Thanks to the (in)famous efforts of Dr Thaler and his DABUS machine, much of the discussion in patent law has been on the interrelated questions of whether the products of AI machines can list the AI as an inventor and, critically, whether the output is patentable at all. There has been much less discussion around the patentability of the AI machine itself. The approach, in the UK, has been to rely on the usual rules of patentability.  

This laissez faire approach is now in question. The case of Emotional Perception AI Ltd v Comptroller General of Patents, Designs and Trade Marks [2023] EWHC 2948 (Ch) in the UK High Court has caused something of an earthquake to the question of AI patentability. Removing the applicability of the program for a computer exemption, the decision of Sir Antony Mann makes the UK a more welcoming place for artificial neural networks (ANN) patents than it was mere months ago.

The case, in 84 short paragraphs, will be of interest to all owners of ANNs as they consider where to patent. In removing from consideration the “program for a computer” exception to patentability, the case caused the UKIPO within weeks to issue guidance confirming that ANN will not receive an objection under the computer program exception.  

In the case itself, Emotional Perception had applied to patent an ANN which was capable of recommending to a user a media file (which for the sake of argument was considered to be a music file). After training, the ANN made this recommendation by considering the “proximity” of one music file to that of another. The “closer” the file, the more likely a user was likely to enjoy it as a recommendation.

The patent was initially refused by the UKIPO examiner on the basis that it was for a program for a computer and did not provide for a technical contribution, applying the well-established, and relatively strict, tests for considering software patents.

On appeal, Sir Antony Mann considered two questions. First, whether an ANN was a program for a computer. Second whether, if it was, there was a technical contribution.

On both counts Sir Antony found in favour of Emotional Perception. It is the decision in the former which blows open the UK’s approach to ANN patentability.

First, he concluded that neither a “hardware” ANN (effectively a black box containing the ANN) or a “software emulation” of a hardware ANN is a program for a computer. The ANN fell outside the exception. Although a human programs the parameters of the ANN, it proceeds without further intervention. An ANN was not within the exception. The appeal succeeded.

This decision is significant. It suggests that virtually no ANN will fall within the exception. It puts AI machines in a more favourable position on patentability than a traditional software implementation achieving the same result. I would question whether the drafter of the exception had this type of loophole in mind, but the court is not the place for the needed policy discussion.

On the second point Sir Antony found that as a (music) file was recommended to a user this amounted to a technical contribution. This recommendation was external to the computer and, therefore, the ANN was outside the exemption.

This element of the decision arguably builds on, albeit in an expansive manner, earlier decisions which consider similar types of output. It may therefore be of wider impact to computer programs seeking patent protection in suggesting that the production of a file to a user can be a technical contribution. Had the decision rested on this point, the earthquake would be more a rumble. It is, to some degree a fact specific question.

In contrast, by finding that an ANN, hardware or emulated, is not a program for a computer, a general principle is established, removing an otherwise significant hurdle.

The decision has unexpectedly liberated the ANN patent space without a discussion of the policy implications. Sir Antony was, however, hampered by the concession made by the UKIPO that hardware ANN was not within the exemption. This gave the court limited room for manoeuvre, regretfully recognised by Sir Antony when he noted “I think that the debate would have been interesting had the concession not been made”.

The IPO was also precluded from arguing that an ANN, at a certain level of abstraction, is a mathematical model as this had not been properly raised. It remains an open question to be taken up by another.

It is unfortunate that the IPO made the concession and on receiving the decision has immediately changed its practice. The case marks the opening of a new front in the debate around AI and patents. The UK is undoubtedly now a more friendly place for securing patent protection in an ANN. It would, however, be good to see this properly debated, from a policy perspective outside the constraints of the court. For now ANN owners should be looking at UK patents with interest and with less trepidation than before.

Update (2 January): There have been some reports that the IPO may have sought leave to appeal the decision to the Court of Appeal.  This would perhaps provide an opportunity for further detailed debate around the issues.  Whilst there are still wider policy issues to discuss, if the appeal is launched it will be one to watch with interest.

Peter Vaughan is a chartered trademark attorney and a senior lecturer at Nottingham Law School.