With the fourth industrial revolution, applications of artificial intelligence (AI) are appearing in various fields and the technology is a hot topic of discussion. The Patent Examination Guidelines of the Korean Intellectual Property Office (KIPO) provide principles and examples related to patent eligibility and inventive step of AI related inventions, and KIPO has provided some AI-related patent examination cases in a recently-issued case book.

Although the definition of an AI-related invention is not explicitly specified, KIPO defines it as a computer/software-related invention for performing specific functions using AI training. Thus, the general Korean patent rules related to computer/software inventions can still apply to AI-related inventions. However, due to the AI-specific details involved in the various aspects of components, additional AI-specific guidelines are also provided by KIPO. For example, the recent case book generally states that AI-related inventions may involve generating trained models based on training data and a learning model (AI learning modeling invention), or performing specific functions by applying the training data and/or the trained model to various technical fields (AI application invention).

Patent Eligibility

Article 2 of the Korean Patent Act defines an “invention” as “the highly advanced creation of a technical idea utilizing the laws of nature”. Although software inventions (including AI inventions) are not considered to wholly utilize the laws of nature, the Patent Examination Guidelines prescribe requirements under which computer-related inventions are considered to be patent-eligible where information processing by software is implemented in detail using hardware. More specifically, the Guidelines state that computation or processing of unique information according to the purpose of use should be implemented by specific means or methods in which software and hardware co-operate, and thereby a unique information processing apparatus or an operating method thereof according to the purpose of use should be established. In addition, with regard to AI-related inventions, the Guidelines further require that the same effect must be repeatedly achievable without intervention based on human mental activity. Accordingly, if an algorithm corresponding to an AI invention is implemented using hardware, the invention may be recognized to be patent-eligible under these requirements.

Description Requirements 

Like inventions in other technical fields, AI inventions should follow the Korean patent rules which provide that a detailed description of an application should be clearly and concretely described such that a person skilled in the art can easily practice the invention, and the invention of claims should be supported by the detailed description.

The recently issued AI invention case book includes a case that did not meet the requirements with regard to AI inventions, explaining that if a claim recites features related to a correlation between input training data and output data of a trained model, if the detailed description merely lists a plurality of items of training data and does not concretely describe the correlation, the detailed description is not clearly described. Furthermore, in this case, the claim may not be considered to be supported by the detailed description.

Although these conclusions were based on general Korean patent rules, the examination result shows an example of explicit standards for determining whether an AI-related invention meets the description requirements.

Assessment of Inventive Step

As the Patent Act and Examination Guidelines related to inventive step are quite inclusive, neither the relevant parts of the Act nor the Guidelines have been majorly affected by the appearance of new concepts of invention. Although it seems that KIPO tends to maintain a high threshold when judging inventive step for AI-related inventions, the conventional general standard related to inventive step has not itself changed due to AI inventions. Nevertheless, the Patent Examination Guidelines and the case book provide some additional guidance related to the AI-specific characteristics.

As a general rule prescribed in the Guidelines, simply implementing a previously or generally known technology with artificial intelligence technology without specifying unique information processing performed via a learning model is not considered to involve an inventive step. Additionally, as mentioned above, AI-related inventions may involve components corresponding to the training data and the learning model. The cases introduced by KIPO fall into several categories based on the components:

1. Case 1 is where a claimed invention and a cited invention are substantially the same in terms of technical field and training data, but are different in terms of a learning model. If an improved effect can be achieved by the difference, inventive step can be recognized.
2. Case 2 is the same as case 1 except that the difference in the learning model is merely a simple design modification. In this case, inventive step is not recognized. However, if there are differences in specific features other than the training data and the learning model and the differences exhibit an improved effect, inventive step can be recognized.
3. Case 3 is where a claimed invention and a 1^st cited invention are substantially the same in terms of technical field and learning model, but are different in terms of training data. If a 2^nd cited invention has the feature corresponding to the training data, and there is no difficulty in combining the cited inventions, inventive step is not recognized.
4. Case 4 is where a claimed invention and a 1^st cited invention are substantially the same in terms of technical field and training data, but are different in terms of learning model. If a 2^nd cited invention has the feature corresponding to the learning model, and there is no difficulty in combining the cited inventions, inventive step is not recognized.

It seems that KIPO tends to apply a higher threshold when assessing inventive step concerning a difference in training data compared to that of a learning model.

Conclusion

Under Korean patent law, it currently appears that AI-related inventions are still considered to come under computer/software inventions, albeit with some unique additional aspects. Although there remains some doubt as to whether the Korean Patent Examination Guidelines sufficiently reflect the characteristics of AI-related inventions, such details or characteristics of AI-related invention are gradually being reflected more clearly. Accordingly, while complying with the current guidelines, it will be necessary to keep an eye on future developments in the Guidelines as well as related trends.