Sufficiency At The EPO For AI Inventions

Sufficiency At The EPO For AI Inventions

To print this article, all you need is to be registered or login on

At the European Patent Office, for the purposes of examination,
applications disclosing and claiming subject matter relating to
Artificial Intelligence or Machine Learning are treated in a
similar manner to applications disclosing and claiming subject
matter relating to other types of mathematical methods or
algorithms implemented on a computer (see the EPO’s Guidelines
for Examination (GfE) on AI inventions).

The guidance provided in relation to such
computational and mathematical methods focusses on analysing
whether the method contributes to the technical character of the
invention. Only those features of a claim that contribute to the
technical character of an invention are taken into account when
assessing inventive step of the invention.

However, an additional issue that has started to appear in
connection with AI inventions at the EPO – that has
traditionally been less of an issue for other types of
computational and mathematical method inventions – is that of
sufficiency of disclosure in the patent application.

Sufficiency at the EPO

Article 83 EPC requires that a European patent
application “shall disclose the invention in a manner
sufficiently clear and complete for it to be carried out by a
person skilled in the art”

The GfE expand on what is required for sufficiency of disclosure
in this context: “A detailed description of at least one
way of carrying out the invention must be given. Since the
application is addressed to the person skilled in the art, it is
neither necessary nor desirable that details of well-known
ancillary features are given, but the description must disclose any
feature essential for carrying out the invention in sufficient
detail to render it apparent to the skilled person how to put the
invention into practice.”
(GfE F-III 1)

The GfE further indicate what would be regarded as insufficient
disclosure in this regard: “Sufficiency of disclosure
cannot be acknowledged if the skilled person has to carry out a
research programme based on trial and error to reproduce the
results of the invention, with limited chances of
(GfE F-III 3)

For more ‘deterministic’ or ‘predictable’
computational and mathematical method inventions, a functional
description of each step or feature of the method would typically
be considered to satisfy the sufficiency requirement, without
needing to describe how the steps are to be programmed in the

However, some recent decisions from the EPO’s Boards of
Appeal suggest that this may not be ‘sufficient’ for
AI-related inventions, which may be more ‘probabilistic’ or
less predictable in nature. In particular, the described
functionality of an AI system may be achieved only for certain
trained parameters of the AI system, e.g. weighting values at
nodes, or when certain training data is used to obtain the system

As such, for AI inventions it is important to consider: has the
skilled person really been provided with enough information in the
patent specification to be able to implement the invention to
achieve the desired functionality without undue burden? Whereas a
relatively high-level, functional description of an algorithm may
be sufficient for mathematical or computational methods of a more
predictable or deterministic nature, for AI inventions training
data or model parameters may need to be disclosed in more detail
for the skilled person to be able to reproduce the invention and
achieve the described functionality in order for the sufficiency of
disclosure requirement to be met. For instance, one or more
examples of specific sets of training data or model parameters may
be provided.

Sufficiency of disclosure was raised as an issue in each of the
two Technical Board of Appeal decisions summarised briefly below,
where the application in each case related to AI subject

T 0161/18

The application in this case disclosed a method for determining
cardiac output from an arterial blood pressure curve measured at a
peripheral region. This was described as less invasive than known
measurement methods. Weighting values to estimate aortic pressure
from peripheral blood pressure were determined using an artificial
neural network (ANN).

The patent specification disclosed: the input data for training
the ANN should cover a wide range of patients of different ages,
genders, constitutional types, health conditions, etc.; a
particular, standard training technique; and, a particular,
standard ANN architecture.

The EPO Board of Appeal decided that the application did not
sufficiently disclose the invention because the application did not
disclose which input data were suitable for training the ANN
according to the invention, or at least one data set suitable for
solving the problem at hand. In particular, the Board of Appeal
decided that the skilled person could not rework the ANN and so the
application was refused.

It is notable in this case that sufficiency was only first
raised as an issue at the appeal stage of proceedings.

It is also of note that no sufficiency (or similar) objection
was raised during prosecution of the corresponding application in
the US.

T 1191/19

The application in this case related to using a meta-learning
scheme to train multiple identifiers related to personalised
interventions in neurorehabilitation. The EPO Board of Appeal
decided that the application did not sufficiently disclose the
invention, and noted the following points.

