“Don’t be afraid to fail. Be afraid not to try.”
— Michael Jordan

Patentability in Artificial Intelligence

As technology advances, machine learning continues to transform industries, creating numerous opportunities for innovation. However, the intersection of machine learning and intellectual property, particularly in terms of patentability, remains a complex and evolving landscape.

Exclusions from Patentability

In the context of patent law, certain categories of inventions are excluded from patentability. These exclusions are consistent across major patent offices, such as the European Patent Office (EPO) and the United States Patent and Trademark Office (USPTO). The following are commonly excluded:

• Scientific theories and discoveries
• Mental acts
• Computer programs
• Mathematical methods

These exclusions are framed as “such subject-matter or activities as such,” which implies that certain inventions may be deemed non-patentable if they do not meet specific criteria.

Are AI-Based Inventions Patentable?

The short answer is yes—AI-based inventions can be patented, provided they meet the requirements set forth in the European Patent Convention (EPC). The key requirements include:

• Technical Character: The invention must be implemented on a computer or through technical means.
• Technical Effect: There must be a technical effect achieved that leads to inventiveness.

Technical Character and the EPO Guidelines (first hurdle)

According to the EPO guidelines, if a claim pertains to a method that employs technical means (like a computer) or to a device, the subject matter possesses a technical character and is thus not excluded from patentability under Articles 52(2) and (3) of the EPC.

Assessing Technical Effect in AI (second hurdle)

When evaluating the inventive step of an AI-based invention, it is crucial to consider all features contributing to its technical character. If a mathematical method is employed, its contribution to the technical effect must be examined. A mathematical method may enhance the technical character of an invention if it produces a technical effect serving a specific technical purpose, either through its application to a technological field or by being tailored for specific technical implementations.

Examples of Technical Purposes

Some examples of how a mathematical method may contribute to technical character include:

• Controlling specific technical systems or processes: For instance, operating an autonomous vehicle.
• Digital enhancement: Applications in audio, image, or video processing.
• Speech recognition: Techniques for separating sources in speech signals.
• Simulation: Modelling the behaviour of defined classes of technical items.

Case Law Examples

Several case law examples elucidate the principles of patentability concerning AI and machine learning:

• T 598/07: Heartbeat Monitoring Method Based on a Neural Network

This case involved a neural network in a heart monitoring device to identify irregular heartbeats. The Board of Appeal (BoA) recognised the claimed subject matter as technical, showcasing a specific application of AI technology.

• T 1784/06: Classifying a Set of Data Records

In this instance, the Board determined that the classification of data records was non-technical, as it served only the purpose of classifying without any technical application.

• T 1286/09: Image Classification of a Digital Image

This application focused on enhancing the diversity of images for training a semantic classifier. The BoA concluded that the claimed subject matter contributed to solving a technical problem, thus qualifying as technical.

• EP 1569128 B1

This case involved optimising machine learning techniques using a graphics processing unit (GPU). The claimed subject matter provided a technical effect by adjusting learning rates based on measured progress.

Conclusions

When drafting or reviewing applications for AI in both the European and US contexts, the following considerations are critical:

• Specificification drafting: Clearly articulate the application of the method, including its technical components.
• Technical Advantages: Highlight any advantages related to the computing or electronic devices involved.
• Sufficiency of disclosure: Ensure that training and data sets used in the method are well-defined and sufficient.

By adhering to these principles, innovators can better navigate the complexities of patentability in the realm of machine learning and artificial intelligence.

For further inquiries, please contact Hanane Fathi Roswall.