As artificial intelligence (AI) continues to advance and transform various industries, the protection of AI-related inventions through patents becomes increasingly important. Artificial intelligence encompasses a wide range of technologies, including supervised and unsupervised classifiers, natural language processing, computer vision and neural networks.
Each patent office has their own guidelines for the assessment of patentability of patent applications and the approaches taken by the different patent offices worldwide can vary considerably. This is especially true for computer-implemented inventions, which of course includes many inventions relating to AI. Europe is an important market for many industries and a convenient way to obtain patent protection in multiple European countries is by filing an application at the European Patent Office (EPO).
A summary of the main issues that should be considered when trying to obtain protection from AI-based inventions is provided here. Of course, this is not an exhaustive list and there can be other concerns (for example, clarity of the claims) but this article provides a general overview.
Excluded subject matter
To be eligible for patent protection at the EPO, an invention must have a “technical character”. AI inventions often involve complex algorithms, data processing techniques, and predictive models. In isolation, the EPO considers that these underlying algorithms and computational models are abstract mathematical methods, which are not eligible for patent protection, regardless of whether the computational models could be used to solve technical problems or trained with technical training data. In order for the EPO to consider an invention “technical”, rather than “abstract”, the invention must possess technical character as a whole. Otherwise, the invention would be excluded from patentability.
Practically, most AI inventions are computer-implemented inventions. The EPO considers a computer a technical means for solving a problem. To demonstrate that an AI innovation goes beyond mere mathematical models or algorithms and achieves a practical technical effect, therefore, it is usually sufficient to say that the invention is implemented by a computer processor (or other similar technical means).
Once it is established that the claimed subject-matter as a whole is not excluded from patentability, the invention is examined in respect of the other requirements of patentability. These can include inventive step and the “technical character” of the invention, which is again considered in this assessment, as will be discussed below.
Incentive Steps
Inventive step is a crucial requirement for patentability. For an invention to be considered inventive, the EPO requires it to solve a technical problem in a non-obvious manner, in view of all the prior art available at the filing date of the application. Implementing a known AI technique with an existing data set to solve a known problem is unlikely to be considered inventive on its own, even if the problem has not been addressed using AI in the past. AI-related patent applications often face challenges in demonstrating inventive step due to the availability of vast amounts of prior art and the overlap of AI techniques across different technical fields. To assess inventive step, it is important to describe the new technical features and resulting technical effects of the invention clearly, especially when building on existing AI methodologies.
The EPO applies the same assessment criteria for inventive step to AI-related patent applications as it does for other computer-implemented inventions. However, some specific considerations within the field of AI can be distinct.
In accordance with the EPO’s approach to assessing inventive step, the invention must solve a technical problem in a non-obvious manner. A first step to assessing inventive step is to identify an objective technical problem solved by the invention. Artificial intelligence may be applied in various fields of technology and that field is often key to the choice of problem. Some applications are intrinsically technical, whilst others are not. Classifying text documents solely in respect of their textual content is not regarded as a technical problem by the EPO (rather, it is regarded as a linguistic problem). Classifying data records, such as telecommunication network data records, without any indication of a technical use being made of the resulting classification, is also not deemed technical by the EPO, even if the classification algorithm is more robust than previous methods. In contrast, the use of a neural network in a heart monitoring apparatus for the purpose of identifying irregular heartbeats would be considered to solve a technical problem. Moreover, classification of digital images, videos, audio or speech signals based on low-level features (e.g. edges or pixel attributes for images) is generally also considered technical.
The EPO may interpret any features that only address non-technical aims as constraints in the formulation of the technical problem (i.e., these features are effectively ignored when deciding if the invention is non-obvious). Therefore, if the only problem solved by an AI method is not intrinsically technical, the EPO will consider that the technical problem solved by the invention is simply implementation of a non-technical method on a computer system, which they will view as obvious.
