EPO Publishes Revised Guidelines on Computer-implemented inventions

EPO Publishes Revised Guidelines on Computer-implemented inventions

EPO headquarters, the Isar building in Munich.

The European Patent Office Guidelines 2018 were recently published on the European Patent Office (EPO) website. All substantial changes in the new Guidelines relate mainly to sections discussing the First Hurdle, the EPO equivalent to patent eligibility. Although the First Hurdle can be overcome simply by adding the presence of a computer, the number, quality of and relationship between technical features are essential in dealing with the Second Hurdle, or inventive step. A thorough analysis of whether each claimed feature is technical, or not, is essential to claim drafting and prosecution of a computer-implemented invention at the European Patent Office and many also believe may help assessing eligibility and patentability before the United States Patent and Trademark Office. Thus, U.S. patent practitioners working with Computer-implemented inventions (CII) would do well to review the new EPO 2018 Guidelines.

The new Guidelines entered into force on November 1, 2018. Like the previous edition, this year’s Guidelines include substantial and valuable improvements with regard to guidance on the eligibility of computer-implemented inventions (CII). These sections of the Guidelines have been discussed with the European Patent Institute (epi), in particular with the ICT Thematic Group of the European Patent Practice Committee within the epi. The EPO website includes a useful html index for the Guidelines, including those sections relating to computer-implemented inventions.

The changes within the 2018 Guidelines appear mainly in Parts F-IV and G-Vll of the Guidelines and provide further, detailed guidance on well-established practices.  Such practice discussed in the guidelines consists of the so-called “two-hurdles” approach:

  1. First Hurdle. The claimed subject-matter must have a technical character, but claims may contain a mix of technical and non-technical features; and
  2. Second Hurdle. All features contributing to the technical character are taken into account for assessment of inventive step, including mathematical method steps which contribute to the technical character of the invention. The inventive step is assessed by the well-established “Problem and Solution Approach for claims comprising technical and non-technical features” (see G-VII, 5.4), which has not undergone substantial changes.

The Guidelines also explain that claims in a patent application concerning a computer-implemented invention may comprise:

  1. A method claim;
  2. A computer program product claim;
  3. A corresponding data carrier signal carrying the computer program product;
  4. A computer readable (storage) medium or data carrier.

For these reasons, for the very first time, the Guidelines include new sections on Artificial Intelligence and Machine Learning, and on Simulation, design or modelling, among other changes.

What follows is discussion of the main sections found in the 2018 EPO Guidelines on CII.

Mathematical methods, basics (G-II 3.3)

The section on mathematical methods has been completely revised and re-drafted, adding a distinction between contribution in producing a technical effect that serves a technical purpose, by its application to a field of technology and/or by being adapted to a specific technical implementation.

Concerning contribution by way of application to a particular field of technology, many examples are given, among others:

  • De-noising, detecting persons in a digital image, estimating the quality of a transmitted digital audio signal;
  • Mapping a speech input to a text output;
  • Error-correction coding of data for transmission over a noisy channel, compression of audio, image, video or sensor data;
  • Optimizing load distribution in a computer network;
  • Determining the energy expenditure of a subject by processing data obtained from physiological sensors;
  • Providing a genotype estimate based on an analysis of DNA samples, as well as providing a confidence interval for this estimate so as to quantify its reliability;
  • Providing a medical diagnosis by an automated system processing physiological measurements.

It should be noted that each of the above examples are rather specific, since it is stated that a generic purpose is insufficient. The Guidelines, however, fail to provide much clarity on the level of specificity required. However, in this situation we suggest relying on relevant case law (e.g. T1029/06) which states that use of meta-wording that includes the words “technical” or “physical” is insufficient, since otherwise anyone could argue that a claim was sufficiently technical by mere use of meta-language.

The Guidelines suggest (as an example) that sufficient specificity in the technical purpose can be achieved by functionally linking and limiting a method claim to the technical purpose, either explicitly or implicitly.

Concerning the contribution by way of specific technical implementation, a mathematical method also may contribute to the technical character of the invention independently of any technical application when the claim is directed to a specific technical implementation of the mathematical method and the mathematical method is particularly adapted for that implementation in that its design is motivated by technical considerations of the internal functioning of the computer. Here also, a generic technical implementation must be avoided, and alleging that the mathematical method is algorithmically more efficient than prior art mathematical methods does not solve the problem. Indeed, we must remember that technicality is to be assessed independently of the prior art, according to well-established case law.

In application to a field of technology as well as specific implementation, merely specifying the technical nature of the data or parameters of the mathematical method may not be sufficient to overcome the First Hurdle, because a mathematical method can be fed with real data but still be an abstract method without a technical effect or purpose, such that it represents mere mental acts.

