Background: Patent Eligibility at the EPO
Very briefly, the European Patent Convention (EPC), and specifically Article 52 EPC, dictate that, (amongst other exemptions) computer programs and mathematical methods shall not be regarded as inventions. This may be perceived to mean that this type of subject matter is not patentable in Europe. However, Article 52(3) EPC provides a caveat to these exceptions in that they are excluded from patentability only to the extent to which a European patent application or European patent relates to this subject-matter ‘as such’. Without getting into too much detail, this caveat means that, in reality, inventions can be patentable in Europe even if they include or overlap with the exclusions.
In particular, an invention involving a mathematical method or a computer program can nevertheless be patentable if it is of technical character and produces a technical effect beyond the mere execution of the mathematical method or program. In Europe, this ties into the assessment of novelty and inventive step. In general, only the technical features of the invention that contribute to the technical effect are considered when determining whether the invention is novel and inventive over the state of the art. This usually means that patent eligibility itself is actually dealt with during the assessment of inventive step, through the problem-solution approach.
AI-related inventions are often closely related to computer programs or mathematical methods, and are thus considered in a similar manner in terms of patentability, novelty and inventive step. However, AI-related inventions have further, unique requirements to consider when drafting, some of which appear in the EPO guidelines for examination, and some which have been developed from case law.
Key Considerations for Drafting AI Patent Applications at the EPO
Technical Character and the Problem-Solution Approach
As discussed above, AI inventions are often classified as mathematical models. The EPO recently updated the Guidelines for Examination (G-II, 3.3.1) to specifically set out their approach to assessing inventions relating to AI and machine learning (ML), emphasizing that they must exhibit a technical character and solve a technical problem. Indeed, the EPO even note that the pre-existing guidance on mathematical methods in the Guidelines (G-II, 3.3) also generally applies to the computational models and algorithms used to implement AI and ML.
So what does this mean in practice? Usefully, the EPO has also provided a number of examples in the Guidelines (G-II, 3.3.1) indicating what may be considered technical and non-technical.
Firstly, the EPO has indicated that simply defining inventions in terms such as "support vector machine", "reasoning engine" or "neural network" may merely refer to abstract models or algorithms. Therefore, using these types of terms on their own does not necessarily imply the use of technical features as required for providing a technical effect.
Rather, the use of AI in a technical feature, such as the use of a neural network in a heart monitoring apparatus for the purpose of identifying irregular heartbeats, is more likely to be perceived to make a technical contribution. Similarly, the classification of digital images, videos, audio or speech signals based on low-level features such as edges or pixel values are further typical technical applications of classification algorithms.
On the other hand, the EPO has indicated that classifying text documents solely in respect of their textual content is for a linguistic purpose rather than a technical one (see decision T 1358/09). The EPO has also indicated in the Guidelines (G-II, 3.3.1) that classifying abstract data records or even "telecommunication network data records" without any indication of a technical use being made of the resulting classification is non-technical, even if the classification algorithm may be considered to have valuable mathematical properties.
More generally, the EPO appears to assess the technical character of AI inventions in the same way they have for mathematical methods and computer programs. In this manner, it may be useful to look to other areas of the Guidelines, including those for computer program type inventions. In the Guidelines for Examination, G-II, 3.6, it is stated that a computer program product requires a (further) technical effect which goes ‘beyond the "normal" physical interactions between the program (software) and the computer (hardware) on which it is run’. The normal physical effects of the execution of a program, e.g. the circulation of electrical currents in the computer, are not in themselves sufficient to confer technical character to a computer program. Generally, as indicated by case law, this means that computer programs require some sort of technical effect imparted on the real world (e.g. an algorithm that makes the spin-cycle of a washing machine use water more efficiently, or a process that fundamentally improves the performance of the computer it runs on). Since many AI inventions are computer-implemented, it would be useful to consider what sort of real-world applications the invention has when drafting a patent application for filing at the EPO.
To summarise, AI-related inventions face a similar level of scrutiny at the EPO as other computer-program products and mathematical methods, and the EPO has adopted similar approaches when determining the presence of a technical contribution, or lack thereof, when assessing AI subject matter.
Adequate Disclosure to Avoid Insufficiency Objections
A growing challenge in drafting in Europe with respect to AI-related inventions is ensuring that the requirement of sufficient disclosure is met. The EPO has been raising more insufficiency objections for AI inventions, particularly when applicants fail to explain the technical implementation of the AI model in adequate detail. The Guidelines for Examination (G-II, 3.3.1) touch on sufficiency with respect to AI-related inventions that require training data, stating that ‘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’. Even before these additions to the Guidelines, decisions such as T0161/18 helped demonstrate the EPO's approach to sufficiency for AI-related disclosures. In this case, the EPO upheld an insufficiency objection because the applicant had not adequately described how the AI system was trained or how the claimed invention achieved the alleged technical improvements. This decision and the updated Guidelines stress the importance of providing detailed technical information relating to an AI invention, such as the architecture of neural networks, data input/output parameters, and training methodologies, especially where such features are crucial to the technical effect and intended use of the invention.
