The High Court judgement in Emotional Perception AI Ltd v Comptroller-General of Patents, Designs and Trade Marks  EWHC 2948 (Ch), which was handed down at the end of November, has triggered a major shift in UKIPO practice for examination of inventions involving Artificial Neural Networks (ANN).
The UK Intellectual Property Office (UKIPO) have undertaken wide ranging consultations and several round table events on AI and IP. We have previously compared the approaches of the UKIPO and European Patent Office (EPO) here, and published an analysis of the guidance for “Examining patent applications relating to artificial intelligence (AI) inventions” published by the UKIPO in October 2022 here. In the following discussion, we take a look at what this new judgement, and the associated statutory guidance issued by the UKIPO, will mean for applicants in this space going forward.
What has just happened?
In Emotional Perception AI Ltd v Comptroller-General, a decision by a UKIPO hearing officer was appealed to the High Court. The hearing officer decision refused a patent application directed to an ANN based invention under the “program for a computer” exclusion from patentability (UKPA 1977 S1(2)(c)). The court disagreed with the hearing officer, stating essentially that an ANN is not itself a program for a computer and therefore would fall outside the “program for a computer” exclusion from patentability. Following this unexpected turn of events, an appeal against the judgement from the UKIPO seemed likely. Instead, within a week of the judgement being handed down, the UKIPO released associated statutory guidance explicitly stating that “the office is making an immediate change to practice for the examination of ANNs for excluded subject matter. Patent Examiners should not object to inventions involving an ANN under the “program for a computer” exclusion of section 1(2)(c).”
This change in practice appears to be excellent news for applicants seeking to patent ANN inventions that may previously have struggled due to the “program for a computer” exclusion. However, it remains to be seen exactly how the judgement will be implemented by the UKIPO. In particular, the Manual of Patent Practice (MoPP) and the UKIPO’s guidance for examining patent applications relating to AI inventions (more generally, rather than specifically ANN) have yet to be updated; future updates to both are promised in the recently issued statutory guidance.
While already extremely helpful for applicants looking to protect inventions “involving ANNs”, it is possible that the judgement could have a more wide-reaching impact on UKIPO practice relating to excluded subject matter. For example, some of the discussion in the judgement concerning the application of the Aerotel test (for a full discussion of this test, see our article comparing the approaches of the UKIPO and EPO, linked above) could be seen as being applicable to the examination of computer implemented inventions more broadly, and perhaps even to the mathematical method exclusion (for completeness, the mathematical method exclusion is not directly considered in the judgement). The new development represents a further divergence between EPO and UKIPO practice regarding the patentability of AI inventions, albeit a divergence that this time favours the UK in terms of identifying which jurisdiction is more favourable for AI patenting. So what exactly does the judgement say, and what does it mean for day-to-day prosecution, and for the broader strategic decisions facing applicants in this area?
A deep dive into the judgement
Briefly, Emotional Perception AI Ltd v Comptroller-General relates to an invention for providing media file recommendations using an ANN. In a specific example, music files are classified by being passed through a first ANN which analyses semantic properties (for example, labels such as “happy”, “sad”, or “relaxing”) and a second ANN which analyses physical properties (for example, metrics such as tone, timbre, speed, and volume). The second ANN is then trained using back-propagation in order to adjust the weightings of the second ANN such that semantic similarity identified by the first ANN between tracks would be reflected by physical similarity identified by the second ANN between those same tracks. Once the second ANN training process is completed, the second ANN can use physical properties of songs without semantic labels to recommend to an end user songs which are nonetheless semantically similar, i.e., it has learned to discern semantic similarity from physical properties.
To determine whether it is appropriate to classify an ANN as “a computer program as such”, the judgement splits ANN inventions into two types: a hardware ANN (referred to in the judgement as “a physical box with electronics in it”) and an emulated ANN (in which a conventional computer runs a piece of software which enables the computer to emulate the hardware ANN). The judgement considers, for each type of ANN and during each phase of execution (training and inference), where the computer is and where the program is. There appeared to be consensus amongst the parties that in the case of a hardware ANN, there was no computer program to which the exclusion might apply. When considering implications of this for an emulated ANN, the judgement makes a distinction between the implementation of instructions input by a human (such as a programmer) and a trained ANN, stating in paragraphs 54 to 58 (emphasis added):
“[for a hardware ANN] the hardware is not implementing a series of instructions pre-ordained by a human. It is operating according to something that it has learned itself … I do not see why the same should not apply to the emulated ANN. It is not implementing code given to it by a human…
I therefore consider that the “decoupling” can be achieved and is correct and the emulated ANN is not a program for a computer for these purposes.”
