Navigating the New Frontier: AI Patentability at the European Patent Office

Understanding patentability of AI at the EPO

AI and machine learning (ML) hinge on computational models and algorithms, characterized by their abstract mathematical nature. The EPO states that although these mathematical methods address technical issues across various sectors, their patentability is limited when claimed alone, as per Art. 52(2)(a) and (3) European Patent Convention (EPC). This limitation arises because patent claims based solely on abstract mathematical methods, without any technical application, lack the required technical character for patent eligibility.

This changes when patent claims involve technical means, like a computer, or relate to a specific device, as this grants the claims a technical character and overcomes the non-patentability restrictions of Art. 52(2) and (3) EPC. Further, it's crucial that for a mathematical feature to enhance an invention's technical nature, it must directly contribute to a technical effect or address a technical problem.

Therefore, AI needs to not only be executed by technical means but also demonstrate a technical purpose in order to be patentable. Clearly despite its inherent mathematical basis, AI can significantly contribute to technological fields. For instance, using a neural network in a heart monitoring device demonstrates a technical contribution, which would therefore be considered patent eligible subject matter.

Assessing inventive step in AI

The concept of an inventive step is a fundamental pillar of patent law, designed to ensure that patents are awarded for innovations that represent a significant advancement over existing technologies. For AI inventions, this entails demonstrating that an algorithm or AI system achieves a technical effect beyond mere theoretical application, emphasizing the practical utility of AI technologies in solving problems that the EPO considers technical in nature, excluding for example problems that are purely financial, commercial, administrative or organisational in nature.

Inventive step assessment for AI patents requires a detailed examination of how an AI's underlying mathematical model or training of such model contributes to the technical character of the invention. The guidelines state that for a patent claim to be considered inventive, it must not only utilize technical means (e.g. a computer) but also result in a specific technical effect that serves a specific technical purpose, by its application to a field of technology and/or by being adapted to a specific technical implementation. EPO lists the following examples of technical implementation: digital audio, image or video enhancement or analysis; providing a medical diagnosis by an automated system processing physiological measurements of technical applications; assigning the execution of data-intensive training steps of a ML algorithm to a graphical processing unit (GPU); and preparatory steps to a standard central processing unit (CPU) to take advantage of the parallel architecture of the computing platform.

Sufficiency of disclosure and training data

Closely intertwined with the inventive step is the principle of sufficiency of disclosure. In order to show technical contributions and inventive step the patent application must disclose appropriate details. The updated guidelines underscore the necessity for patent applications to provide enough information about the AI model and the characteristics of the training data to enable a skilled person to reproduce the invention. This requirement addresses the delicate balance between transparency and the protection of sensitive information inherent in AI technologies. The guidelines also suggest alternatives for disclosing sensitive data, such as the learned coefficients or weights of a model, offering patentees a pathway to protect their innovations while adhering to the sufficiency of disclosure requirements.

The EPO provides the following explanation, which outlines well the requirement of sufficient disclosure for AI and ML inventions:

"The technical effect that a machine learning algorithm achieves may be readily apparent or established by explanations, mathematical proof, experimental data or the like. While mere allegations are not enough, comprehensive proof is not required, either. If the technical effect is dependent on particular characteristics of the training dataset used, those characteristics that are required to reproduce the technical effect must be disclosed unless the skilled person can determine them without undue burden using common general knowledge. However, in general, there is no need to disclose the specific training dataset itself" (EPO Guidelines: G-II-3.3.1).

AI inventorship: affirming the human element

Another significant update concerns the criteria for inventorship in the context of AI inventions. Reflecting the legal stance established in J 8/20 (DABUS), the updated guidelines affirm that an inventor must be a natural person. This clarification addresses the burgeoning debate on AI inventorship and aligns the EPO's position with the broader view that AI systems, despite their capacity to generate inventive outputs, cannot be recognized as inventors.

Practical implications for AI patent applicants

AI patent applicants are encouraged to:

• clearly articulate the technical contribution of their AI inventions, specifically the technical effects achieved by AI algorithms and the role of training data
• ensure patent applications provide sufficient technical description of the AI model, input and output data, training data and training process
• recognize the necessity of human inventorship, reflecting on the attribution of invention in AI-generated innovations

Looking ahead

The 2024 guidelines reflect a nuanced approach to AI patentability, urging patent applicants to ensure their patent applications contain enough technical information for successful examination. While not a fundamental shift in practice, these updates highlight the importance of detailed technical disclosure in AI and ML inventions. As the AI landscape evolves, staying informed about these guidelines will be crucial for inventors and legal professionals, ensuring that AI innovations are effectively protected under the EPO's strict examination criteria.

Read more in our global trade secrets and patents series:

• The Unified Patent Court: emerging trends from the first nine months
• A changing landscape for AI patenting in the UK
• Standard Essential Patents: key developments for SEPs and FRAND commitments
• Supplementary Protection Certificates (SPCs): recent developments in EU case law
• The limitation of patents in ongoing revocation procedures in Spain
• What’s plausible? The UK Court of Appeal revisits the ever shifting concept
• Global trade secrets update