Patenting Emerging Computing Technologies before the EPO – 20 Decisions You Should Have Heard Of
M. M. Fischer (DE)
Emerging computing technologies, such as Artificial Intelligence (AI), Machine Learning (ML), Blockchain Technology, Big Data, Self-Driving Vehicles (SDV) have become a “hot topic” within - but of course also well beyond - the patent community. The EPO with its well-established and stable jurisdiction regarding computer-implemented inventions (CII) and its ambition to be the worldwide benchmark in patenting CII has placed special emphasis on patenting emerging computing technologies by, for example, organizing conferences on patenting AI and Blockchain Technology. It updated its Guidelines for Examination to provide more guidance on the patentability of CII in general and inventions in the field of AI, Machine Learning, Big Data, etc. in particular (see sections F-IV, 3.9.3; G-II, 3.3, 3.3.1, 3.3.2 etc.) and has published a study on Self-Driving Vehicles. The following article explicitly does not want to dwell on the basics of patenting CII, such as the COMVIK approach, but intends to shed light on a compilation of decisions that may be helpful in arguing why an invention is patentable or not before the EPO. Although some of the emerging computing technologies mentioned above are not entirely new, their large-scale industrial application and thus their relevance for the field of Intellectual Property has only recently become the focus of attention. While there are not so many decisions by the Boards of Appeal in these fields yet, the principles set out in the following decisions may be applied or extrapolated thereto and may thus assist in drafting, prosecuting and opposing future patent applications/patents before the EPO.
In the first section, I will discuss decisions that give guidance on how to deal with mathematical/algorithmic features in the claims. In the second section, I will present a selection of decisions relating to inventions from different technical fields some of which are patentable and some of which are not. The third section deals with user interaction and user interfaces, while the fourth section deals with decisions that rely upon the concept of a notional business person and make a rigorous distinction which features may be used in the statement of the problem. All 20 decisions are discussed within a context/network of related decisions.
a) Mathematical/Algorithmic Features in the Claims
1. T297/86 of 29.09.1989: Automatic Control of Printing Press
The patent underlying this decision deals with the use of linear regression analysis to correlate subjective and objective harmonic analysis data related to print quality and obtain regression parameters. These parameters, once learnt, can be used to predict the subjective data based on objective data obtained.
The Board held that the use of linear regression and harmonic analysis data was indeed inventive when applied to the specific problem. The inventive step appears to lie in the mathematical step: “It is the merit of the present invention that it has been recognized that the harmonic analysis approximates most closely the actions of the machine operator.”
An early important decision applicable to AI, Machine Learning and Big Data inventions where the inventive step relies on mathematical steps for a technical purpose.
2. T1227/05 of 13.12.2006: Simulation of a circuit subject to 1/f noise
Mathematical/algorithmic features can be regarded as technical features if they serve an adequately defined technical purpose. Another important aspect of the decision is that an enhanced speed of the algorithm, when compared to other algorithms, cannot be sufficient to establish a technical character of the algorithm. (“Algorithmic efficiency is not a technical effect.”, cf. T2418/12, T1784/06, T42/10 and T1370/11). If enhanced speed (or better use of resources) is obtained within a technical process, such as image compression, then this effect can be considered to be a technical effect.
It is recommendable to describe this adequately defined technical purpose associated with the mathematical/algorithmic features at least in the description so that it can be included in the claims, if necessary. In T953/94, the Board found allowable a method of controlling a "physical" process using a mathematical model, although a reference to an unspecified "physical process" might, according to T1227/05 be rejected as a "meta-specification".
