In a recent High Court decision, Justice Mann overturned the UK Intellectual Property Office's (UKIPO) refusal to grant a patent for an artificial intelligence system utilising artificial neural networks (ANNs) to recommend similar songs to users[1].
The invention developed by Emotional Perception AI Ltd leverages a deep learning model based on ANNs. ANNs comprise layers of interconnected nodes like neurons in a brain, which process and transmit data signals. By training the ANN model on numerous song pairs labelled with both musical properties and semantic descriptions, it learns to map songs with proximate vectors in musical property space to songs with similar vectors in semantic space. Once trained, the ANN model can analyse an input song's properties, generate a corresponding semantic vector, and find the songs with the closest vectors in its database to recommend.
Basis of the appeal?
The UKIPO initially refused the patent application under section 1(2)(c) of the Patents Act 1977, asserting that the invention constituted a computer program 'as such.' The Hearing Officer went on to argue that the innovation's contribution amounted to nothing more than a routine computer programming activity. Moreover, the recommendation of semantically similar files to users was deemed to lack any "technical effect beyond the ordinary execution of a program on a computer."
Emotional Perception AI contested this decision on two grounds: first, that the computer program exclusion was not applicable, and second, that even if considered a computer program, the invention contributed technically and should not be deemed a computer program 'as such.'
High Court’s Ruling
On appeal, Justice Mann determined that the subject matter of the invention did not trigger the statutory exclusions in any way. This was because the Artificial Neural Network (ANN) was considered not to be a computer program, irrespective of whether it was implemented in hardware or emulated in software on a computer. The crux of this conclusion rested on the fact that the ANN did not execute a set of instructions crafted by a human; instead, it functioned based on its independently acquired knowledge. In the case of software emulation, the judge found that the ANN operated on a different level compared to the underlying computer software. While the emulation of the ANN's architecture (including its nodes and layers) resulted from programming efforts, the actual functioning of these elements was autonomously generated by the ANN itself, without direct human input.
Further, the judge held that the invention was not actually a computer program, emphasising a crucial distinction between the software used for initial model training and the autonomous functioning of the trained model. Specifically, he ruled that the invention's focus was on the specific training approach using song pairs, not the ancillary software. Notably, the trained ANN operated independently, applying its learned weights and biases without reliance on human instructions.
Assessing Technical Effect
Addressing the hypothetical scenario where the invention could be deemed a computer program, Justice Mann assessed whether it provided a 'technical effect' beyond the confines of the software. Citing precedents such as Protecting Kids[2], he determined that the notification of a music recommendation file to a user constituted a technical effect, despite the subjective nature of music recommendations tied to user enjoyment.
In Protecting Kids, similar logic was applied to the identification of inappropriate content and its notification to an adult, deeming it a technical effect. Justice Mann concurred with Emotional Perception that, akin to Protecting Kids, there was an external effect through the music recommendation file provided to users.
Contrary to the earlier decision by the UKIPO’s Hearing Officer, he dismissed the argument that the user's subjective experience rendered any effect non-technical, emphasising that the artificial neural network's output of semantically similar songs to users had tangible, objective consequences.
Significance and Key Learnings
The decision will be welcomed by those looking to obtain patent protection in the UK for AI inventions involving ANNs and we could see the UKIPO being a more favourable environment for the grant of patents relating to these types of inventions. However, there are a couple of caveats to bear in mind:
- the assessment of technical contribution in this case focused on whether there was a technical effect outside the computer (one of the “signposts” from Re AT&T Knowledge Ventures LP [2009] EWHC 343 (Pat), which is a key factor indicating that claimed subject matter is outside statutory exclusions). The judge considered that there was such a technical effect outside the computer, because the claims specify the step of providing the identified file to the external user. If the claims had only involved identifying a file using the ANN without the additional step, the outcome might have been different. Therefore, where relevant to the invention, it is important to include in the claims a step of providing an output. A similar rationale was applied in the Protecting Kids case, where the court found that the claimed subject matter did not pertain to excluded subject matter.
- the UKIPO has put the updating of its examination guidelines on hold, pending this decision. In the past, when there have been court decisions in the field of computer-implemented inventions that have been favourable to patentees, the UKIPO has not updated its guidelines in a manner that is as broad as such favourable decisions. This has led to the UKIPO applying the criteria for patent eligibility rigorously. Therefore, the impact of this recent decision on actual practice by the UKIPO remains to be observed. Additionally, the UKIPO may possibly appeal to the Court of Appeal, which potentially could result in the overturning of the current judgment. In such a scenario, the existing state of affairs would persist. Conversely, if the Court of Appeal upholds the judgment, it could signal the commencement of a more AI-friendly era at the UK Intellectual Property Office (UKIPO).
In summary, there is reason for measured enthusiasm, for the moment, as this decision holds the potential to significantly enhance the likelihood of success for patent applications concerning AI at the UK Intellectual Property Office (UKIPO).
[1] Emotional Perception AI Ltd v Comptroller-General of Patents, Designs and Trade Marks [2023] EWHC 2948
[2] Protecting Kids The World Over (PKTWO) Ltd [2011] EWHC 2710