Japan Patent Office Case Examples on Artificial Intelligence Offer Guidance for Other Offices on Treating AI Inventions

By John M. Rogitz
February 28, 2019

The Japan Patent Office’s recent guidance on examination of artificial intelligence-related inventions could provide a useful framework for other patent offices struggling with how to treat new technologies

https://depositphotos.com/227010482/stock-photo-futuristic-silver-cyborg-touching-artificial.htmlThe Japan Patent Office recently added ten new case examples pertinent to artificial intelligence-related technology to Annex A of its Japanese Patent Examination Handbook. The examples are meant to facilitate understanding of the description requirements and the inventive step requirement in Japan as applied to AI-related inventions. In doing so, they provide a useful preview for how other patent offices might begin treating AI-related inventions. The examples are also very useful for any practitioner with clients in the AI space who intend to file in Japan.

Description Requirements

There are two prevailing points to take away from the new case examples in relation to the Japanese “description requirements” – e.g., enablement and written description.

(1)       Correlations

First, when an invention applies AI to a technical field, one should disclose a “certain relation, such as a correlation.” For example, it is not enough to say “run photos through a neural network to detect emotion on faces.” One should also give example correlations for the neural network to work with, such as “a smile means happy” and “a frown means sad.” However, the correlation can also be shown “in view of common general technical knowledge,” which obviously has its pitfalls and drawbacks.

(2)       Test Results Required

Second, when a computer product is presumed to have an AI function, one should show actual test results or other proof of validation of the model “unless an estimation result by AI can be a substitution for an evaluation on a product that has actually been made.”

Thus, for example, for a new glue with a specific composition obtained using a neural network to optimize ratios of constituents to thereby achieve certain cure times and strength, one should show test results from running the model, or proof that the model was otherwise verified as accurate.

Inventive Step Requirement

As far as the inventive step requirement goes in Japan in relation to AI inventions, Example 34 from the new case examples provides some guidance. Example 34 itself deals with providing an estimation system of a hydroelectric power generating capacity. The example seems to indicate that novel input data might satisfy Japan’s inventive step requirement. Claims 1 and 2 from this example will be discussed below to illustrate.

First, regarding Claim 1, according to Example 34 the claim does not have a patentable inventive step because, “[i]n the technical field of machine learning, it is well-known that an estimation process of an output in the future is carried out based on an input of time series data in the past, by using a trained neural network which has been trained with a training data containing an input of time series data in the past and a certain output in the future.”

Per Example 34, this result is reached based on the fact that, prior to the invention set forth in this example, “[i]t [was] well known that an estimation process of an output in the future is carried out based on an input of time series data in the past, using a trained neural network. The neural network has been trained with a training data containing an input of time series data in the past and a certain output in the future. [The first cited reference] and the well-known art are common with each other in estimating a certain output in the future based on an input of time series data in the past, with reference to a correlation among data. Therefore, a person skilled in the art could easily derive a configuration that enables estimation of a hydroelectric power generating capacity, by applying the well-known art to [the first cited reference] and adopting a trained neural network in substitution of a regression equation model [emphasis added].”

However, Claim 2 is different as it recites novel temperature data as input data that results in a “highly accurate” estimation output and thus satisfies the inventive step requirement: “The invention of Claim 2 uses a temperature of the upper stream of the river for estimation of a hydroelectric power generating capacity. There is no prior art found disclosing such use of a temperature of the upper stream of the river. Accordingly, it is not a common general technical knowledge that there is a correlation between a temperature and a hydroelectric power generating capacity [emphasis added].”

The basis for this finding is that:

“[g]enerally, an input of data of which correlation is unknown may cause a noise in machine learning. However, the invention of Claim 2 uses an input data containing a temperature of the upper stream of the river during a predetermined period between a reference time and a predetermined time before the reference time. This enables a highly accurate estimation of a hydroelectric power generating capacity, taking an increase of inflow rate due to meltwater in the spring into consideration. It is a significant effect that a person skilled in the art cannot expect [emphasis added].”

Accordingly, it [is] not considered to be a mere workshop modification that can be carried out in application of the well-known art to [the first cited reference] by a person skilled in the art to contain, in an input data in an estimation of a hydroelectric power generating capacity, a temperature of the upper stream of the river during a predetermined period between a reference time and a predetermined time before the reference time.”

Thus, when filing AI patent applications in Japan, one should disclose example correlations that the AI-related invention might make. For certain AI inventions, one should also disclose test results or validation of the AI model. Furthermore, novel input data and output data may be enough to establish an inventive step worthy of patent protection in Japan. Whether other patent offices will approach AI-related inventions in a similar way remains to be seen. Here in the United States., there are other issues that might need to be addressed first—namely, the current chaotic state of the law on patentable subject matter and whether artificial intelligence inventions will be rightly and consistently treated as patentable in the first place.

Image Source: Deposit Photos
Image ID: 227010482
Copyright: AndrewLozovyi 

The Author

John M. Rogitz

John M. Rogitz is a registered patent attorney with his own practice in San Diego, CA. His background includes preparation and prosecution of a large number of patent applications for high-tech Fortune 500 companies in a wide range of technologies. John has also been active on behalf of his clients in the acquisition of patent portfolios. He writes frequently for various publications on developments in patent law and also lectures on intellectual property for DeVry University. Previously, John was engaged in civil litigation at the Watkins Firm, a San Diego-based law firm. Prior to that, he worked as a web developer for Loyola Marymount University. John received his J.D. in 2009 from California Western School of Law.

Warning & Disclaimer: The pages, articles and comments on IPWatchdog.com do not constitute legal advice, nor do they create any attorney-client relationship. The articles published express the personal opinion and views of the author and should not be attributed to the author’s employer, clients or the sponsors of IPWatchdog.com. Read more.

Discuss this

There are currently No Comments comments.