Has the High Court given the green light to AI inventions?
A recent development in UK case law may prove pivotal to the success of patent applications in the burgeoning field of Artificial Intelligence (AI). The judgment handed down by Sir Anthony Mann on 21 November 2023 concerns an appeal at the High Court of a UK Intellectual Property Office (UKIPO) Hearing Officer decision (BL/O/542/22). In Emotional Perception AI Ltd v Comptroller-General of Patents, Designs and Trade Marks, the High Court decided to overturn the UKIPO’s application of the exclusion from patentability for “a program for a computer… as such” under section 1(2)(c) of The Patents Act 1977.
The Invention
In their patent application, Emotional Perception AI Ltd claim to have developed an Artificial Neural Network (ANN) based system for providing media file recommendations to an end user, a system which they say can be used to offer suggestions of similar music in terms of human perception and emotion.
The system is trained in a multi-step process. Pairs of music files are taken, each accompanied with natural language “semantic” descriptions, and a first ANN analyses the descriptions and produces vectors in semantic space, such that semantically similar music files have closer coordinates. A second ANN analyses the music files for pre-set physical properties such as tone, timbre, and speed, producing another set of vectors in a physical property space. The second ANN is then trained through back-propagation to converge or diverge the physical property space vectors for the music files so that they more closely reflect their semantic space vector counterparts. For example, if the semantic space vectors in the pair of music files are closer than their corresponding physical property space vectors then the second ANN is re-trained such that the physical property space vectors are closer together, and vice-versa. Then, after the system has been trained, the configuration (weights and biases) of the second ANN is fixed and can be used to devise semantically similar music to any given track and make a recommendation of a similar music track by sending a message and a file to a remote user.
Basis of the Appeal
The Hearing Officer considered whether the subject-matter defined by independent method claim 4 was excluded using the four-stage approach in Aerotel Ltd v Telco Holdings Ltd [2007] RPC 7, and then assessed the technical contribution of the claim with reference to the five signposts from AT&T Knowledge Venture v Comptroller of Patents [2009] FSR 19. The Hearing Officer concluded that the actual contribution involves a specific approach for training an ANN: “…the programmer defines the problem and the training approach, and the ANN operates within those boundaries to build a suitable model…”, making the invention no more than a computer program and thus excluded from patentability.
The questions for Mann to consider were along two lines of thought:
a) Is the exclusion for “a program for a computer… as such” engaged?
b) If the exclusion is engaged, is there nevertheless a technical contribution?
The Decision
Mann started by addressing two fundamental questions: what is the computer, and where is the program? By considering a hardware ANN first – a piece of hardware containing the nodes and layers of the ANN – Mann concluded that the processing of data qualified the hardware ANN to be a computer itself. A software emulated ANN necessarily involves a computer to facilitate the emulation. Mann then departed from the interpretation of the UKIPO by decoupling the computer program used to implement the emulated ANN from the computer program used to train the ANN. Having established, in agreement with the UKIPO’s position, that the hardware ANN does not involve a program, Mann explained that the emulated ANN should also be considered to operate in the same way (albeit facilitated by underlying software). Thus, the only program in the claim – during the training stage – is a subsidiary part of the claim to the key feature of using pairs of files for training and setting the training objective and parameters accordingly. Mann therefore concluded that the claim is not directed towards a computer program at all, so the computer program exclusion in not invoked.
Mann also considered the patentability of the application if the computer program exclusion was invoked. By explaining how providing a file constitutes an external technical effect that can lift the ANN out of the exclusion on patentability of computer programs and further by suggesting that a trained ANN is capable of providing a technical effect in itself, Mann has offered a strong word of support for the patentability of ANN-based inventions.
This decision is a departure from the approach previously taken by examiners at the UKIPO in assessing the patentability of ANNs, making the UK perceptibly friendlier towards ANN-implemented inventions. However, it should be noted that Mann did not consider whether training or implementation of an ANN might fall under the exclusion for a mathematical method provided by Section 1(2)(a) of the Act. Regardless, this decision may provide a new, more ANN-positive direction for the UKIPO than current guidelines of the European Patent Office (EPO) appear to permit.
Conclusion
The judgment is encouraging for anyone seeking protection for an AI-implemented invention, particularly those relating to Artificial Neural Networks (ANNs) and will help persuade Examiners at the UKIPO to refrain from raising an excluded matter objection for trained ANNs. However, the impact of the difference between the approach traditionally taken by the UKIPO and the decision of Mann in assessing the patentability of ANNs is yet to be seen. It is possible that the UKIPO will appeal to the Court of Appeal, where the judgment could be overturned. Whilst the UKIPO considers the decision, they have temporarily suspended their guidelines for examining patent applications relating to AI inventions published in September 2022. Nevertheless, and at least for now, this decision appears to measurably improve the chance of a successful outcome for UK patent applications involving ANN-based inventions.
For further information on this ground-breaking High Court decision and its implications for AI patentability, please feel free to reach out to us at docketing@secerna.co.uk. Our team of experts at Secerna LLP is ready to assist and provide insights into the evolving landscape of AI-related inventions and patent applications. We welcome your inquiries and look forward to engaging in meaningful discussions to support your intellectual property needs.