Artificial Intelligence and Machine Learning (G-II 3.3.1)

The 2018 EPO Guidelines for the very first time provide a section relating to artificial intelligence and machine learning, which are first defined as computational models and algorithms for classification, clustering, regression and dimensionality reduction. They are considered per se to be of an abstract mathematical nature, irrespective of whether they can be “trained” based on (real) training data.

The applicant or practitioner drafting a claim related to artificial intelligence and machine learning should pay particular attention to expressions such as “neural network” or “AI routines”, since these does not contribute per se to the technical character of  the claim.

Instead, the applicant or practitioner seeking protection in Europe in this field should establish a causal link to the technical purpose e.g. use of mathematical method in a heart monitoring apparatus for the purpose of identifying irregular heartbeats, classification of digital images, videos, audio or speech signals based on edges or pixel attributes, and avoid using expressions that may encompass cognitive aspects of data (e.g. textual content of a document).

In this connection, it is quite helpful that the new Guidelines now specify that steps of generating the training set and training the AI models also may contribute to the technical character of the invention if they support achieving a technical purpose.

In addition, although non addressed by the new Guidelines,  expressions such as ‘neural network’, ‘artificial intelligence’ and ‘deep learning’ – although relatively well accepted and clear (eg., decision T1285/10 of the boards of appeal) – are not normally used in claims but mostly in the specification, as optimized models go well beyond these terms. Rather, buzzwords such as ‘trained recognizer’, ‘adapted model’, ‘learner component’ or even ‘determining’ an output (as in standard method steps) are often used in the claims and are further described in the specification. In addition, when more specific terms such as ‘neural network’ are used in the claims, they usually appear only in the dependent claims.

Simulation, design or modelling (G-II 3.3.2)

Claims directed to methods of simulation, design or modelling can be considered technical under the same principles of the other computer-implemented inventions.

For example, the computer-implemented simulation of the behavior of an adequately defined class of technical items, or specific technical processes, under technically relevant conditions qualifies as a technical purpose, as is the case of an electronic circuit subject to 1/f noise or of a specific industrial chemical process. This section of the new Guidelines makes it clear that lack of a final physical step of production of a product is not required for a simulation method to be patent-eligible.

This applies not only to simulation of behavior of a technical item but also in the context of computer-aided design of a specific technical object (product, system or process): determining a critical parameter such as a refractive index in an optical system or an operating parameter of a nuclear reactor constitute a technical contribution.

In drafting a method claim, the applicant or practitioner should avoid including steps that link achieving technical purpose to human decisions, as well as mere design-like features such as how to describe a multi-processor system in a graphical modelling environment.

Schemes, rules and methods for performing mental acts, playing games or doing business (G-II 3.5)

A section dedicated to schemes and rules for abstract purposes such as mental acts, playing games or doing business was already present in previous versions of the Guidelines. However, in the new version, extensive and consistent treatment of the topic has been provided, in order to help applicants and practitioners identify eligible subject-matters in these fields that have historically low rates of allowance.

This section is rather long with many, diverse examples, and may be summarized as follows.

If a method claim encompasses a purely mental/gaming/business realization of all method steps, it falls under the category of non-eligible methods. For example, when a simulation involves real world values obtained by a technical measurement, this is not eligible subject-matter unless:

  • A corresponding claim includes the step of carrying out the technical measurement, or
  • The use of other technical means (e.g. a computer, a measuring device etc.) or if it provides a physical entity as the resulting product, or
  • If it serves a technical purpose.

Steps involving ineligible subject matter can be included, however, they will be taken into account only if, in the context of the invention, they contribute to producing a technical effect serving a technical purpose.

In this framework, it is important for practitioners and applicants to pay attention to claim language, in that it could encompass both a technical meaning and a non-technical meaning, e.g. terms like “system” or “means” will be looked at carefully, because a “system” might e.g. refer to a financial organization and “means” to organizational units if it cannot be inferred from the context that these terms refer exclusively to technical entities.

According to the new Guidelines, examples of eligible subject-matter include:

  • When implementing a condition of matching random numbers for game paying, the use of a computer calculating a pseudo-random sequence or of mechanical means such as cubic dice or uniformly sectored reels;
  • A networked implementation of a game of chance like bingo, in which numbers physically drawn by an operator undergo a random mapping prior to transmission to remote players, since the scrambling of results has the technical effect of securing a data transmission; and
  • Interactive control of real-time manoeuvers in a game world, the display of which is subject to conflicting technical requirements.

