Not quite eight months have gone by since, at a snap, the High Court judgement in favour of Emotional Perception AI Ltd upended the patentability (or otherwise) of artificial neural networks (ANNs) in the UK. The UK IPO first suspended their AI patentability guidelines, and then re-introduced them with a (narrow) ANN carve out, and, at the eleventh hour, went to the Court of Appeal to contest the decision. The world has turned and now the Court of Appeal has allowed the UK IPO’s appeal against the decision. We find ourselves snapped back to the pre-November 2023 practice on ANNs (complete with further UK IPO statutory guidance hot off the press), but why?
The High Court decision held that:
“…firstly, claims to ANNs were not claims to computer programs as such and as such were simply not caught by the exclusion to patentability of computer programs under UK law; and
secondly, even if a claim to an ANN was a claim to a computer program as such, in this case there was nevertheless a technical contribution which again took the claim outside of the exclusion.”
When is a computer program not a computer program?
The first contention is undoubtedly the one that grabbed the most headlines, led to a stark change in the UK IPO guidance, and over which the most ink was spilled. At the High Court the view was taken that the various “weights” which dictate how an ANN functions do not constitute a computer program. The High Court seemed to take quite a narrow view on a computer program, very much viewing this in terms of human provided “if-then-else” type stepwise logical instructions.
That the Court of Appeal has taken a more expansive view on what a “computer program” is perhaps the least surprising outcome given the myriad programming paradigms that don’t quite follow “if-then-else…”
Here Lord Justice Birss, with Lord Justice Arnold and Lady Justice Nicola Davies in agreement (a judicial team with an extraordinary amount of patent expertise), took the view that
“[the weights of the ANN] are a set of instructions for a computer to do something. For a given machine, a different set of weights will cause the machine to process information in a different way. The fact the set does not take the form of a logical series of ‘if-then’ type statements is irrelevant”
noting with some satisfaction that this was in line with the approach of the European Patent Office (EPO).
Thus, ANNs are brought squarely back into the computer program exclusion – swiftly dispatching the ANN carve out in the UK IPO’s suspended AI guidelines.
When is a technical effect not a technical effect?
Perhaps the more interesting (and potentially more consequential) aspect of the original High Court judgement was the argument in the alternative, that even should an ANN be caught by the computer program exclusion, in this case there was nevertheless a technical contribution which again took the claim outside of the exclusion.
Broadly, Emotional Perception’s innovation was to be able to recommend a file of content that is semantically similar to another file of content (typically, albeit not a limitation of the claim, music data files), using intrinsic properties of the data files. This is supposedly to be achieved by manipulating a property space of an ANN such that it reflects semantic similarities. In other words, two data files that a close together in this manipulated property space should also be semantically similar. Examples of the properties would be physical properties, such as tone, timbre, speed, loudness. The semantic similarity (or otherwise) comes from how people would describe the content, e.g. happy, sad, jazz, rock, etc.
Broadly speaking the orthodox position at the EPO, and previously at the UK IPO, has been that processing data based on subjective criteria, so as to only achieve a particular effect in the mind of a user is not a technical effect, and not enough to escape the computer program exclusion. Indeed, this is the approach currently being taken by the EPO in the corresponding Euroepan patent application which is currently under examination.
At the High Court, a different view was taken. with the Court concluding that regardless of whether the semantic similarity criteria were subjective or not the
“…ANN has certainly gone about its analysis and selection in a technical way. It is not just any old file; it is a file identified as being semantically similar by the application of technical criteria which the system has worked out for itself. 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.”
The Court of Appeal has, it is fair to say, poured no small amount of cold water on this analysis. Birss LJ found that whether a file is identified or not by the system is down to
“…its semantic qualities. This is a matter of aesthetics or, in the language used by the Hearing Officer, they are subjective and cognitive in nature. They are not technical and do not turn this into a system which produces a technical effect outside the excluded subject matter.”
throwing in, for good measure, refence to an EPO Appeal case which states
“In the Board’s view, 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. While making “good” or “bad” recommendations may lead to different user reactions and thereby, in the end, to different technical results (the user might for example play more or fewer songs, or issue more or fewer search queries in order to find other songs), such results do not qualify as a technical effect of the recommendations, as they depend on subjective choices made by the user”.
Plus ça change, plus c’est la même chose…
The Court of Appeal has, in this case, clearly seen no good reason to depart from how the EPO would assess such a claim and, as a result, has mostly closed the doors to patentability previously opened by Sir Anthony Mann in his High Court judgement.
However, it remains that the assessment of the technology of the subject-matter, at least in the Court of Appeal written judgment, is quite brief and could have been written ten years ago about a “classic” computer software invention. One cannot help but notice however that the challenge ML inventions pose to the current law have not gone away.
There is an assumption running through the orthodox approach espoused in the Court of Appeal judgement (and the EPO Board of Appeal judgement cited) that there is a clear line to be drawn between on the one hand objective (technical) properties, and on the other hand subjective (non-technical) criteria. Here, it seems that, for Emotional Perception’s innovation to work, there has to be a mapping that holds between the so-called semantic space, and the property space – i.e. there needs to be some arrangement of the property space whereby the semantic (supposed subjective) properties can be divined from the (objective) measured properties.
Humans are typically good at finding patterns in data even where none exist, whereas machine learning systems’ strengths often lie in finding patterns in data where no-one thought they existed. If machines, and machine learning, become more and more adept at repeatedly and accurately determining “subjective” qualities from objective properties, then it may challenge a system that relies on a neat distinction between the two.
Where to now?
The Court of Appeal decision has again underlined a willingness from the UK courts to maintain at least a degree of harmonization between UK patent practice and that of the EPO. This may be important should future cases be brought regarding, say, simulation type inventions, an area where EPO practice has more recently diverged from establish UK patent practice. Currently applicants for UK patents benefit from a lacuna where simulation-heavy inventions seem more acceptable before the UK IPO than the EPO.
In the meantime, on the ANN front, both the Court of Appeal and the new UK IPO statutory guidance is clear in stating that ML innovations are still potentially patentable, in the same way any software innovation is, they just need to satisfy the same technicality requirements, which clearly the UK IPO and the Court of Appeal felt were not met in this case.
The Emotional Perception camp had, before the Court of Appeal judgement, already been dropping heavy hints that they would try and take the case to the Supreme Court should the judgement go against them. It remains to be seen whether such a further appeal is lodged (and whether the Supreme Court would hear the case) but this may very well not be over yet.