The judgement is satisfied that while a hardware ANN did not involve a program for a computer, and an emulated ANN was also not a program for a computer, programming activity was involved in the training phase, and so “the only remaining candidate computer program is therefore the program which achieves, or initiates, the training” (paragraph 59). However, when considering the extent to which the training claim actually claimed the computer program that achieves or initiates the training, the judgement states that what is special about the training process is not contained within that computer program: “What is said to be special is the idea of using pairs of files for training, and setting the training objective and parameters accordingly. If that is right, and I consider it is, then the actual program is a subsidiary part of the claim and is not what is claimed. The claims go beyond that.” (Paragraph 61).
Although the above paragraphs of the judgement refer specifically to ANNs, similar arguments would appear to be applicable to AI inventions of other types, in which a Machine Learning model can be implemented using dedicated hardware, or through software running on a conventional computer. It is easy to envisage arguing that claims to a particular ML process (that is not necessarily an ANN), whether implemented in hardware or as a software emulation, are not claims to a computer program. Similarly, while training of a model may be initiated or achieved using a computer program, if what is special about the training regime goes beyond what can be encompassed by the computer program involved in the training, then presumably a claim to that training method would also not be a claim to a computer program. It will be interesting to see what the updates made to the MoPP and the UKIPO’s guideance for examining patent applications relating to AI inventions have to say on these and other points.
Helpfully, the judgement considers the question of technical contribution, in the event that that the judgement were to be wrong on the question of whether the training process involved a claim to a computer program; this is very relevant to questions 3 and 4 of the Aerotel test. One possible interpretation of technical contribution turned on the effect achieved by transmission of the recommended file. Paragraphs 76 to 78 of the judgement state (emphasis added):
“The Hearing Officer seemed to consider that a subjective appreciation of the output of the system was just that, subjective and in the user, and therefore not a technical effect. I do not consider that to be the correct analysis… The correct view of what happened, for these purposes, is that a file has been identified, and then moved, because it fulfilled certain criteria… So the output is of a file that would not otherwise be selected. That seems to me to be a technical effect outside the computer for these purposes, and when coupled with the purpose and method of selection it fulfils the requirement of technical effect in order to escape the exclusion. I do not see why the possible subjective effect within a user’s own non-artificial neural network should disqualify it for these purposes.”
The above reasoning, as stated in the judgement, looks to the end result as being something which helps to take the case away from being a case of a computer program “as such”. However, the judgement indicates that there is another way of approaching the matter if it is assumed, contrary to the earlier conclusions, that the computer program is either the training program or the overall training activity. In this circumstance:
“[I]f one is assuming for these purposes that the computer program is either the training program or the overall training activity… the resulting ANN, and particularly a trained hardware ANN, can be regarded as a technical effect which prevents the exclusion applying… I therefore consider that, insofar as necessary, the trained hardware ANN is capable of being an external technical effect which prevents the exclusion applying to any prior computer program. There ought to be no difference between a hardware ANN and an emulated ANN for these purposes.”
The judgement thus provides multiple routes by which claims involving an ANN (and possibly therefore other model architectures?) may avoid being caught by the computer program exception, either through the relevant claim not being directed to a computer program at all, or via the technical effect achieved by the claim.
Further, although the judgement does not (for procedural reasons) consider whether the “mathematical method” exclusion applies in the present case, it would seem possible that the reasoning in the judgement could have ramifications for how this exclusion is applied going forward. For example, the hearing officer decision is quoted as stating:
“A claim to an ANN or the algorithm by which it is trained, in a general and abstract sense, relates wholly to a mathematical method and it fails at step 3 [of Aerotel]. Even if there is something more than a mathematical method present, I cannot see how it is technical in nature and so it would not satisfy step 4.”
