ARTICLE
7 May 2024

AI Inventions: The Importance Of Description, EPO Guidelines And Practical Considerations

NG
Novagraaf Group

Contributor

Novagraaf has been helping iconic brands and innovative organisations drive competitive advantage through intellectual property (IP) for more than 130 years. One of Europe’s leading IP consulting groups, Novagraaf specialises in the protection and global management of IP rights, including trademarks, patents, designs, domain names and copyright. Part of the Questel group, Novagraaf has 18 offices worldwide and a network of more than 330 IP attorneys and support specialists.
Meeting the requirements for sufficiency of description can be challenging for all types of EU patent applications, but especially ones for AI inventions.
Netherlands Technology

Meeting the requirements for sufficiency of description can be challenging for all types of EU patent applications, but especially ones for AI inventions. Catherine Caspar highlights the difficulty of accurately describing the training data used in AI models and summarises the practical guidance offered by the EPO's updated Guidelines for Examination.

Sufficiency of description is one of the important benchmarks for patent validity. Computer-implemented inventions, including those in the field of artificial intelligence (AI), are no exception to this requirement.

According to this criterion (defined by Article 83 of the European Patent Convention), an invention must be set out in the patent application in a manner sufficiently clear and complete for a person skilled in the art to be able to execute it.

AI inventions: Descriptions of training data

For AI inventions, one major difficulty here concerns the description of the training data used in AI models, including determining the necessary level of precision for descriptions of the training data.

As we have previously covered, the use of AI makes the application of the criteria for assessing the sufficiency of description more challenging due to the complexity of AI systems and the complex training data they use during the learning phase.

For example, the patentability of an AI-based solution may lie in different aspects: the use of a specific algorithm, a specific training method involving a specific definition of the cost function and/or a specific method of adjusting the algorithm parameters during training, the choice of the hyper-parameters of the algorithm, the nature and typology of the training data, the combination of the AI algorithm with other types of algorithms, preprocessing training data. Very often, it is a combination of these different aspects that enables the AI-based solutions (or "AI inventions") to be effective for a given application.

AI Inventions: EPO Guidelines for Examination

The European Patent Office (EPO) updated its Guidelines for Examination earlier this year to provide some practical guidance for drafting patent applications related to AI inventions, summarised below.

An important criterion is to determine to what extent the training data produces technical effects and contributes to the solution of a technical problem.

For example, in the case of a classification algorithm, the method of generating the training data and the training method of the classification algorithm can be decisive for this algorithm to be efficient, providing specific advantages relevant to the state of the art and thereby meeting the EPO patentability criteria.

The technical effects produced by the generation of the training data and/or the training method of the classification algorithm must thus be identified and established.

According to the EPO Guidelines, these technical effects can be established by different means, such as: explanations, mathematical proofs, experimental data or other factual elements. Mere claims of benefits, without any substantiation, are generally not enough for AI inventions. A causal link must be established between, on the one hand, the properties of the training data and/or the training method; and, on the other hand, one or more technical effects provided by the invention.

For example, if one of the technical effects depends on particular properties of the training data, these must be described, and it must be indicated how these properties achieve this effect. However, fully describing the training dataset is generally not required.

These general rules provide very useful practical guidelines for drafting, but it remains necessary to analyse each invention in detail, to identify, case by case, the characteristics conferring patentable character to the AI inventions and thereby the limits of what must be described or not.

The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.

Mondaq uses cookies on this website. By using our website you agree to our use of cookies as set out in our Privacy Policy.

Learn More