Where next for patentability of AI inventions in the UK?

In December 2023, it was announced that the UK Intellectual Property Office (UKIPO) had received permission to appeal the High Court’s ruling in Emotional Perception AI Ltd v Comptroller-General of Patents, an appeal that will be heard on 14 May 2024. The Court of Appeal will therefore be revisiting the question of the patentability of software inventions and will have its first opportunity to comment on the patentability of artificial neural network (ANN)-based inventions.

The ongoing Emotional Perception case has generated significant interest in the artificial intelligence (AI) industry and patent profession alike, at a time when global interest in use and development of AI technology is sky rocketing – indeed, the UK Government recently identified AI as one of the five technologies of tomorrow critical to helping drive future discoveries and economic growth in its recent Science and Technology Framework. Emotional Perception is the first UK patent court case to focus on AI technology, and the importance of the case is underlined by the fact that the UKIPO, immediately following the first instance judgement, changed their practice for examination of ANN inventions – a change that, at least for the time being, effectively removes ANN inventions from the scope of the computer program exclusion and has led to a flurry of AI-related patents being allowed.

Seen by many as a long-awaited breakthrough towards the recognition of AI as a technical field in its own right, eligible for patent protection commensurate with its increasing role in driving innovation worldwide, the first instance judgement is also highly controversial, as it represents a significant divergence from established European Patent Office (EPO) practice and has sparked debate and has polarised views amongst the patent community. If not overturned on appeal, the Emotional Perception case will make the UK a far more favourable patentee-friendly place for patenting AI inventions for the foreseeable future.

The decision at first instance

The invention at the heart of the Emotional Perception appeal relates to a system and method for providing improved media file recommendations to an end user which is implemented using an artificial neural network (ANN) trained in a specific way which is said to align its output more closely to how a human semantically perceives the content of an input file.

Initially, the UKIPO rejected Emotional Perception’s application on the grounds that the invention was excluded from patentability for being a computer program as such. That rejection was then reversed by the High Court which concluded (among other things) that the claimed invention, in so far as it relates to the training and use of a trained ANN, did not involve a computer program at all so did not engage the exclusion, and even if it did, the novel method of identifying and providing files to a user was a sufficient “contribution” and sufficiently “technical” to prevent the invention from being considered a “computer program as such”. It is this decision which is now under appeal.

The Law

Since 2007 the UKIPO have used the test set out in Aerotel Ltd v Telco Holdings Ltd  to determine whether an alleged invention relates to excluded subject matter. This test involves the following steps:

• Properly construe the claim.
• Identify the actual contribution (although at the application stage this might have to be the alleged contribution).
• Ask whether it falls solely within excluded matter.
• If the third step has not covered it, check whether the actual or alleged contribution is actually technical.

A key question at the heart of the Emotional Perception case was: what was the actual contribution of the claimed invention? Counsel on both sides accepted that the invention related to an ANN-based system for providing improved file recommendations and that the fundamental insight on which the invention was based was the training of the ANN to analyse physical properties of a file using pairwise comparisons of training files. Once trained the trained ANN could then be used to identify, swiftly and accurately, files from a database which correspond semantically to a target file, and to provide file recommendations to a user device.

The judge concluded that although potentially implemented in software, a software ANN was not operating a program in the traditional sense of a set of instructions provided by a programmer because, rather than processing data on a step-by-step instructional basis, machine learning uses training data to self-learn and reconfigure an ANN to solve a specific problem. No other candidate program was identified, and for this reason the “contribution” was held not to engage the computer program exclusion at all. The judge went on to consider step 4 of the Aerotel test anyway and concluded that the contribution was technical in nature and thus also not a computer program as such, because: (i) the end result of sending the file recommendations to an end user provides a relevant external technical contribution; and (ii) a trained ANN itself can also amount to a technical contribution external of the training it received.

Differences of approach in the UK and the EPO

The UKIPO and EPO’s approaches to the assessment of patentable subject matter diverged in the early 2000s when the EPO changed their practice in PBS Partnership/Pension Benefits T931/95. Following that decision, the EPO adopted an “any hardware” approach whereby if a claim involves the use of technical means (e.g. a computer) or is directed to such technical means it avoids the exclusions. As a result, the EPO do not currently raise excluded subject matter objections to “computer-implemented” inventions unless a claim is directed to a purely abstract concept.

