[Supervised by a patent attorney] What is the deep relationship between generative AI and patents? Thorough explanation from infringement risks to protection of inventions

The evolution of ``generative AI'' such as ChatGPT and Midjourney is dramatically changing our business and development processes. It can be said that the use of AI is now essential in improving business efficiency and creating new services.

However, because technology is evolving so rapidly, it is also true that legal arrangements and interpretations have become complicated. Particularly in the field of **patents**, along with new opportunities, there are also risks that cannot be overlooked.

• "Can ideas created by AI be patented?"

• "If we incorporate generation AI into our service, will it infringe on the patents of other companies?"

• "How can we monopolize AI-related inventions?"

For business owners and development managers who have questions like these, this article provides a detailed explanation of the current status of ``Generative AI and patents''** from the perspective of a patent attorney, as well as strategies to protect your business.

1. Current status of "intellectual property" behind the generative AI boom

First of all, intellectual property rights issues related to generative AI can be broadly divided into "copyrights" and "patent rights." ``Unauthorized use of training data'' and ``generation of similar images,'' which are often discussed in the news, are mainly copyright issues.

On the other hand, patent rights are extremely important from the perspective of protecting the core technology of your business or overcoming the technological barriers of other companies.

AI technology itself and inventions using AI

In the world of patents, it is necessary to distinguish between the following two types.

1. Invention of AI technology itself (core technology)

   • New neural network structures, more efficient learning algorithms, etc. This is an area where platform operators such as Google and Microsoft are fiercely competing.

2. Inventions using AI (applied technology)

   • ``A system that automates the design of specific products using image generation AI'' and ``An app that supports medical diagnosis using language models.'' For many Japanese companies, this is the main battleground for acquiring patents.

Currently, the number of applications for AI-related inventions at the Japan Patent Office (JPO) is steadily increasing. This is evidence that companies are increasingly trying to protect their AI-based business models with patents and eliminate competitors.

2. Can "inventions created by AI" be patented?

Here are answers to frequently asked questions. "If you give a problem to a generative AI, can the solution (idea) that the AI comes up with be patented?"

"Inventor" must be human

In conclusion, under current Japanese patent law (and the laws of major countries),AI itself is not allowed to file a patent application as the "inventor".

The general interpretation is that an "inventor" under patent law is limited to a "natural person (human being)." As a famous example, there was an application in which the AI ​​"DABUS" was named as the inventor, but it was rejected in many countries including Japan.

Human involvement makes things special

However, this does not mean that all inventions using AI cannot be patented.

If a human uses AI as a "tool", gets an idea, and completes a specific configuration, the inventor would be a "human" and could be able to obtain a patent.

What is important is "Which part had human creative contribution (ingenuity)?".

Even if you simply enter one word of the prompt and apply the result as it is, there is a high possibility that it will not satisfy the "inventive step (something that cannot be easily thought of)" or "enforceability requirements" and will be rejected. However, when humans make improvements based on the output of AI or incorporate AI into a system to solve a specific problem, it becomes a great invention.

3. "Patent strategy" to protect business using generative AI

What kind of patent strategy is needed to protect AI solutions developed in-house or services incorporating AI from imitation by other companies?

① It is difficult to obtain rights for "trained model" alone

In practice, it is extremely difficult to identify and obtain rights for the "trained model (collection of parameters)" itself in patent claims. Also, copyright protection as a program is limited.

② The aim is "specific uses of AI" and "pre-processing/post-processing"

The following points make it easy to obtain a patent for AI-related inventions and have high business value.

• Application to specific use cases

  • Rather than just "predicting with AI," we want to create a system that inputs data from 〇〇 sensors and weather data from △△ into AI, predicts when equipment will fail, and automatically orders maintenance parts.Utilization of input data, processing, and output results: Things that are specifically linked tend to be patentable.

• Improvements in data pre-processing and post-processing

  • When unique data processing (pre-processing) is performed to improve the accuracy of AI, or when the interface part processes the AI output to make it easier for users to see (post-processing), technical features are likely to emerge and become patentable points.

③ Overcoming the barrier of "inventive step"

AI technology is becoming more generalized. There is a growing risk that simply saying "AI was applied to ____ work" will be rejected as "any person skilled in the art would have thought of it (no inventive step)".

To avoid this,

"Why are you entering that data?"

"Why that learning method?"

"As a result, what are the significant effects compared to conventional technology?"

must be logically asserted in the specification. This is the highlight of a patent attorney who is strong in AI technology.

4. Can I be sued without knowing? Risks of using generative AI

Patents are not only important for "defense" but also for "offense" (attacks from other companies). Is there a risk of patent infringement just by using generative AI?

System use and patent infringement

In general, users (end users) of cloud services such as ChatGPT rarely infringe on the "AI algorithm patents" of other companies simply by using those services. This is because many AI patents specify the configuration of the "system providing side (server side)".

However, you need to be careful in the following cases.

• When creating your own app by incorporating the API

  • When you develop and sell your own apps using APIs such as OpenAI, there is a possibility that the functions and processing flows of those apps may conflict with the "patents for AI-based services" held by other companies.

  • For example, another company may already have a patent for ``a flow that uses a chatbot to automate customer response and link it with a specific database.''

Black box risk

AI, especially deep learning, tends to be a black box in its processing. Even if you think your company has developed the technology, there is a risk that the resulting process will be considered to be the same process as existing patented technology.

Therefore, before releasing a new service, it is essential as a management safety measure to conduct an FTO investigation (infringement prevention investigation) to confirm whether it has touched the patent network of another company.

5. Patent applications in the AI era, points to choose a patent attorney

Technology related to generative AI is a complex intertwining of knowledge in IT, mathematics, and law. There are an increasing number of cases in which conventional mechanical and chemical knowledge alone cannot handle the situation.

Who is a patent attorney who specializes in AI patents?

When requesting an AI project, we recommend checking the following points.

1. Understanding the "IoT/AI Examination Handbook"

   • The Japan Patent Office has clarified the examination standards for AI-related inventions. Do you know these things well and understand what kind of logic should be used to describe them to make it easier to obtain a patent?

2. Understanding technical terminology

   • Can you understand terms such as "supervised learning," "reinforcement learning," "token," and "vectorization" without explanation, and accurately incorporate them into the specification?

3. Proposals from a business perspective

   • In addition to simply obtaining a patent, can you propose ``to what extent should we obtain a patent to keep competition in check'' and ``how to create strong rights that will not be designed to circumvent it?''

Early start is the difference between winning and losing

Technology in the AI field deteriorates quickly, so patent applications are also a matter of speed. By consulting with a patent attorney at the "idea stage" or "PoC (proof of concept) stage," rather than consulting after development is complete, you can build a stronger patent network while correcting the direction of development.

6. Summary: Consider AI utilization and intellectual property strategy as a set

Generative AI has made it easy for anyone to use advanced technology. That's why where you find your company's unique value and how you legally protect it will determine your future struggle for survival.