There’s no point getting emotional about Emotional Perception

In December, Emotional Perception was granted permission to appeal to the Supreme Court. This follows the initial refusal of its patent application for an AI-based music recommendation system by the UK Intellectual Property Office (UKIPO) in June 2022. The case will now be heard by the highest court in the land, after the refusal was overturned by the High Court and reinstated by the Court of Appeal. 

Background

The patent application was refused on the grounds of being unpatentable subject matter under Section 1(2) of the Patents Act 1977, which states that programs for computers are unpatentable, but the exclusion only applies ‘to the extent that a patent [...] relates to that thing as such’.  

The world of AI has moved on significantly in the last 15 years, but the case law on software patentability has not. The current position (see Aerotel Ltd v Telco Holdings Ltd [2007] RPC 7 and AT&T Knowledge Ventures v Comptroller [2009] EWHC 343 (Pat)) is that only a relatively narrow band of AI inventions can be patented, namely those that provide a ‘technical contribution’. What makes a ‘technical contribution’ is not defined in the Act and it is narrowly construed by the UKIPO, limiting patentability to software that has some effect outside the computer (e.g., controlling machines in the real world or using sensor data) or that is tied to computer architecture at a relatively low level (e.g., embedded systems). 

You would hope that this case would allow the court to consider whether the software patent status quo is fit for purpose in an increasingly digital age. You might also hope that it would open up the possibility of protecting a wider range of AI innovations and perhaps aligning with the UK government’s recently stated desire to ‘turbocharge’ AI. However, we are highly sceptical that anything so positive will come from this decision, based on the way the arguments that have unfolded in the lower courts. 

To understand why, first some brief background. The application relates to an artificial neural network (ANN), a common type of AI model inspired by the brain. It has layers of interconnected nodes that process data and each node is associated with weights, which are applied to the inputs to the node to produce its output. The weights are learned by training the model with data, so as to provide a desired output when provided with an input. 

Emotional Perception’s ANN is trained to recommend music tracks. The ANN processes the track and gives a numerical output (‘embedding’) that can be compared to embeddings derived from text descriptions of music files to determine their similarity. This is particularly useful where a new track without text metadata needs to be used to search very large music databases for similar tracks. It is also a good example of the ‘cold start’ problem, a common engineering problem in AI systems, which need to provide output before sufficient data is available. Emotional Perception’s application provides a clever ‘trick’ (in the words of Birss LJ, giving the leading judgment) to solve it by training the ANN to provide output similar to the model that generates the text embeddings.