Albeit not explicitly confirmed by the Boards of Appeal, it appears that simulation of a technical or physical process or product is considered to be technical, while simulation of non-technical subject-matter is not. For example, in T1265/09, which also involved simulation, the subject-matter of determining an efficient schedule for a plurality of scheduled agents in a telephone call center was not deemed technical. Incidentally, in T489/14 (“Modelling Pedestrian Crowd Movement”), the Board is currently (February 2019) considering to submit a set of questions relating to the field of simulation to the Enlarged Board of Appeal. The Applicant suggested the following questions:
- Can a computer-implemented method of simulation based on laws of physics or calculating values which represent physical quantities for aiding the design of technical aspects of a physical system or technical product or for aiding the technical operation of a physical system or technical product be considered to be or to serve a technical purpose provided the technical purpose is adequately defined?
- If the answer to question I is “No”, in a claim directed to a method of designing, making or operating the physical system or technical product and which recites steps in a method of simulation for aiding the design of technical aspects of the physical system or technical product or for aiding technical operation of the physical system or technical product, would the simulation method steps be considered as contributing to the technical character of the claim and, thus, be considered in an assessment of inventive step?
- III. Can a computer-implemented method of simulation involving values which represent physical quantities which can be influenced by or driven by non-physical factor(s) (such as aggregated human behavior) and yet still be accurately simulated and be technically relevant such that the simulation is still able to aid the design of technical aspects of the physical system or technical product or the technical operation of the physical system or technical product still be considered to be or to serve a technical purpose provided the technical purpose is adequately defined.
T1227/05 is certainly a landmark decision and is, for example, highly relevant for the field of “Big Data”, in which data from (known) sensors are analysed by novel and inventive algorithms and used to control a (known) technical device.
3. T625/11 of 19.01.2017: Determining a threshold of a parameter for operation of a nuclear reactor
The Board discussed whether a technical use of the output of a computer process can be a factor in determining technical character of the claim. In the present case, the Board stated that recognition of a technical character connected with the use of the threshold value resulting from the simulation for proper operation of a nuclear reactor would lead to a significantly more specific redefinition of the objective technical problem solved by the invention, namely determining at least one threshold value of a nuclear reactor operating parameter in order to allow better use of the reactor. Since the claimed solution was not disclosed in any of the cited prior art documents, the Board held that the claimed method did involve an inventive step.
The Board concluded that the determination of the value, being a threshold value of an operating parameter for a nuclear reactor, conferred a technical character on the claim and, further, that the technical character went beyond the mere interaction between the simulation algorithm and the computer system.
The claimed method itself does not bring about a technical effect. It was sufficient that only the use of a parameter determined by the method brings about a technical effect.
By contrast, in T471/05, the Board confirmed the refusal of a patent application which related to the mere design of an optical system, which can be carried out as a purely mental act or as a purely mathematical design algorithm and, consequently, encompasses embodiments falling within the category of methods for performing mental acts as such and within the category of mathematical methods as such both expressly excluded from patent protection under Art. 52(2)(a) and (c) in conjunction with Art. 52(3) EPC.
4. T914/02 of 12.07.2005: Designing a core loading arrangement for loading nuclear reactor fuel bundles into a reactor core
The whole method may be performed mentally, based on the appropriate, available data pertaining to the geometry of the core, the number of fuel bundles, the respective reactivities of the bundles, the reactor design rules etc. Moreover, as a result, the claimed method provides a design of a core loading arrangement which may be a purely mental, abstract scheme of how bundles could be arranged in an actual, real-world nuclear reactor core, rather than a concrete, physical reactor core loading.
The involvement of technical considerations is not sufficient for a method which may exclusively be carried out mentally to have technical character. One of the few cases where an application has been refused under Art. 52(2)(c) EPC because it did not exclude that it could be performed purely mentally. Inserting language such “using a computing device” in the claim could have overcome at least this objection.
b) Tour d`horizon through different technical fields
5. T598/07 of 19.05.2010: Heartbeat monitoring method based on a neural network
The patent application underlying this decision refers to the use of a neural network in a heart monitoring apparatus to identify irregular heartbeats, which was considered to be technical. The decision deals with a nice example of a technical application of a neural network, in particular a Kohonen neural network.
