The Emotional (Perception) Rollercoaster Continues: Supreme Court agrees to hear an appeal on Patentability of AI Inventions

United Kingdom

In July 2024, the UK Court of Appeal handed down its judgment in Emotional Perception v Comptroller, ruling that Emotional Perception’s deep learning AI recommender system was a computer program “as such” and therefore excluded from patent protection. The decision not only reversed Sir Anthony Mann’s pro-AI High Court judgment but also went further than the UKIPO’s ruling by holding that hardware artificial neural networks (ANNs) constitute computer programs (something previously applicable only to software emulations).  The UK Supreme Court has now agreed to take up the case for review.  The UK Supreme Court is expected to hear the case next year and the result will likely have wide reaching implications for patentability of computer implemented inventions in the UK.

Nature of the invention

The invention (to recap our previous article (here)) involves a deep learning AI system designed to recommend similar songs based on both semantic and musical properties. The system utilises two ANNs: one for semantic properties (e.g., mood descriptors like happy or sad) and one for musical properties (e.g., tone, timbre, beats per minute). The training data comprises pairs of music files, each labelled with either semantic or musical properties, enabling the ANNs to learn to cluster similar songs together. The innovation lies in combining the results of the two ANNs, adjusting the musical similarity scores based on semantic similarities, and vice versa. The final ANN integrates both semantic and musical similarities to provide song recommendations. The system can be implemented in either a hardware form (hardware ANN) or through software emulation (software ANN).

Grounds 1 and 2: computer programs ‘as such’

The first ground of appeal argued that the High Court erred in holding that the exclusion for "a program for a computer ... as such" was not engaged. The second ground argued that the High Court had wrongly relied on a “concession” from the Comptroller that a hardware ANN was not a computer program.

Is an ANN a computer program?

Birss LJ, giving the Court of Appeal’s unanimous judgment, gave broad interpretations to “computer” and “computer program”, holding that the term “computer” is a “machine which processes information” and a “computer program” is a “set of instructions for a computer to do something”. On this basis, it held that (a) an ANN qualifies as a computer because it processes information and (b) the set of weights in an ANN, which determine how information is processed, are a set of instructions for the computer, thus constituting a computer program. The Court found this consistent with precedents like Gale’s Application and Aerotel, which described a computer program as a set of instructions for processing information. It rejected various arguments from Emotional Perception, including that a computer program requires a set of if-then statements and the involvement of a human. Birss LJ  regarded these as irrelevant and unpersuasive.

Any distinction between hardware-implemented and software-implemented ANNs?

On the second ground, the Court held that the distinction between hardware and software implementations of ANNs does not alter the fact that the weights and biases constitute a program for a computer. There was no difference therefore between hardware and software ANNs. This constitutes a remarkable turnaround in result for the Comptroller given that, in the High Court, Sir Anthony Mann clearly understood (whether rightly or wrongly) the Comptroller’s position to be that the computer program exclusion was not engaged at all for hardware ANNs.

Ground 3: the mathematical method exclusion

While this ground was not considered by the Court of Appeal (given its ruling on grounds 1 and 2), Birss LJ did note that such an argument “might well have had traction if the conclusion was that weights and biases of an ANN were not a computer program”.

Ground 4: substantive technical contribution

The fourth ground dealt with whether the claimed invention involved a substantive technical contribution. Applying the Aerotel four-step test, the High Court had ruled (obiter) that the ANN’s ability to recommend songs based on both semantic and musical properties constituted a technical effect, thus making it patentable.

The Court of Appeal disagreed, siding with the Comptroller that the beneficial effect of the file recommendations based on semantic similarity was of a subjective and cognitive nature, not a technical one. The judges referenced the decision in Yahoo by the EPO, which held that the quality of song recommendations does not amount to a technical effect. The Court concluded that the ANN’s method of processing data to provide song recommendations, while inevitably technical given that it involved a computer, ultimately produced a subjective and cognitive result, not a technical contribution. Accordingly, the invention was not patentable because it did not provide a technical effect outside the excluded subject matter.

CMS comment

The UK Court of Appeal's decision was another change in the direction of the law around AI patentability. By reinstating (and expanding on) the UKIPO’s original ruling, the Court has returned to a situation similar to that of the European Patent Office regarding what qualifies as a "computer program as such" under s.1(2) of the Patents Act 1977. Decisions of the Technical Boards of Appeal of the EPO, such as T702/20 Mitsubishi/Sparsely connected neural network, are mentioned as “applying exactly the same approach”. 

Some argue that the Court of Appeal decision has gone even further than the UKIPO by including hardware-implemented ANNs into the “computer program as such” exclusion.  Paragraph 70 of the decision notes that “Nor is there any difference for this purpose [engaging the computer program exclusion] between a hardware ANN and a software ANN.  However, it is implemented, the weights (by which I mean weights and biases) of the ANN are a program of a computer and therefore within the purview of the exclusion”.  However, this statement is limited to the weights and biases of the ANN.  Other aspects such as particular analog or digital circuitry needed to implement a novel type of ANN are arguably still outside the computer program exclusion.

The decision introduced a need for an immediate change to practice for the examination of ANNs, where patent examiners should treat ANN-implemented inventions like any other computer implemented invention for the purposes of s.1(2).  The AI patent examination guidance from the UK IPO, which included scenarios with draft claims, were suspended pending consideration of the outcome of the Court of Appeal’s decision, although a practice note was issued stating that Examiners should apply the Aerotel approach.  A strong emphasis, however, is made in the judgment that ANN-implemented inventions may be patentable in some cases; it simply means there will be an application of the Aerotel approach to assess technical contribution along with the aid of the AT&T signposts. So, for applicants, this decision highlights the need to show a clear technical contribution beyond the computational methods used.

Now that the UK Supreme Court has granted Emotional Perception AI Ltd permission to appeal there is good opportunity for the law to develop around patentability of fast moving computer implemented technologies such as artificial intelligence and machine learning.