Once the technical problem has been identified, the EPO will identify the features of the invention that contribute to the technical character of the invention. Computer-implemented inventions involving AI will claim a mix of technical and non-technical features, with the algorithmic parts of the AI initially being considered non-technical. However, non-technical features – including any AI algorithm – may be taken into account when assessing inventive step, but only if they contribute to solving the technical problem. Non-obviousness of algorithmic features of the AI will be considered if those features provide a “technical contribution”.
A technical contribution can manifest in various ways, such as improved efficiency, increased accuracy, enhanced user experience, or the solving of another specific technical problem. The EPO provide some examples of how mathematical methods may make technical contributions, which include:
“Digital audio, image or video enhancement or analysis, e.g. de-noising, detecting persons in a digital image, estimating the quality of a transmitted digital audio signal;”
“Separation of sources in speech signals; speech recognition, e.g. mapping a speech input to a text output;”
“Optimising load distribution in a computer network;” and
“Providing a medical diagnosis by an automated system processing physiological measurements.”
For the AI method to provide a contribution, the claim must also be functionally limited to the technical purpose by establishing a sufficient causal link between the technical effect produced by the claim and the AI method steps defined therein.
Alternatively, AI methods may contribute to the technical character of the invention – independently of any technical application- if the AI methods are specifically designed/adapted for implementation on a specific computing system or network (and where the design is motivated by technical considerations). To satisfy this requirement, the AI method must be designed to exploit specific technical properties of the system, in order to bring about a technical effect, such as efficient use of computer storage capacity or network bandwidth.
One useful example provided by the EPO of an AI method being adapted for particular system is:
“Assigning the execution of data-intensive training steps of a machine-learning 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. The claim should be directed to the implementation of the steps on the GPU and CPU for this mathematical method to contribute to the technical character.”
Where an AI method makes a technical contribution, the steps of generating training data and training the AI itself may also contribute to the technical character of the invention. If the novel/inventive character of the invention resides in the method of training the AI or the generation of training data, these steps may be considered during assessment of inventive step. If the AI method does not make a technical contribution, simply applying technical training data to an abstract AI method will not make the method technical. Such methods would therefore be excluded from consideration for assessment of inventive step.
Once the features that make a technical contribution have been identified, the final step is to determine whether the features contributing to the solution. This starts from the closest prior art and the objective technical problem – and whether this would have been obvious to the skilled person.
Sufficiency
In order for an invention to be patentable, it must be described in sufficient detail for a person skilled in the art to put the invention into effect. Where an AI invention includes a known neural network model that is trained to produce a desired effect, the requirements for sufficiency at the EPO can be quite high. One reason for this may be that the neural network is treated as a “black box”, whose inner workings are not disclosed. The only way to enable someone to reproduce the effect of such a neural network is to describe the process of training the neural network, to produce the desired outcome. A patent application involving a known neural network should be carefully drafted so that sufficient details regarding the training data are included. For example, input data suitable for training the neural network should ideally be described in detail (or at least one data set suitable for solving the technical problem of the invention should be included). Otherwise, the EPO may argue that the skilled person would be unable to reproduce the invention.
In some cases, it may be impractical or undesirable to disclose entire sets of training data. Instead, it may be preferrable to describe the training data that is needed in sufficient detail (i.e., how to select and process suitable data), or else modify the example data to produce a minimum set that satisfies the sufficiency requirements.
Conclusions
The EPO applies its established criteria for assessing patentability when evaluating AI-related patent applications. While recognising the unique challenges posed by AI technologies, the EPO focuses on the technical character, technical effects, and non-obviousness of the AI invention. By ensuring a robust assessment process, the EPO promotes the protection of truly inventive and technologically advanced AI innovations, fostering continued progress in the field of artificial intelligence.
When drafting a patent application relating to an AI-based invention, it is crucial to clearly describe the technical advancements, innovative algorithms, technical effects, and specific technical challenges overcome by the invention. It is also helpful to provide experimental data, comparative results, and evidence of unexpected or advantageous effects, which can further support the presence of an inventive step.
Please contact your usual advisor at Boult for more information.