The decision is relevant for all sorts of technical applications of neural networks. In recent times, Deep Learning algorithms, e.g. convolutional neural networks (CNN), which are special types of neural networks, have gained a lot of attention since they were able to reduce the error rate in computer vision speech recognition applications. CNN are often efficiently trained using GPUs (graphics processing units) which by itself may be a basis for a technical effect.
Many AI applications involve a training phase and an operating phase. Although not explicitly stated in the updated version of the Guidelines for Examination, it is recommended to direct an independent claim to each phase.
6. T1784/06 of 21.09.2012: Classifying a set of data records
The application deals with the problem of classifying a set of data records. The Board held that the problem was non-technical since the automatic classification of data records serves only the purpose of classifying the data records, without implying any technical use of the classification.
In some cases, a method which relates to an algorithm which is per se considered non-technical may be considered technical if it refers to a special architecture, e.g. adapting a machine learning algorithm to a GPU (graphics processing unit), cf. EP Patent EP 1 569 128 B1.
7. T1358/09 of 21.11.2014: Document classification
In this decision, the Board held that the classification of documents based on their textual content is non-technical. As a consequence, in T22/12, it has been decided that classifying an e-mail as spam is also non-technical.
8. T2418/12 of 14.07.2017: Suggesting a related term
The algorithm underlying claim 1 serves the overall purpose of suggesting query terms that are semantically related to the various "senses" of a particular input term. This is not a technical problem, for whether terms are "related" to each other is a cognitive or linguistic matter and not a technical issue (cf. T1358/09, T2230/10 and T2439/11).
Subject-matter relating to linguistic and/or semantic aspects of texts is often considered to be non-technical.
9. T1316/09 of 18.12.2012: Analysing content of an incoming electronic message
Methods of text classification per se do not produce a relevant technical effect or provide a technical solution to any technical problem.
Critical voices may argue that semantic analysis of texts is conceptually similar to object recognition in images. (Following T208/84, object recognition in images as a sub-discipline of digital image processing is considered to be technicaly). Therefore, it is not understandable why the first is not considered to be technical in contrast to the latter.
10. T208/84 “VICOM” of 15.07.1986: Image Processing
The underlying patent application relates to a method of digitally processing images in the form of a two-dimensional data array.
The Board held that a basic difference between a mathematical method and a technical process can be seen, however, in the fact that a mathematical method or a mathematical algorithm is carried out on numbers (whatever these numbers may represent) and provides a result also in numerical form, the mathematical method or algorithm being only an abstract concept prescribing how to operate on the numbers. No direct technical result is produced by the method as such. In contrast thereto, if a mathematical method is used in a technical process, that process is carried out on a physical entity (which may be a material object but equally an image stored as an electric signal) by some technical means implementing the method and provides as its result a certain change in that entity. The technical means might include a computer comprising suitable hardware or an appropriately programmed general purpose computer.
In this landmark decision, an image (although its representation could be seen to be nothing more than a two-dimensional array of numbers) has been considered as a physical entity. This decision is applicable to all sorts of image, video and audio processing processes and in particular to the fields of image, video and audio compression and analysis (e.g. objection recognition, face recognition, speech recognition), cf. T1586/09 (Quality of transmitted digital audio signals).
The decision, albeit old, is still highly relevant and may, for example, be important for self-driving vehicles relying on analysis of camera images.
11. T1286/09 of 11.06.2015: Image classification of a digital image
The gist of the present invention consists essentially in increasing the diversity of exemplar images used to train a semantic classifier by systematically altering an exemplar colour image to generate an expanded set of images with the same salient characteristics as the initial exemplar image. More specifically, an exemplar image may be altered by means of "spatial recomposition", i.e. by cropping its edges or by horizontally mirroring it.
