AI-related inventions are often categorized as computer-implemented inventions. When applying for a patent, the Applicant should pay attention to the patent eligibility of the subject matter, the content disclosed in the specification for patent enablement, and the improvements or unexpected technical effects over the prior art technologies.
Nowadays, artificial intelligence techniques are being utilized in many areas. Per the statistics of WIPO report, AI inventions from a single patent family may appear in more than two technologies, functions or application fields. In Taiwan, one patent application must be filed and applied for each invention, unless two or more inventions are technically related to each other and have a broad inventive concept. In this regard, the patent applications of AI technologies may arise the problem of unity of invention due to the complexity of the technical content, especially for inventions using neural networks as the core technology.
Per Taiwan patent examination practices, the scope of patent claims is examined on a claim-by-claim basis. Basically, if the claims share the same or corresponding technical features, the patent application shall be considered to have unity of invention. For those claims that lack unity of invention, the Examiner will usually notify the Applicant before conducting prior art searches to determine whether these claims meet the requirements of novelty and inventive step. Under such circumstances, the Applicant can remove the said claims from the original application by filing a divisional application thereon. Accordingly, the amendments and divisional applications can be used to overcome the rejection reason that the original application does not comply with the unity of invention.
