In Emotional Perception[1], the potentially ground-breaking decision for patent-rights holders at the end of last year, the High Court found that claims to a trained artificial neural network (ANN) did not fall foul of the statutory non-patentable subject matter exclusions for being a "computer program as such".

The UK Intellectual Property Office (IPO) is appealing the decision, and has published its grounds of appeal.

With the Court of Appeal set to hear the appeal on 14 May 2024, we have taken a look at the issues under appeal and the possible consequences for the patentability of artificial intelligence (AI) inventions in the UK. As noted in our previous reporting of this case ('Get ready for patents for AI' and 'Rational reaction to Emotional Perception?'), there was a general consensus the High Court's decision was more favourable to patentees than expected, and likely at odds with what the European Patent Office (EPO) would have decided were the claims assessed under its framework.

The grounds of appeal

Four grounds of appeal have been raised:

  1. The judge was wrong to find that the claims did not engage the exclusion from patent protection for "computer programs as such".
  2. The judge was wrong to rely on the IPO's 'concession' during written submissions to the judge that a hardware ANN would be unlikely to engage the exclusions as a basis for finding a software-implemented ANN is not caught by the computer program 'as such' exclusion.
  3. The judge was wrong to exclude consideration of the 'mathematical method as such' exclusion.
  4. The judge was wrong to hold that the claimed invention involved a substantive technical contribution.

There is optimism that the formulation of these grounds will lead to helpful guidance from the Court of Appeal on how the patentability of ANNs is to be assessed in the UK, whichever way they decide the issues.

The framework for assessing the patentability of ANNs

In the decision the judge found that, as a matter of construction, the claim was not directed to a computer program at all. This meant that the issue of whether the supposed program was a mere program 'as such' and not patentable, or whether there was disclosure of a 'technical contribution' and therefore the program was more than a program 'as such' and patentable (at least with regard to the computer program exception) wasn't part of the judge's eventual decision.

However, in case the judge were found to be wrong on any appeal he also considered whether the claim did provide a technical contribution.

These two strands of the judge's reasoning are the focus of the first and fourth grounds of appeal, respectively.

There is no precise definition (in statute or case law) as to what constitutes a 'technical contribution', despite it being the key issue in analysing whether a computer-implemented invention relates to non-patentable subject matter. Consequently, any further guidance or examples from the Court of Appeal of what satisfies the requirements of a technical contribution that result from the fourth ground of appeal will be welcomed by the industry and patent professionals alike.

However, it is the first ground of appeal that is of potentially wider significance to applicants in the field of AI. If this ground is dismissed and the Court of Appeal upholds the judge's finding that a trained ANN is not a computer program at all (and so no consideration of technical contribution is required in assessing patent eligibility), it paves the way for clams directed to trained ANNs being outside the computer program exclusion for patentable subject matter, even if the ANN is used for what might conventionally be understood as a non-technical purpose.

This leads to another important issue raised by the grounds of appeal.

The role of the mathematical method exclusion

Despite its minor role in the case so far, the potential significance of the mathematical method exclusion should not be overlooked. It is arguably an equally relevant exception in the field of AI related inventions.

In the Hearing Officer's decision that preceded the case before the High Court, the IPO twinned its mathematical method and computer program exclusion objections by arguing:

  1. that the software-emulated ANN cannot truly be decoupled from the underlying software platform that supports it, supporting the position the ANN is a type of computer program; and
  2. insofar as the emulated ANN can be decoupled from the underlying supporting software platform (and so arguably is not itself a computer program), it is an abstract model that relates wholly to a mathematical method.

For procedural reasons, the judge did not consider the issue of whether the claim should be excluded for relating to a mathematical method 'as such'. This means the decision of the High Court does not place any new restrictions on the IPO's capability to raise objections to claims directed to ANNs under this limb of the exclusions. This leaves open the possibility that excluded subject matter objections will in practice pivot from being raised under the computer-program exclusion to the mathematical method exclusion. This is the typical approach of the EPO, who generally view AI and machine learning as being based on computational models of an abstract mathematical nature as opposed to computer progams.

In this context, it can be appreciated that consideration by the courts of the mathematical method exclusion will be important for gaining a clearer and more complete understanding of the patentability of ANNs in the UK. Let's hope the Court of Appeal take the opportunity to do so.

For further information, or to discuss any of the points raised in this article, contact Arnie Francis or Alex Brodie.

Footnotes

[1] Emotional Perception AI Ltd v Comptroller-General of Patents, Designs and Trade Marks [2023] EWHC 2948 (Ch) (21 November 2023).