The patentability of artificial intelligence (AI) involves two primary areas: the ability to patent inventions in AI technology itself and the rules surrounding inventions made with the assistance of, or solely by, AI. A key distinction across most jurisdictions is that an inventor must be a natural person.
Patenting AI Technology
AI-related technologies are patentable provided they meet standard patentability requirements (novelty, non-obviousness, and industrial applicability) and do not fall into excluded subject matter categories, such as abstract ideas, mathematical methods, or computer programs “as such”.
- Technical Contribution is Key: To be patent-eligible, an AI invention must make a “technical contribution” or achieve a “technical effect”. This means it must solve a technical problem or operate a computer in a new and technical way, rather than just processing data or implementing an abstract algorithm.
- Examples of Patentable AI Inventions: Patents are granted for specific applications like neural network architectures for specific medical imaging tasks, improved training methods for large language models (LLMs), unique AI hardware components (like Google’s Tensor Processing Unit or TPU), and systems that enhance existing technical processes (e.g., noise reduction, object detection).
- Documentation: When patenting AI technology, it is recommended to clearly define the input and output, the neural network architecture, training data and cost functions, and the specific structural components to demonstrate the technical contribution.
Inventorship and AI-Assisted Inventions
A crucial legal challenge is the “human inventor requirement”.
- Human Inventor Required: In most countries, including the United States, the European Patent Office, and the UK, only a natural person can be named as an inventor on a patent application. Applications listing an AI system (like the “DABUS” AI) as the sole inventor have been rejected.
- AI as a Tool: When a human uses AI as a tool to assist in the inventive process, the resulting invention can be patentable, provided at least one human being made a “significant contribution” to the conception of the invention.
- Significant Human Contribution: Merely presenting a problem to an AI or simply reducing an AI’s output to practice is not enough for human inventorship. The human contribution must be a substantial part of the inventive concept, such as defining the specific problem, selecting the data set, setting unique parameters, and evaluating the results in a meaningful way.
The “Best Mode” Requirement
The “best mode requirement” in US patent law obligates the inventor to disclose the best method contemplated for carrying out the invention at the time of filing. For AI inventions, this means adequately describing how the AI works, including potentially specific data correlations, to enable a skilled person to reproduce the invention.
For detailed guidance, you can refer to the official resources from patent offices, such as the EPO’S AI resources page the USPTO’s AI resources page or the WIPO patent landscape report on Generative AI.
(This article was prepared with the assistance of AI).