Common Pitfalls to Avoid in AI Patent Drafting
Given the complexity of AI technologies and the EPO’s approach to the assessment of patentability, several pitfalls can arise when drafting AI patent applications. Below are some common pitfalls to consider.
Overly Broad or Vague Claims
One potential mistake in drafting is to draft too broadly. Overly broad claims and specifications that attempt to cover generic AI methods without detailing how the invention addresses a specific technical problem may risk falling foul of the requirements of clarity and sufficiency of disclosure.
Lack of Detailed Technical Disclosure
Another related potential pitfall is failing to provide sufficient detail about the AI model’s technical architecture, inputs, outputs, and training data. As highlighted by recent objections and the EPO’s guidelines, applicants may be required to disclose such features in greater detail than before. Simply stating that an AI method improves predictions, optimizes performance, or processes data is not likely to be adequate - rather, the low-level features of the invention required to provide these results may need to be disclosed to meet the requirement of sufficiency of disclosure.
On this point, the Guidelines for Examination (G-II, 3.3.1) further state that ‘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’.
Lack of Explanation of Technical Effect
Another potential pitfall may include failing to explicitly link the AI aspect of the invention, such as a machine-learning model, to a technical effect. As noted above, the EPO appear to consider the patentability of AI-related inventions in a similar manner to mathematical methods and computer programs, and as such, the focus when drafting an application to an AI-related invention should be similar as it is for these subject matter types. In particular, it may help to frame the AI-related invention with respect to a technical application linked to a technical effect that has real-world advantages.
Best Practices for Structuring AI Patent Applications
To improve patent drafting for EPO applications directed to AI inventions,applicants may wish to follow a structured approach that aligns with the EPO’s requirements.
1. Clearly Define the Technical Problem and Solution
A well-drafted patent application may begin by clearly identifying the technical problem the invention addresses. Linking the AI aspects of the invention to this technical problem, the technical solution to the problem, and any technical advantages associated therewith may aid in more clearly illustrating a technical contribution.
For example, if an AI model is used to optimize resource allocation in a telecommunications network, a patent application to such an AI model may describe the problem of network latency, and the solution of how the optimization using the model reduces network latency or improves throughput, providing a potentially technical effect.
2. Provide Detailed Descriptions of the AI Model
When drafting an application concerning an AI invention, a detailed discussion of the implementation of the AI, such as the model’s architecture, training process, and operation is likely to be beneficial in terms of satisfying the sufficiency of disclosure requirement. As noted above, the Guidelines for Examination (G-II, 3.3) indicate that the EPO have considered specifically that disclosure of how the AI model functions, how it is trained, and what data it uses may be needed.
Applicants may wish to include technical descriptions of the AI model's components, such as neural networks, machine learning algorithms, and data inputs/outputs. Examples of use cases or experimental results can also help for sufficiency purposes and may additionally help clarify the technical contribution of the AI model.
3. Emphasize the Technical Features responsible for the Technical Effect in the Claims
To meet the requirements of novelty and inventive step, applicants may wish to ensure that each claim emphasizes the technical effect of the invention. As discussed above, the assessment of patentability in terms of eligible subject matter usually occurs in earnest during the assessment of inventive step. For this reason, it may be beneficial to draft claims such that they clearly define the technical features responsible for the technical effect. These should be more than standard features such as a computer or a processor.
For instance, if an AI system improves the energy efficiency of an industrial process, the claims should clearly state how the AI model achieves this improvement, (through features that interact with the industrial process) such as through better control of system parameters or actuators, or more efficient allocation of resources, for example.
4. Include Real-World Examples
Real-world examples and applications of the invention may help to demonstrate how an AI invention is used in practice, and may thus help to tie the invention to a real-world technical effect. Such examples may beneficially show how the AI invention solves a specific technical problem, and may further indicate useful technical advantages.
Conclusion
The practice of drafting patent applications to AI-inventions may still be in its infancy, but already the EPO has taken positive steps to set out a framework of guidelines and case law for patent practitioners to consider for best practices. At present, it would appear to be beneficial for applicants to demonstrate a clear technical contribution, provide sufficient disclosure, and ensure that the AI invention solves a technical problem. It will be interesting to see how the EPO’s framework continues to develop in the future.
Here at Solve Intelligence, we are committed to building AI-powered platforms that can help practitioners draft in every technical field, including those relating to AI-inventions. Our platforms assist with every aspect of the patenting process while keeping patent professionals at the helm of these powerful tools. In this way, we give patent practitioners the control needed to reap AI's benefits while mitigating its associated challenges. Our Patent Copilot™ helps with patent drafting, patent filing, patent prosecution, office action analysis, patent portfolio strategy and management, and future patent infringement analyses. At each stage, our Patent Copilot™ works with the patent professional, keeping them in the driving seat, thereby equipping legal professionals, law firms, companies, and inventors with the tools to help develop the full scope of protection for their inventions.