Are “better” neural networks patentable in the UK?

Following on from the Court of Appeal’s judgment in Comptroller-General of Patents, Designs and Trade Marks v Emotional Perception AI Ltd [2024] EWCA Civ 825 (Emotional Perception) (covered here), the UKIPO has published a brief statutory guidance note indicating how the judgment will affect examination of patent applications involving artificial neural networks (ANNs).

The bulk of the guidance note provides a summary of the judgment and as expected, confirms that “patent examiners should treat ANN-implemented inventions like any other computer implemented invention. Examiners should use the Aerotel approach to assess whether the claimed invention makes a contribution, which is technical in nature.”

The reasoning in Emotional Perception indicates that one way in which an ANN-implemented invention can make a technical contribution, and thus escape exclusion from patentability, is where its function is technical in nature; this appears in practice to be similar to the manner in which the European Patent Office (EPO) has considered ANN inventions directed towards a particular technical application. However, the EPO’s approach has tended to be more restrictive where the innovation lies in the nature of the ANN itself, rather than being tied to a particular use case.

With this in mind, one intriguing aspect of the Court of Appeal’s judgment is the manner in which a distinction is drawn between an ANN itself and the weights and biases of the ANN when trained. The Court of Appeal is clear that these latter parameters should be understood as defining a “program for a computer” but is equally clear that the ANN itself, whether implemented in hardware or software, is to be understood as a “computer” in its own right. As noted in our previous analysis, the judgment recognises that in the right cases, a technical contribution could arise where the invention relates to a “better computer”. Since the decision considers an ANN to be a computer, it follows that, in the right cases, a technical contribution may also potentially arise in the case of a “better ANN”.

This possibility is officially recognised in the guidance note, which states that “[a]s with any other computer implemented invention, the AT&T signposts may be a helpful aid for examiners in considering whether an ANN-implemented invention makes a technical contribution. Examiners should apply the signposts carefully, particularly in so-called “better computer” (i.e. better ANN) cases where the invention is said to improve the functioning of the computer (i.e. the ANN) itself.”

The exact conditions under which an ANN will be considered a “better ANN” which improves the functioning of the ANN itself are, at present, unclear. At the EPO, neural networks and their structures are considered to be of an abstract mathematical nature, but such structures can in certain cases be seen as contributing to technical character, for example if their design is motivated by technical considerations relating to the internal functioning of the underlying hardware that the neural network is implemented on. On the other hand, the guidance note from the UKIPO makes clear that in the UK, the question of whether there is a “better ANN” will be determined based on the AT&T signposts.

For those who are unfamiliar, the AT&T signposts are a set of guiding questions that can be used to help in determining whether an invention provides a technical contribution and is therefore not excluded from patentability. The first four signposts read:

(i) whether the claimed technical effect has a technical effect on a process which is carried on outside the computer;

(ii) whether the claimed technical effect operates at the level of the architecture of the computer; that is to say whether the effect is produced irrespective of the data being processed or the applications being run;

(iii) whether the claimed technical effect results in the computer being made to operate in a new way;

(iv) whether the program makes the computer a better computer in the sense of running more efficiently and effectively as a computer.

The UKIPO has confirmed that these signposts are applicable to ANN-implemented inventions, whilst also highlighting the finding by the Court of Appeal that an ANN is a computer. Thus, for example, there now seems to be the possibility that signpost (ii) could be met for an invention that relates to an improvement in the ANN itself (e.g. the architecture of the ANN) that is independent of the data that is processed by the ANN and the application being run (i.e. the weights and biases forming the ANN computer program).

The Court of Appeal decision in Emotional Perception acknowledged that “better computer” cases are “often difficult ones”. The situation is more complex still in the case of a “better ANN”. The AT&T signposts and the existing case law on their interpretation are all based upon inventions using conventional computers. Given the differences between ANNs and conventional computers, there are many questions around the extent to which current principles can be mapped to ANNs. More detailed guidance is expected to be issued by the UKIPO in due course.

However, at present, we think there is a hint of an opening of a new path forward for inventions relating to ANNs in the UK, that is not available at the EPO. Thus, whilst the Court of Appeal’s ruling tended towards re-alignment with EPO practice in comparison with the High Court judgment it overturned, the alignment is not necessarily absolute: some divergence may yet still exist, potentially providing applicants with more opportunity to protect core AI innovations in the UK. Watch this space.