On the contrary, as an example, computation which determine a game score or a skill rating for players, even if computationally complex, are usually considered non-technical.

Finally, concerning the Second Hurdle above, a few examples are given for eligible subject-matter where, although a technical problem exists, it is not solved but circumvented:

  • A reduction of memory, network, or computational resources achieved by limiting the complexity of a game does not overcome a technical constraint by a technical solution;
  • An automated accounting method that avoids redundant bookkeeping does not solve a technical problem if such reduction of workload results from a reduction of the number of operations to be performed due to the business constraints.

Inventions realized in a distributed computing environment (F-IV 3.9.3)

Finally, a new section has been added to the Guidelines, the new section which is related to computer-implemented inventions realized in a distributed computing environment, in order to give guidance on unity requirements.

The new Guidelines specify that it may be necessary to refer to the specific features of the different entities in the environment and to define how they interact to ensure the presence of all essential features, unless this is not essential to performing the invention. The different entities participating in the distributed system can be claimed without incurring a non-unity objection, however it may happen that not all claimed entities are new and inventive. This is the case when for example an entity encodes information in a more efficient way, but an information-receiving entity decodes such encoded information in a conventional way: the information-receiving entity is normally neither new nor inventive.

Further Comments

On May 30, 2018 the EPO held an international conference on patenting AI in Munich to explore this topic in detail. The author of this article was invited as a speaker. Material and videos of the conference have been published on the e-courses section of EPO website.

However, the Guidelines only include a short section on the matter with respect to what was discussed in the conference and other related events.

For example, application of equivalence according to Article 69 of the European Patent Convention has no guidance to date. There is no hint in the Guidelines as to how to respond to the following questions:

  • Can a patentable AI method claim be considered equivalent to a classic method claim with a definite, detailed sequence of steps, for example, in a decision tree format?
  • What prior art effect would a disclosure directed to classic computer-implemented inventions have on claims directed to AI methods related to the same technical purpose, and vice versa (i.e. AI disclosure compared to classic method claim).

Applicants and practitioners may consider alternative solutions to the same technical problem and try to encompass both classical and AI steps in a claim, if the claim does not become too general. Alternatively, a separate patent application directed specifically to decision tree and/or AI methods may be filed, in order to prevent any design-around.

Another important point where the new Guidelines are silent is how to assess the requirement of sufficiency of disclosure in the case of AI patents. In this situation, we suggest providing as much detail as possible, including comparative examples and parameter ranges, as well as exemplary training data sets.

Finally, no change in the definition of the skilled person is specifically included in the new Guidelines, although this topic was discussed in the above-referenced conference. This leaves open issues such as whether the skilled person has AI tools to perform routine work and whether the skilled person can be substituted by a machine trained in the art, and further whether this can be fictional or should refer to a real state-of-the art machine.

These issues, as well as other problematic issues like inventorship, may be assessed in future cases before national Courts or before the EPO Boards of Appeal.


The ICT parts of 2018 EPO Guidelines include very important clarifications of ICT patentability requirements, as a result of valuable work performed over the last three years. No other significant structural revisions are anticipated in the coming years, but rather only fine-tuning of the wording depending on user feedback.

The new Guidelines address artificial intelligence inventions and clarify that they are subject to the same criteria as the other computer-implemented inventions, by relying on stable jurisprudence. In any event, careful consideration of patenting strategy is critical for obtaining the desired protection.


Image Source: EPO. CC BY-NC-ND 2.0.


Warning & Disclaimer: The pages, articles and comments on IPWatchdog.com do not constitute legal advice, nor do they create any attorney-client relationship. The articles published express the personal opinion and views of the author as of the time of publication and should not be attributed to the author’s employer, clients or the sponsors of IPWatchdog.com.

Join the Discussion

3 comments so far.

  • [Avatar for Ternary]
    November 15, 2018 01:32 pm

    The EPO is most certainly not an organization that is known for issuing patents in a haphazard way. Technical utility is an important consideration for patent eligibility. The practical and well documented approach to come to some sort of agreement on patent eligibility is a type of practical approach that I would have associated with the US patent system. No more.

    This type of careful consideration seems no longer possible here. My own preference is “anything under the sun.” But I understand that different interests may require some form of compromise. I find “technical utility” a very acceptable criterion. It would be sublime if EPO would make the process of obtaining a patent more affordable.

    Great job Andrea in summarizing and explaining this complex matter. Very timely also.

  • [Avatar for PatentGuy]
    November 15, 2018 10:51 am

    Thank you for the detailed explanation!

  • [Avatar for Anon]
    November 15, 2018 07:55 am

    Many thanks