Artificial intelligence: English Court of Appeal decides artificial neural network is not patentable

Background

An ANN is a form of artificial intelligence (AI) structured as a collection of connected nodes, called artificial neurons, designed to model the neurons in a brain. The neurons are arranged in layers, and each neuron is connected to other neurons. Each can process inputs and then produces an output which is passed on to other neurons in other layers. The last layer produces an output from the system, which may be a decision, a recommendation or something similar.

The case concerned a patent application for a trained ANN which could provide an improved system for providing media file recommendations to users (new music with the same “feeling” as the music a user already enjoys). 

Issues

The key questions the Court of Appeal had to consider were:

• Is an ANN a “computer”?
• If it is a computer, is an ANN a program for a computer “as such”, within the meaning of section 1(2) of the Patents Act 1977 (meaning that it would be excluded from patentability unless the next bullet applies).
• If the ANN is a program for a computer, did the ANN nevertheless fall outside the computer program exclusion (and so be patentable) because it had a “technical contribution” outside itself?

Computer program

Lord Justice Birss (with whose judgment Arnold LJ and Nicola Davies LJ agreed) observed that:

• A “computer” is a machine which processes information.
• A “computer program” is a set of instructions for a computer to process information in a particular way.

Applied to the facts, Birss LJ found that:

• An ANN is a computer – it is a machine for processing information.
• Its “weights and biases” (effectively internal correlations between data points in the neural network, configured during training and determining the output of the processing) are a computer program, as they are a set of instructions for a computer to process information.

While the High Court held that a hardware ANN was not a computer (as opposed to a software ANN which it held to be a computer without a program), Birss LJ considered both versions to be clear examples of a computer, each having the same implementation of “weight and biases”, which Birss LJ found was a computer program. Accordingly, the distinction between a hardware and software ANN was not relevant to the question of patentability before the Court of Appeal.

Technical contribution

Birss LJ held that:

• The training of the ANN was part of the creation of the program and subsidiary to the patent claim, and therefore was not a technical contribution.
• The outputs of the ANN (its music recommendations) were not a technical contribution. The improved recommendations that the system could produce were a matter of aesthetics. They were subjective and cognitive (in the hands of the user) in nature. They were not technical and did not turn the ANN into a system which produced a technical effect outside the excluded subject matter.

Key takeaways

The judgment does not mean that an ANN-implemented invention can never be the subject of patent protection. It simply means that ANN-implemented inventions are in no better or worse position than other computer-implemented inventions under English law. Many computer-implemented inventions are outside the computer program “as such” exclusion, and may be patentable as a result.

As at the date of publication, the IPO’s Guidelines for examining patent applications relating to AI have been temporarily suspended pending consideration of the Court of Appeal judgment in the Emotional Perception case (the Guidelines included scenarios that were amended to align with the High Court Emotional Perception judgment, which had held that the ANN was patentable). Any update is likely to reflect the Court of Appeal’s view that an ANN is a computer program. 

The Court of Appeal made clear that training a neural network was part of the creation of the program and therefore was not a technical contribution. That finding is of particular relevance to generative AI systems, where training is a fundamental part of system development. Patent claims relating to generative AI will need to take account of this view.

Moreover, the Court, in focussing its assessment on whether the outputs from the ANN were technical or non-technical in nature, made clear that non-technical outputs an ANN produces are not relevant in establishing patentability of an AI computer program in the same way they are not relevant when assessing any other computer implemented invention.