As the corporate name suggests, Emotional Perception had applied for a patent about an improved way of characterising, suggesting and sending to users, recommendations for media files. Music files were of particular interest, but the patent is said to be useful for other file types too, such as video, and text files, such as academic / scientific literature.
In basic terms, the point was instead of just suggesting similar songs by reference to artist or genre, Emotional Perception’s innovation could use an artificial neural network which would be trained to suggest songs having a similar mood or feeling, such as happy or calming, by relating text-based descriptions, to physical properties, such as timbre, tone, or speed of the music.
Outcome for Emotional Perception
This patent application in the way the owners had claimed the invention, could not pass the UK legislative criteria for patentability. This is due to the exclusion from patentability of programs for computers. The computer program exclusion is interpreted by decades of case law. In practice, the exclusion does not mean a blanket ban on software patenting at all, but the invention in question has to achieve something more than just being a program (program code being protected by copyright instead of patents), it needs to achieve further technical effect.
Another legislative exclusion on patenting relates to mathematical methods – which was thought to be highly relevant in a case about artificial neural networks, but unfortunately due to some procedural oddities with how the court documents were prepared and filed, the mathematical method exclusion was not fully decided upon, although the judges and UK IPO have commented on it at each stage of the litigation so far.
The case may yet proceed to a hearing at the UK Supreme Court. Meanwhile in the Court of Appeal, given the current huge interest in AI, Birss LJ had to point out that artificial neural networks (ANNs) are by no means new. But this is the first English case required to decide on the patentability of an ANN.
What is a computer? What is a program?
A lot of the Court of Appeal hearing was devoted to these two fundamental questions, on which it may be helpful to have this clarity going forward.
Birss LJ explained that Emotional Perception contended the weights (of weights and biases) in an ANN were not a computer program.
He defined a computer as: ''a machine which processes information. Neither party came up with a better definition and I believe that is a useful on'' (para 61).
When defining a computer program, the Judge referred back to earlier cases: Gale’s Application [1991] RPC 305 ''a sequence of instructions'') and Aerotel Ltd v Telco Holdings Ltd, Macrossan’s Patent Application [2006] EWCA Civ 1371 [2007] RPC 7 (Jacob LJ: "a set of instructions").
Does it matter whether a human is involved?
In contrast to much of the argument which considered the involvement or not, of a human computer programmer, Birss LJ said ''I do not believe that referring to a human programmer is relevant or helpful. I can think of no principle which would justify that as a necessary aspect of the definition and the authorities in this area have never drawn a distinction of that kind…. There is no justification for drawing a distinction in law between instructions created by a computer and those created by a human."
What about the training stage?
Birss LJ also said that it was not relevant the machine learns for itself via a training process what the particular values are for the weights. He said that argument "is focussed on the manner in which the instructions are produced. As I have said, I do not accept there is justification for that either in principle or in the [Patents] Act (or the international conventions; EPC [European Patent Convention] or TRIPS [WTO Agreement on Trade-Related Aspects of Intellectual Property Rights]). How the program came into being is irrelevant".
He re-iterated that the training activity, of pairwise comparison between the music files’ co-ordinates in the notional spaces of the network based on semantic and physical characteristics, produces a useful system, and is part of the contribution but still, that training is effectively part of creating the program.
An artificial neural network is clearly a computer….. the weights are a program
Birss LJ continued: "Turning to an ANN, the first point to make is that however implemented, such a machine is clearly a computer – it is a machine for processing information. Focussing on the weights of an ANN, in my judgment irrespective of the manner in which an ANN is implemented (hardware or software), the Comptroller [of Patents at the UK IPO] is right that these weights are a computer program. They are a set of instructions for a computer to do something…."
The Court of Appeal brought up an EPO Technical Board of Appeal decision - Mitsubishi (T702/20), involving a neural network, and thought this also supported the Court’s approach.
Artificial neural networks are no worse (or better) off than other computer-implemented inventions
Importantly, given the pre-occupation with AI and his earlier comment in the judgment that ANNs are very much not new, not to mention how important they are in areas such as diagnostics and drug discovery, Birss LJ also emphasised the fact the computer program exclusion is engaged for an ANN implemented invention, does not make it unpatentable.
He gave examples of computer implemented inventions which have successfully been through the courts or European Patent Office, such as: controlling an X-ray machine, software for drill bit design, as well as software for multi-touch devices such as phones or tablets. He said these examples would have been just as patentable if the computer involved was an ANN.
Likewise, he said in the failed patentability cases such as Merrill Lynch’s Application [1989] RPC 561(automated stock trading methods) and Macrossan [2007] RPC 7- mentioned above (automation of company formation documents), using an ANN would not have saved them from exclusion - "ANN implemented inventions are in no better and no worse position than other computer implemented inventions."
Not a technical contribution, only a subjective / cognitive contribution
For software (or business method) patent cases, in short the basic approach the UK courts use to see if the supposed invention is software as such, or does something more by way of technical contribution to the field, is a 4-stage test (from the Aerotel/Macrossan case):
- Properly construe the claims in the patent
- Identify the actual contribution
- Ask whether that contribution falls solely within the statutory exclusions to patentability
- Check whether the contribution is actually technical
Referring back to other case law, where (to oversimplify for brevity) technical contribution could be found if the program contributed to making the computer run faster or more powerfully or better in some other way, Birss LJ also said the current situation is "not a better computer case" and pointed out that the step of sending a file recommendation message, engages the "presentation of information" exclusion to patentability (unless it can be saved by involving a technical contribution).
Birss LJ agreed with the Hearing Officer that the beneficial effect of how the system works, is subjective and cognitive in nature, using the semantic qualities of the file for the recommendation; these effects are not technical. He also referred to a YAHOO! EPO decision (T0306/10) which he summarised as saying that whether song recommendations are good or bad, does not make a technical effect.
Updated UK IPO guidance mirrors Court of Appeal on conceptual neutrality to ANNs
Whereas after the first instance court decision, the UK IPO changed its patent examination guidance to say patent applications for ANNs should not be rejected for that reason under the computer program exclusion, after the Court of Appeal decision, the UK IPO has again updated its guidance https://www.gov.uk/government/publications/examining-patent-applications-involving-artificial-neural-networks, mirroring what the Judge said- that ANNs are in no better nor worse position on patentability than other computer implemented inventions, and the Aerotel approach needs to be used for assessing whether there is a technical contribution by the computer program in question.
What happens next?
While the Court’s decision is of great interest on patentability of neural networks, it is important to remember the perspective that it was decided on the basis of the particular innovation and patent claim language used in the patent application in question. The Court has commented, but not decided on, any of the other exclusions from patentability – especially the mathematical method exclusion. On that question, the Court simply indicated that it might have traction as an exclusion in this case (and was relevant in the Mitsubishi EPO neural network case). The Judge did briefly mention the "presentation of information" patent exclusion though.
As Birss LJ said, neural network patents are in no better and no worse position than any other computer-implemented inventions. Plenty of software related patents are being granted all the time. Success in doing so is a matter of the art of patent drafting, and successfully showing that the invention in question makes a technical contribution to the field. The case has gone different ways at each turn from the Patent Office to the Court of Appeal, so there is no telling at the time of writing whether it will turn again at the Supreme Court, but the presence of the heavy technical field and patent expertise in Birss and Arnold LLJ, could mean reversing the Court of Appeal decision would be a steep climb.
A link to the full Court of Appeal decision is here: https://www.bailii.org/ew/cases/EWCA/Civ/2024/825.html
and first instance Patents Court decision: https://www.bailii.org/ew/cases/EWHC/Ch/2023/2948.html