The present application relates generally to the field of digital image processing and, in particular, to a method for improving image classification by training a semantic classifier with a set of exemplar colour images, which represent "recomposed versions" of an exemplar image, in order to increase the diversity of training exemplars. All features are considered to be technical. Image classification can be assumed to be technical. Image classification is one big field of application of AI (e.g. Neural Networks, etc.)
T1148/05 also deals with image classification and thus similar technical subject-matter but was considered to lack an inventive step.
12. T1285/10 of 23.05.2014: Diagnosing and recommending treatment for a physiological condition using AI routines
The decision does not discuss whether claim 1 involves any non-technical steps. At least step (i) includes a physical/technical analysis step (“optical reader or scanner”).
The decision appears to suggest that there are no problems regarding technicality with patenting medical diagnosis systems based on AI. As a general rule, to render a claim technical, it is advisable to include “sensors” or any steps relating to a technical (physical/chemical/physiological, etc.) measurement.
13. T2050/07 of 19.03.2013: Providing a genotype estimate based on an analysis of DNA samples
The mathematical features have been considered to contribute to the technical character of the claims.
The question arose whether non-technical features can establish novelty vis-à-vis an earlier application according to Art. 54 (3) EPC. The question was left unanswered since the features in question were considered as contributing to the technical character of the claim.
The decision opened up the field of bioinformatics for patent protection before the EPO. Not many decisions by the Boards of Appeal in this field have followed so far.
14. T1326/06 of 30.11.2010: Encryption/Decryption/Authentication
Methods for encrypting/decrypting of electronic messages must be considered as technical even if they essentially rely on mathematical methods.
Cryptography (encrypting, decrypting or signing) is technical, cf. T27/97, T556/04.
Making the system more secure/more reliable is generally accepted as a technical effect. The decision may unfold relevance in the field of Blockchain technology.
15. T1842/10 of 30.04.2014: Computer-aided modelling method for the behavior of steel volume
The claim is not only directed to a modelling method but also comprises a control method for a means that influences a steel volume.
Control of an external device (X-ray apparatus, etc.) or the operation of the computer itself (memory management, load distribution T318/10, etc.) is normally considered to be technical.
c) User interaction/User interfaces
16. T1670/07 “Broken technical chain fallacy” of 11.07.2013: Facilitating shopping with a mobile wireless communications device
The decision says that a technical effect may arise from either the provision of data about a technical process, regardless of the presence of a user or its subsequent use, or from the provision of data (including data that on its own is excluded, e.g. produced by means of an algorithm) that is applied directly in a technical process. In the case at issue, the data was produced by means of an algorithm and was not applied directly in a technical process, so that neither possibility applied.
The Appellant argued that a reduction in use of resources would be achieved. The Board countered that this was only caused by the way the brain of the user perceives and processes the visual information given by a particular way of presenting information. This was considered as a “broken chain” of technical effects and not accepted. It is not advisable to argue that a technical effect is achieved if this effect is only achieved if the user reacts in a certain way to the information presented. The following two decisions show exceptions:
17. T528/07 of 27.04.2010: Providing a business-to-business relationship portal
If the cognitive content of the information presented to the user relates to an internal state prevailing in a technical system and enables the user to properly operate this technical system, it has a technical effect. An internal state prevailing in a technical system is an operating mode, a technical condition or an event which is related to the internal functioning of the system, may dynamically change and is automatically detected. Its presentation typically prompts the user to interact with the system, for example to avoid technical malfunctions.
This decision refers to an exception of how to escape from the “broken technical chain fallacy”. While psychological factors may not contribute to the technical character, cf. e.g. T862/10, physiological effects (such as physical properties of the human visual perception, T509/07) may be considered to make a technical contribution.
18. T2035/11 of 25.07.2014: Navigation System
The Board distinguished between
- a navigation system (broader interpretation) which calculates the route with the cheapest cost by using a Travelling-Salesman-Algorithm without any position-determining device (This is a purely mathematical pencil & paper problem) and
- a navigation system (narrower interpretation) with a position determining device and being configured to provide route-guidance information in dependence on the actual real-world position of the system.
