In this first of two articles, we summarise our recommendations for drafting patent applications for inventions concerning software and algorithms. As these are subject matter areas that have gained a reputation for being difficult to successfully prosecute before the European Patent Office (EPO), we aim to provide some helpful tips to increase the chances of these applications granting before the EPO.
What kinds of software can be patented at the EPO?
EPC Article 52, relating to what constitutes a patentable invention, states that mathematical methods and programs for computers are not inventions to the extent that the claims relate to this subject matter “as such.” Much case law has been and continues to be the focus of what those two words, “as such”, mean but it is established in our case law that if the invention is “technical”, then it does not fall within the exclusion. Therefore, the conclusion is that only “technical” subject matter is patentable. However, this effectively moves the discussion from how the words “as such” are to be understood to what does/does not constitute a “technical” invention.
At the EPO, a computer-implemented invention (“CII” invention) is one which involves the use of a computer, computer network or other programmable apparatus, where one or more features are realised wholly or partly by means of a computer program.
Artificial Intelligence (“AI”)/Machine Learning is always computer implemented, and the EPO considers AI to be a mathematical method.
The most recent March 2021 edition of the Guidelines for Examination in the EPO specifically addressed these kinds of inventions and the EPO’s practice for AI and machine-learning:
“Artificial intelligence and machine learning are based on computational models and algorithms for classification, clustering, regression and dimensionality reduction, such as neural networks, genetic algorithms, support vector machines, k-means, kernel regression and discriminant analysis. Such computational models and algorithms are per se of an abstract mathematical nature, irrespective of whether they can be “trained” based on training data. Hence, the guidance provided in G‑II, 3.3 generally applies also to such computational models and algorithms” (Guidelines for Examination in the EPO, Chapter G-II, 3.3.1).
Although CII subject matter can be patented, the case law about what exactly is technical (and therefore patentable) is complex. However, the EPO has clarified that an invention can be technical in two different ways (dimensions):
Dimension 1: a feature is particularly adapted for a specific technical implementation (in the sense that the design of the feature is motivated by technical considerations of the internal functioning of a computer). For example, a claim may improve the inner workings of the AI algorithms themselves, e.g. in hardware of software hierarchy. For this dimension, an AI algorithm needs to be adapted for a specific implementation in a computer and therefore needs to be motivated by considerations of the internal functioning of a computer.
Dimension 2: a feature has a technical application to a field that the EPO deems technical (the feature thereby gains “technical” status by virtue of its application in a technical field – e.g. applied AI to finding a new chemical composition or to image processing). For this dimension, an AI algorithm needs to be adapted for a particular technical application, for example to solve a particular technical problem in a technical field.
Thus, either the invention must fundamentally make a computer better, or it is applied to a specific field of technology considered technical. In the latter case, there is a large body of complex case law as to what is, and what is not, a technical area.
How to push a draft AI application in the right direction for grant
Only features which are both inventive and viewed by the EPO as having a technical contribution can confer an inventive step before the EPO. This makes drafting an application which is suitable for the EPO in this area a question of distinguishing over the prior art in a way which is deemed technical.So we need the right kind of technical detail in the application at the start to help us. It is of course necessary to include a description of all the standard software and hardware that is used in the invention. However, emphasis should be placed on a full disclosure of the parts of the software and/or hardware that distinguish over the prior art. For AI in particular, if the invention is in the training/learning phase for example, you need to provide as much detail as you can “squeeze out” of the inventors to explain everything that is new in that phase. Also, for inventive step at the EPO, it is very useful to add advantages of the invention details into the specification at the drafting stage, because inventive step arguments can be based on these advantages.
Although a mathematical method must serve a technical purpose to be deemed technical, a generic purpose (such as “controlling a technical system”) is not sufficient – the purpose must be specific. Furthermore, the claim should not define a technical purpose so broadly as to be seen by the EPO as “administrative.”
Some examples of what the EPO considers to be technical purposes are:
- digital audio, image or video enhancement or analysis
- separation of sources in speech signals, speech recognition
- encoding data for reliable and/or efficient transmission or storage (and corresponding decoding)
- compilation of software
- encrypting/decrypting or signing electronic communications
- facilitating data access using data structures (although we consider this a borderline case)
In contrast, areas generally viewed by the EPO as not technical are:
- inventions related to cognitive data (of interest to humans) rather than functional data (related to a machine)
- output not fed to a machine/technical process
- anything that seems linguistics, or business or admin related
- very broad technical scope or lacking specific technical implementation (provide narrower fallback positions)
A graphical user interface (GUI) comprises feature of presenting information and receiving input in response as part of a human-computer interaction. Features defining user input are more likely to be deemed to have a technical character than those solely concerning data output and display. On the other hand, features concerning the graphic design of a menu (such as its look and feel) which are determined by aesthetic considerations, subjective user preferences or administrative rules do not contribute to technical character. However, a claim feature concerning a GUI that “assist(s) the user in performing a technical task by means of a continued and/or guided human-machine interaction process” has technical character (Guidelines for Examination, Chapter G-II, 3.7).
