AI Implemented Invention for Making Music Recommenations Not Patentable

On July 19, 2024, the U.K. Court of Appeal rejected a patent application by Emotional Perception AI Limited (EPL), essentially because the technical part (the AI) was held to be a program for a computer and there was no external technical effect. The judgment (available here) therefore follows and echoes many of the earlier cases on the patentability of computer implemented inventions.

In summary, the invention, the subject of the patent application, provided media file recommendations. For example, on a music website it was able to provide recommendations of other music similar to the one being listened to. It was said to be better than other similar products as it was able to offer suggestions based on similar music in terms on perception and emotion irrespective of the genre of the music. This output is arrived at by passing music through a trained artificial neural network (ANN). That the ANN was trained is a key component of the analysis of Birss LJ who gave the leading judgment of the Court of Appeal.

Under English patent law, “a program for a computer … as such” is excluded from patentability (section 1(2) Patents Act 1977). A like exclusion applies in Europe and many other jurisdictions. The law, however, has developed so that an invention is not just automatically excluded from patentability just because it is implemented in software. The so-called Aerotel approach (from the case Aerotel Ltd v Telco Holdings Ltd [2007] RPC 7) provide a convenient four-step way of assessing whether a computer implemented invention is excluded from patentability. They are:

1. Properly construe the claim. 
2. Identify the actual contribution (although at the application stage this might have to be the alleged contribution). 
3. Ask whether it falls solely within the excluded matter. 
4. If the third step has not covered it, check whether the actual or alleged contribution is actually technical.

As Birss LJ records:

“The [Aerotel] approach can be summarised loosely as being to work out if the claimed invention makes a contribution which is technical in nature.”

At the U.K. Intellectual Property Office (UKIPO), EPL’s patent application was rejected. This was reversed on appeal to the High Court, from where the question was further appealed to the Court of Appeal.

In summary, the Court of Appeal held that the ANN was highly technical in nature and characteristic, but so too are ‘ordinary’ computers – that does not per se give a technical characteristic to a computer program running on them. Birss LJ held that the training of the ANN was the same in concept as the programming of a computer. The patentable invention could not, therefore, arise in the ANN or its set up, and one had to look at the output and consider whether that was technical in nature.

There is a technical output, namely the transmission of a file on a network. This file, however, only differs from the prior art because the file represents a better recommendation. This difference was held by Birss LJ (and by the UKIPO) as being subjective and cognitive in nature, and not technical in nature. As one had, therefore, something which may be new but is a program for a computer producing something which is not technical in nature the patent application fell foul of the exclusions from patentability.

Does this judgment sound a death knell for patenting AI implemented inventions? Not at all. Like many traditional software implemented inventions before it, this one failed because the output did not have the necessary technical characteristic. Had the output had such a technical characteristic it would not have failed. The judgment does, however, set out clear law that training an ANN will be regarded as “a program for a computer.”