My colleague Juho Aalto outlined the basics of patenting inventions related to AI solutions in his blog post “Artificial Intelligence as Part of a Patentable Invention.”
This post continues Juho’s blog, aiming to clarify what must be disclosed about an AI solution in a patent application.
Parts of a patent application
A patent application consists of a technical description of the invention, drawings, and patent claims. The patent claims define the scope of the protection, while the technical description provides more detailed information about the implementation of the invention in such a way that a skilled person in the field can understand it. (I have written about the concept of a “skilled person” in my blog post “The Skilled Person – Who Are They?”)
Description of the technical problem or environment
Since computer-implemented inventions must solve a technical problem, a mere reference to, for example, AI, a neural network, or a machine learning model is not sufficient—neither in the technical description nor in the patent claims. Therefore, the technical environment or technical problem in which the AI solution is used must be described, for example: “A method for purifying water using a neural network model” or “Facial recognition using a machine learning algorithm.”
Sufficiency of description is essential
Sufficient technical description is also one of the features of an acceptable patent application. Under patent law, the text must provide at least one example of how the invention is implemented. Therefore, the AI solution used in the invention must be described as precisely as possible in the technical description of the patent application.
An AI solution cannot, therefore, be described merely as a black box that takes input data and produces an output set without describing the process in more detail. Referring back to the facial recognition example, it is not sufficient to say that image data is received, a machine learning algorithm is applied to analyze it, and an output is generated stating whether or not the image contained a human face.
What should at minimum be disclosed about the AI implementation in a patent application?
To ensure that the AI-driven process is comprehensively described, the technical description should define the data used (including its variables, source, and structure); the correlation between the input data and the output set; possible preprocessing steps such as normalization or harmonization; the training method and training data used; test examples demonstrating the technical effect; as well as the interfaces and/or data transfer channels between the AI solution and other technical components—especially when the AI output controls another technical system or device.
Using the facial recognition example, the features to be described would include identifying the image file and its source; the structure of the AI model used, e.g., the neural network and its parameters, and how the image file is processed within that structure; but also the data on which the neural network was trained and its origin; and the target of the AI’s output—what is being controlled, measured, or achieved with it. For example, facial recognition might be used to identify a person, which in turn controls an automated access management system.
When discussing the patenting of an AI solution with a patent attorney, it is worth being prepared for the above-mentioned points—they will certainly be asked!