The question of whether or not an invention gives rise to a “technical effect” (i.e. whether or not a claimed invention provides a technical solution to a technical problem) remains part of the EPO’s assessment of patentability. However, rather than being considered as part of an assessment of patentable subject matter, it is considered as part of the EPO’s assessment of inventive step.

For computer-implemented inventions, assessment of inventive step involves an exercise of dissecting a claim into technical and non-technical features. Features which serve a technical purpose and contribute to solving a technical problem are taken into account when determining whether or not an invention involves an inventive step. In contrast, features of a claim which are identified as being non-technical are considered to form part of the background to the “technical problem” that an invention might be said to address and are essentially ignored for the purposes of assessing inventive step. Examples of what are considered technical purposes and processes which enable examiners to assess whether or not particular features are “technical” are set out in the EPO’s Guidelines. If an invention is not considered to address a “technical” problem, then it will be considered to be unpatentable in the EPO.

This approach was explicitly endorsed by the EPO’s Enlarged Board of Appeal in Bentley/Pedestrian Simulation G1/19.

The UK Courts have repeatedly stated that the UKIPO and EPO’s approaches are capable of reconciliation and amount to two different ways to (in principle) arrive at the same result. The first instance Emotional Perception decision, however, puts the two approaches into direct conflict.

First, the High Court’s ruling that ANNs inventions do not engage the computer program exclusion at all will in many cases allow AI-based inventions to pass directly to the inventive step stage and even to grant without any consideration of the technical character of the invention.

In contrast to the EPO, assessment of inventive step in the UK is limited to an assessment of whether or not it would be obvious to a person skilled in the art, using common general knowledge to modify any existing item of prior art to arrive at a claimed invention. All claim features are taken into account as part of this assessment and there is no formal separate assessment of the technical character of a claim at the inventive step stage.

Second, the High Court’s ruling that the end result of providing improved file recommendations provides a relevant external technical effect is in direct conflict with the established jurisprudence of the EPO Boards of Appeal who have consistently refused patent applications relating to recommendation systems and methods for lacking a relevant technical effect because any effects of the recommendations (e.g. playing a song, or requesting further recommendations) depends on the subjective choices of the user (see T 0306/10, T 1869/08, T 1983/18, and more recently T 0183/21).

These decisions underline the EPO’s emphasis on the concepts of technical problem, technical purpose and an objective technical effect. By contrast, there was no consideration by the High Court in the Emotional Perception decision of whether or not the invention served a technical purpose or solved any kind of technical problem, or even whether the output of the ANN was to be used as part of a subsequent technical process. Although the judge acknowledged the subjective effect of the recommendation on the user, he then focused on how the recommendation was generated and concluded that this was sufficient to overcome the “technicality” check in the final step of the Aerotel test. Hence, the judgement supports the notion that a trained ANN, processing data through its nodes in a technical way using logic it has learned itself, fulfils the requirements of the Aerotel test regardless of the purpose and the subsequent use of a machine’s output.

The long term impact of the Emotional Perception decision will depend upon the views of the Court of Appeal.

The conclusion at first instance that the computer program exclusion requires the presence of an identifiable set of instructions provided by a human that a computer is to perform is highly questionable and may well be overturned. However, equally important will be how the Court of Appeal addresses the question of whether the technical purpose or nature of the output of an invention is something which needs to be taken into account when assessing patentability. If it does not, then that would leave the door firmly wide open in the UK to the patentability of a wide range of inventions that leverage AI and machine learning, including those in traditionally non-technical fields such as finance, business and administration which would likely be considered to be unpatentable by the EPO.

That having been said, this second part of the Emotional Perception decision seems to be significantly more grounded in UK precedent. In previous decisions such as Protecting Kids the World Over vs Comptroller and Gemstar v Virgin Media, the UK High Court has considered that transfer of data files is sufficiently “technical” to overcome the final check in the Aerotel test, and that there is therefore no need to consider whether or not the output of an invention has a “technical” purpose.

We can expect the outcome of the appeal sometime in the latter half of 2024.