"In the Board's view, providing real-time route-guidance information to a user in dependence on the user's real-world position is a technical task. It involves an interaction between the user and the navigation system, wherein the navigation system continuously measures the user's position using technical means and, on the basis of these measurements, provides the user with information aimed at enabling the user to manage the technical task of moving a vehicle to a desired destination.
Although the completion of this technical task depends on the user acting upon the provided route-guidance information and hence on an intervention by the user, it does not rely on subjective considerations by the user or on psychological effects. The user may still decide to ignore the route-guidance information, but that does not detract from the technical character of the navigation system as a technical tool to be used interactively in a technical process and not merely in a preparatory phase as a substitution of what could also be done using pencil and paper."
The decision seems to be another way of escaping from the broken technical chain fallacy, cf. T2172/03.
Even if the technical effect of reducing costs is only brought about by the user driving the car as recommended, the system as a whole is considered to be technical.
Cf. Guidelines G-II, 3.3: “Defining the nature of the data input to a mathematical method does not necessarily imply that the mathematical method contributes to the technical character of the invention (T2035/11, T1029/06, T1161/04). Whether a technical purpose is served by the mathematical method is primarily determined by the direct technical relevance of the results it provides.”
This decision may unfold relevance in the technical field of self-driving vehicles.
d) Notional Business Person and Requirement Specification
19. T1463/11 of 29.11.2016: “Cardinal Commerce” - Notional Business Person
The patent application deals with authentication of a consumer via a computer at a centralized merchant authentication processing system.
The Board held that the notional business person (more generally, the non-technical person) cannot normally require even notorious technical means. The reason was that the inventor may have obtained a technical effect using technical means, even notorious means, in a way that would not have been obvious to the skilled person. That was what a patent was meant to reward. To allow the notional business person to prescribe technical means would be to foreclose any discussion of whether they were used in a technically non-obvious way. This should not lead to a proliferation of patents for technically trivial inventions; they would be obvious to the skilled person.
Following the COMVIK approach, this decision introduces the “notional business man” in addition to the “person skilled in the art”. The notional business man gives a requirements specification to the person skilled in the art which has to be free from all (even notorious) technical requirements.
The decision was confirmed in T630/11 and further developed in T144/11 in which the Board stated as a headnote that a problem of the type "implement [the business requirement]" will normally never lead to an allowable claim. Either the implementation will be obvious or have no technical effect, or if not, the implementation will have a technical effect that can be used to reformulate the problem essentially to "achieve [the effect of the implementation]". However, the implementation-type problem is just a starting point that might have to be modified when the implementation is considered. It helps when a technical problem is not apparent at the outset. Examining the business requirements carefully and correctly establishing what is to be implemented ensures that all technical matter arising from the idea of the invention and its implementation is taken into account for inventive step.
This decision may be important for the technical field of Blockchain in which technical problems occur in a business/administrative context.
20. T2079/10 of 19.04.2018: Control of cellular, geographically distributed alarm systems
The patent application underlying this decision relates to electronic control of cellular, geographically distributed alarm systems.
The Board considered claim 1 to include technical features, in particular the feature that a control unit is triggered by physical measurement parameters. Hence, claim 1 is not purely a business method.
The Board expounds that the feature of physical parameters cannot be taken as part of the problem statement under the COMVIK approach. Moreover, it cannot be assumed that the technically skilled person chooses a physical parameter, on the basis of which the technically skilled person gets from the business person a purely abstract business model as a concept for implementation.
The decision may be considered to be a consequence of T1463/11.
This decision could be important for AI inventions. In particular for e.g. neural networks suitable for processing real world data such as video.
