Patent applications with and from artificial intelligence

Patent applications with and from artificial intelligence

The rapid development of artificial intelligence (AI) has not only enabled technological advances, but has also created new challenges and opportunities in the field of patent law. This raises the following questions, among others: How can AI be successfully protected by patent applications? And what influence do AI-related inventions have on patent applications?

Artificial intelligence is regarded as a branch of computer science that poses specific challenges for patenting. Since AI is based on computational models and mathematical algorithms that are considered abstract and are generally excluded from patentability (Art. 52 (2) EPC), the question arises as to how AI can nevertheless be protected by patents. Patents can be granted in particular when AI goes beyond the abstract and is used to solve a technical problem in a technical field. Typical examples of this are the classification and analysis of image, video, audio or sensor data as well as the control of specific systems.

A look at the latest decisions of the European Patent Office (EPO) illustrates the challenges and opportunities in the patenting of AI.

The landmark decision J 0008/20 (Designation of inventor/DABUS) initially stipulated that only a natural person can be designated as an inventor within the meaning of the European Patent Convention (EPC) (also according to the German case law of the Federal Court of Justice X ZB 5/22).

The development underlying decision T 161/18 concerned the transformation of peripheral blood pressure curves into equivalent aortic pressures using an artificial neural network (ANN). The EPO decided that the training data of the ANN was insufficiently disclosed and concluded that the invention was considered non-executable within the meaning of Art. 83 EPC. It can be inferred from the decision that specific details about training data and the input/output parameters as well as processing steps of the ANN are considered crucial for the requirements of practicability.

Another decision T 1669/21 deals with a method for determining the condition of a refractory lining in a metallurgical vessel. Here, the EPO decided that the term "computational model" includes both analytical (non-AI based) and machine-learning models. This decision highlights the need to describe in detail the specific model architecture and concrete input/output parameters.

Decision T 2803/18 concerns a method for the automatic detection of incontinence. The EPO decided that the claimed technical effect, namely an improvement in the accuracy of moisture estimation, was not sufficiently demonstrated by comparative data. This decision underlines the importance of comparative data to demonstrate improvements over the prior art.

In summary, these decisions show that certain "rules" must be observed when drafting patent applications in order to ensure the patentability of AI inventions. Patent applications should provide detailed information about training data, model architectures and technical improvements over the prior art on which the patent application is based. In addition, the naming of human inventors remains essential, even if AI plays a central role in the invention.

‍