Example CII claim forms that are accepted by the EPO are:
- A computer-implemented method comprising steps A, B, …
- e.g. A computer-implemented image processing method of training (or evaluating, or operating) a neural network, comprising steps…
- A data processing apparatus/device/system comprising or a processor configured to perform the method of claim 1
- A computer program comprising instructions which, when executed by a computer, carry out the method of claim 1
- A computer-readable medium comprising instructions which, when executed by a computer, cause the computer to carry out the method of claim 1
See EPO Guidelines for Examination F-IV, 3.9.1: Claims directed to computer-implemented inventions
The EPOs approach to “mixed-type” inventions
Many AI applications will contain both technical and non-technical features since they relate to mathematical algorithms (non-technical) which are computer-implemented (technical). Here, we look at how the EPO applies the “problem and solution” approach to these types of claims. This is referred to as the “COMVIK” approach (as introduced in T 641/00) and is illustrated in the below flowchart.According to this approach, the EPO will first assess the claim for technical character. However, this hurdle is easily overcome by specifying that the method is “computer-implemented.” Next, the EPO will consider whether the claim has any technical features, dividing the claim into those features deemed technical and those that are not. A feature can be technical according to one of the two dimensions defined above. Therefore, the term “technical features” also includes features which, when taken in isolation, are non-technical but do – in the context of the claimed invention – help to produce a technical effect (e.g., features of a machine learning algorithm used to identify an image).
If the claim includes non-technical features, these features may later be used in the formulation of the objective technical problem as part of what is “given” to the skilled person, in particular as a constraint that has to be met. For this reason, it is best to beware of including too many non-technical features into the claims.
Then, as per the “standard” problem-solution approach, the Examiner will identify the differences between the claim and the closest prior art, and the technical effects of these differences in the context of the claim as a whole. If all of the differences make no technical contribution when the EPO can raise an inventive step objection. For example, a computer-implemented machine learning method claim for training a neural network may have differences over the prior art in the training parameters, but if they are not linked to a technical application, they will make no technical contribution.
Using the differences that do make a technical contribution, the EPO will formulate an objective technical problem. The non-technical features of the claim that serve no technical purpose may be used to formulate this problem. The Examiner will then consider whether or not the claimed invention, starting from the closest prior art and the objective technical problem, would have been obvious to the skilled person. For example, if a training algorithm for a machine learning method is not linked in the definition to a technical application and the whole claimed method is therefore considered to be non-technical, the objective technical problem might simply be to implement such a method. At that point any general definitions of computer implementation in the claim will not be considered inventive, and the Examiner may well cite “notorious“ computer systems against these features. If the Examiner considers that it would not have been obvious, no inventive step objection is raised.
Summary and practice points
Despite some notoriety in terms of the perceived difficulty of successfully prosecuting these types of applications before the EPO, various types of AI can be patented. In order to clear the first hurdle to patentability, a claim must possess technical character – this is easily achieved e.g. by claiming a computer-implemented method or process.
The second hurdle then relates to assessing the inventive step of the claimed subject-matter according to the COMVIK approach, wherein the distinguishing features are assessed in terms of whether they solve a technical problem. In order to maximise the chances of success for clearing this hurdle, the application and the claims should ideally be drafted with this approach in mind, and should ideally include as much detail as possible regarding any technical effects achieved by specific claim features. Even features in the claims or description which may not seem to be very important during drafting, since they seem to be merely additions to the central aspects of the invention, may later turn out to be crucial for assessing inventive step in view of the particular prior art documents cited by the Examiner. If the originally filed application includes some technical effects even for these “fall-back” features, it will be easier to come up with a convincing COMVIK-style argument on why these additional features provide a technical contribution over the prior art.
As explained earlier, a technical contribution can occur in either one of the two dimensions: specific technical implementation or technical application.
Further, it is advantageous to present the application in a technical context and to put the emphasis less on any business or user interaction aspects of the invention, but to focus on description of technical features which differ from the prior art and technical hurdles that are overcome by the invention. Ideally, the description would also include some examples of real-life applications of the inventive method or AI system, and include system or device claims directed at obviously technical apparatuses that may also be beneficial for framing the whole invention in a context that is more likely to be accepted as “technical” by the EPO Examiner.
In our next article we will examine sufficiency as applied to AI inventions, including the various types of data that may need to be included in the application to fulfil this requirement. We will also look at what can be learned from the decision G 1/19 (concerning computer simulations) and what can be learned from later decisions that have cited G 1/19.
This is for general information only and does not constitute legal advice. Should you require advice on this or any other topic then please contact firstname.lastname@example.org or your usual Haseltine Lake Kempner advisor.