The Board of Appeal said that the application did not disclose
any example set of training data and validation data which the
meta-learning scheme required as an input. The Board noted that the
application did not even disclose the minimum number of patients
from which training data should be compiled to be able to give a
meaningful prediction and the set of relevant parameters. The Board
further noted that the structure of the ANNs used as classifiers,
their topology, activation functions, end conditions or learning
mechanism were not disclosed.

At the level of abstraction of the application, the Board said
that the available disclosure was more like an invitation to
undertake a research programme. The Board concluded that, under
these circumstances, the skilled person would not be able to
reproduce, without undue burden, the application of the disclosed
meta-learning scheme to solve the problem of predicting
personalised interventions for a patient in processes the substrate
of which is neuroplasticity.

It is again of note that no sufficiency (or similar) objection
was raised during prosecution of the corresponding application in
the US.

Tips for drafting AI-related patent applications

In view of the above, some useful tips for increasing the
chances of successfully prosecuting an AI-related patent
application through to grant at the EPO – taking into account
both sufficiency and inventive step considerations – include
the following.

  • Identify in which part(s) of the disclosed system the novel
    concept lies, and focus the description on this part.
    • An AI system can include several different components, such as
      the architecture of the AI model (e.g., neural networks, deep
      learning, number of layers / nodes), the training method (e.g.,
      supervised learning, unsupervised learning, reinforcement
      learning), the training method (encoding techniques, inputs and
      corresponding outputs, vectors), and integration into a wider
      system (e.g. pre-processing, post-processing). Only those
      components which are key to the invention, or are not standard in
      the AI field, may need to be described in detail.
  • Consider the bigger picture and describe how the novel concept
    fits into an overall system and contributes to an overall technical
  • Include testing, experimental or simulation data to evidence
    recited technical effects and/or reproducibility.
  • Draft the patent application to include the technical
    motivations and/or advantages associated with described features of
    the invention.
    • The application under consideration in Technical Board of
      Appeal case T 2653/16 related to a banknote processing machine
      which could detect which banknotes to reject, e.g. because they
      were damaged. The machine automatically learned various sensor
      values associated with banknotes to be rejected. Beneficially, the
      machine did not need to know which banknotes in the training set
      should be rejected as not fit for circulation, which avoided the
      need to manually pre-sort a training set of banknotes, and this was
      found to be technical in nature by the Board.
  • Consider whether any structural or functional modifications to
    known systems are needed to implement the invention.
    • The application under consideration in Technical Board of
      Appeal case T 1749/14 related to increased security for
      point-of-sale (POS) transactions. It was considered that the
      notional ‘business person’ may come up with the abstract
      idea of avoiding the need for a customer to provide PIN and account
      information to a merchant. However, in this case new
      infrastructure, devices and protocols were needed to implement the
      idea. It was decided that this would go beyond the considerations
      or capabilities of the notional business person, and so was
      considered to be technical in nature.
  • Consider whether any structural or functional modifications to
    known systems are needed to implement the invention.

The content of this article is intended to provide a general
guide to the subject matter. Specialist advice should be sought
about your specific circumstances.

POPULAR ARTICLES ON: Intellectual Property from UK

Batman Won Another (Trade Mark) Battle

K&L Gates

Batman may be a superhero but it is the General Court who has come to the rescue following an invalidation action bought against DC Comics, Warner Bros subsidiary…

Marvellous Mushrooms!

Marks & Clerk

For a few years now, easily compostable packaging material, such as Mushroom® Packaging, which uses MycoComposite™ biomaterial, has been available to create sustainable alternatives…

Unified Patent Court (UPC)

Horten Advokatpartnerselskab

The UPC represents a significant milestone in the field of intellectual property, offering a unified framework for patent enforcement across participating European countries.

AI Tests Intellectual Property Boundaries

Travers Smith LLP

As businesses invest more heavily in AI, what intellectual property challenges are they likely to encounter? How do we avoid AI have a “chilling” effect on human creativity and enable…

Leave a Reply

Your email address will not be published. Required fields are marked *