Outlook – Enforcement of Patents
The initiatives by the EPO in the fields of emerging computing technologies and its stable jurisdiction contribute to an ideal framework for obtaining patents in these fields. However, let us have look at the situation regarding the enforcement of patents in the field of emerging computing technologies. What is the nicest (or better: broadest) patent worth if it cannot be reasonably enforced - either because the courts do not have the technical expertise or even more importantly the infringement of such patents is intrinsically difficult to detect? The first problem could be solved by the Unified Patent Court which offers the possibility to recruit technical judges who have technical expertise in the emerging computing technologies mentioned above. The second problem is more difficult to tackle. Rachel Free and Loretta Pugh have shown in their article “Implications of the General Data Protection Regulation for Detecting Infringement of AI Patents” in EPI Information 3/2018 that the GDPR may be helpful in detecting patent infringement.
Another approach that I would like to add is the standardization process that has just started in the field of AI. “ISO/IEC JTC 1/SC 42” is a committee that was newly established that deals with the standardization of AI. Moreover, if one thinks about safety relevant fields, such as algorithms that are applied in self-driving vehicles, it is clear that such algorithms will have to be examined carefully by a technical authority (e.g. TÜV in Germany, MOT in the UK) and that they will underlie some standardization. It is also thinkable that manufacturers of self-driving vehicles will be forced to openly disclose the algorithms that they employ. The highest number of patents within CII that are enforced are standard essential patents (SEP) in the field of telecommunications because one can easily determine whether someone infringes a patent. If AI is also governed by standards, then enforcing patents in this field will also be possible without undue burden such as reverse engineering.
If you have any questions or comments, please contact the author: firstname.lastname@example.org
The author runs a newsletter on computer-implemented inventions. Please write to the same e-mail address if you wish to subscribe.
- The term “Artificial Intelligence” was coined in 1956 by John McCarthy at the Dartmouth Conference which is considered by many as the birth of AI as a research field. The field did not grow organically and had to cope with setbacks. Since AI could not hold the enthusiastic promises it initially made, the field entered into a depression at least twice (called “AI winters”) – periods during which AI research did not get a lot of funding and research facilities were closed down. Possibly, the current boom period should therefore also be regarded in a realistic manner against this backdrop.
- Moufang, R., «Zur Patentierung von Entwurfs- und Simulationsverfahren in der EPA-Rechtsprechung», GRUR Int. 2018, Heft 12, 1146-1152
- Do not confuse this decision with T625/11 since both deal with nuclear reactor technology.
- AI techniques, such as neural networks, deep learning, clustering, support vector machines, etc., are different from other, let us call them “conventional”, algorithms in that they typically involve a training phase in which they are presented some samples (e.g. images) by means of which the AI algorithm learns to classify the image. In an operating phase, once the algorithm has been trained, it can be shown new samples that it has not seen before and it can classify them. However, the classification it makes depends on the training samples and unlike conventional algorithms it is not well understood why the AI technique comes to a certain decision (“black box”) and cannot be mathematically proven that it works correctly in all cases which may lead to safety problems e.g. in the case of self-driving vehicles. There is a similarity with pharmaceutical inventions in which studies/tests can show that a drug is able to cure a disease but it is not fully understood why this is the case, whether the drug helps persons suffering also from other diseases and whether there are any side-effects. For obtaining a patent for AI inventions, this means that, like for pharmaceutical inventions, it may be advisable to file test results that show that an AI technique indeed achieves a technical effect.
- Analogy: In the digital world, image processing allows for example to sharpen the contrast of an image. In the real-world, an image restorer who restores old paintings can achieve the same effect using mechanical and/or chemical methods. Therefore, both processes, the digital image processing and the methods applied by the image restorer, should be considered to be technical.
- Is the situation different, in other words is the system still technical, if there is no physical/chemical analysis step but only a user answering questions via e.g. a computer-based questionary based on which a medical diagnosis is made? In T1153/02, it was not discussed whether some features were non-technical. The application was refused based on a lack of inventive step in view of the prior art.
- Like the European Medicines Agency (EMA) examines the admission of medicinal products based on tests, clinical studies, etc.