It would seem that at the very least, the contents of the judgment relating to the assessment of technical contribution could be applied to step 4 of the Aerotel test when being used to evaluate a claim with respect to the mathematical method exception.
What about the EPO?
The EPO takes the approach that artificial intelligence and machine learning are based on computational models and algorithms that are in themselves of an abstract mathematical nature, and consequently the guidance for excluded subject matter provisions, and specifically for mathematical methods, should apply. In practice, this means AI applications are assessed for patent eligibility using the COMVIK approach, in which claim features found to fall within an excluded category can only be considered for the purposes of inventive step if they contribute to the solution to a technical problem.
We have previously expressed doubts about the extent to which the UK and EPO approaches to excluded subject matter arrive at the same destination, at least at first instance, but it seems likely that the Emotional Perception AI judgement will result in a significant further divergence between the UK and EPO in the extent to which ANNs (and possibly other AI subject matter as discussed above), are caught by the exceptions to patentability.
The EPO position on the technicality of neural networks has been repeatedly affirmed in case law, including in EPO Board of Appeal decision T0702/20 which defines a neural network as “class of mathematical functions which, as such, is excluded matter”. This decision goes on to state that other “non-technical” matter “can therefore only be considered for the assessment of inventive step when used to solve a technical problem, e.g. when trained with specific data for a specific technical task.” Provision of a file in a recommendation system seems highly unlikely to fall within the EPO’s definition of a specific technical task, as this question has also been considered in past EPO case law. In EPO Board of Appeal decision T1983/18, a computer-based method for generating recommendations of mobile applications to a user by classifying the applications into different categories and determining a “relevance” between categories is considered. The Board of Appeal outlined in the decision that: “this method is not based on technical considerations, but on a set of non-technical assumptions or requirements for estimating a user’s preferences, and thereby determining the contents of the recommendation. The overall effect is that of providing recommendations according to the user’s estimated preferences, which is non-technical”. T1983/18 followed another EPO Board of Appeal decision T0306/10, which considers a system for recommending songs to users, states that “…the selection of an item, for example a song, for recommendation to a user does not qualify as a technical purpose. From a technical point of view it is irrelevant what songs are recommended to a user.” This assessment in EPO case law bears strong similarity to the hearing officer’s approach to technicality in Emotional Perception AI, and appears to shut the door to technicality for recommendation systems.
When the above EPO decisions are juxtaposed with the Emotional Perception AI Ltd v Comptroller-General judgement, the divergence in the approaches to technicality of ANNs and the possibility for technical contribution to be provided by recommender systems, is clear. As the UKIPO have issued statutory guidance based on the judgement, and indicated that more substantial practice updates are on the way, an appeal against the judgement by the UKIPO appears unlikely. Accordingly, in at least the short to medium term, UKIPO practice towards ANN inventions at least (and potentially AI inventions more generally) is likely to be more favourable to applicants than EPO practice. Further, even were the EPO minded to adapt their practice to more closely match that of the UKIPO, there is no straightforward way for the necessary changes to be made. A decision by the Enlarged Board of Appeal could change how the EPC is interpreted, however this would require some divergence in EPO case law, which is not presently the case; as indicated by the decisions cited above, the case law in this area is frustratingly consistent. If there were ever sufficient political will to do so, the European Patent Convention itself could be revised by a diplomatic conference. However, such revision has happened once in the existence of the EPC, and would require years of negotiations. It is therefore likely that the UKIPO practice will remain significantly more friendly to ANN inventions than the EPO practice for some time.
Emotional Perception AI Ltd v Comptroller-General is a positive step for applicants seeking to patent ANN inventions. While we have certainly had false dawns before, it is a stated aim of the UK government to be a location of choice for AI development, and the Emotional Perception AI Ltd judgement is consistent with this aim. Both the definitive language used in the judgement, and the rapidity with which the UKIPO statutory guidance was issued following the judgement, are promising signs. We are cautiously optimistic that the judgement may herald a shift towards a more favourable approach to the patentability of AI at the UKIPO, and look forward with interest to the publication of updates to the MoPP and the UKIPO’s guideance for examining patent applications relating to AI inventions in light of Emotional Perception AI Ltd v Comptroller-General. In the meantime, the High Court has provided some extremely helpful analysis for quoting in the next UK office action response in this subject matter area…