Emotional Perception 2024

Emotional Perception's patent application for an AI music recommendation neural network was refused by UKIPO as a computer program "as such", and therefore excluded from patentability.  The High Court overturned this decision in 2023.  But the UK Court of Appeal has reversed it again, endorsing the view of the UKIPO Hearing Officer.

 

The Bad

In the earlier proceedings, the question of "hardware" vs "software" implementations of an artificial neural network (ANN) was prominent.  However, Lord Justice Birss shifted the focus to the weights of the neural network, concluding that, regardless of the underlying ANN implementation, the weights learned in training were a "computer program" (para. 68).

 

He then considered the effect of the computer program (i.e., the weights) when 'run' on the ANN, concluding that the recommendation of a file based on its semantic properties was "a matter of aesthetics or, in the language used by the Hearing Officer, … subjective and cognitive in nature".  In this respect, the judgement aligned with the EPO's approach that also characterises recommendations as subjective or cognitive.

 

This is a setback for protecting certain categories of AI inventions in the UK, but not all hope is lost.

 

The Good

Earlier we noted the attention given throughout the Emotional Perception proceedings to the question of a "hardware" vs "software" artificial neural network (ANN).  We focused previously on the Court of Appeal's characterisation of the weights of a neural network as a "computer program".

 

Paragraph 68 contains a notable conclusion: "Turning to an ANN, the first point to make is that however it is implemented [hardware or software] such a machine is clearly a computer – it is a machine for processing information" (emphasis added).  The Court of Appeal also seemed to accept Emotional Perception's characterization of a software ANN as "a software emulation of [a hardware] ANN" (para. 16).

 

This is interesting because it is not the characterization adopted by the EPO.  The EPO regards neural networks and other machine learning models as "computational models and algorithms [that] are per se of an abstract mathematical nature, irrespective of whether they can be 'trained' based on training data".  This was the position of the EPO Technical Board of Appeal in T 702/20 (Sparsely connected neural network/MITSUBISHI) 07-11-2022, cited in para. 69 of the Court of Appeal Judgement.

 

Although it did not save Emotional Perception’s invention, this characterization of an ANN raises the possibility that, following the UK Court of Appeal judgement, certain improvements concerning a neural network could be characterized as an improved computer rather than an improved computer program, even if the neural network is implemented in software (for example, improvements to the underlying architecture of the neural network).

 

This is promising in some respects, but unfortunately significant ambiguity remains following the decision.

 

The Ugly

For procedural reasons we have noted previously the exclusion on protecting "mathematical methods" as such received little attention.  This is unfortunate, as the EPO primarily examines the patent eligibility of AI inventions under the mathematical methods exclusion.

 

We previously discussed above how the UK Court of Appeal characterised the weights of a neural network as a ‘computer program’ run on an artificial neural network (the underlying ‘computer’).

 

Having concluded that Emotional Perception's weights were a computer program that failed to produce a technical effect when ‘run’, the court did not need to consider the mathematical methods exclusion, noting only (in para. 84) that

 

"I will only say that I think this objection might well have had traction if the conclusion was that weights and biases of an ANN were not a computer program. It is hard to see why even if they are not to be regarded as a computer program for some reason, they are not in any case a mathematical method and so the very same analysis … would apply with the same result."

 

The ugliness arises in the ambiguity concerning the model itself (the neural network).  On the one hand, the UK Court of Appeal categorizes it as "clearly" a computer even when implemented in software (para 68), and a computer (as opposed to a computer program) is plainly technical. On the other hand, whilst para. 84 of the Court of Appeal judgement appears to characterise the weights as a mathematical method, it does not directly answer the question of whether the model itself could be characterised as a mathematical method.  Going forward, it is unclear what position the UKIPO will take on this point in examination, and whether they will diverge or align with the EPO.

 

So, where does this leave us? 

 

As ever in the rollercoaster of software patents in the UK and Europe…more or less back where we started.  But don't be disheartened…our specialist and growing AI team has been supporting AI innovators in the UK, Europe and worldwide for the better part of a decade, and we are ideally placed to navigate this complexity for you.  

 

Contact Tom Woodhouse for supporting in protecting your AI innovations.

 

This briefing is for general information purposes only and should not be used as a substitute for legal advice relating to your particular circumstances. We can discuss specific issues and facts on an individual basis. Please note that the law may have changed since the day this was first published in July 2024.