Incorporating UKIPO Guidance Into AI Patent Strategies

On Jan. 30, the U.K. Intellectual Property Office published updated guidance for examining patent applications relating to artificial intelligence following the Court of Appeal judgment last year in Comptroller-General of Patents, Designs and Trademarks v. Emotional Perception AI Ltd.

This decision overturned an earlier ruling by the High Court that an artificial neural network, or ANN, including the one underlying Emotional Perception's invention, should not be excluded from patentability under the computer program exemption.

The guidelines have been updated to account for the Court of Appeal's alternative determination that an ANN is fundamentally a program for a computer and therefore may be subject to the computer program exemption.[1]

The U.K. Supreme Court has granted permission for Emotional Perception to appeal the Court of Appeal's decision with a hearing expected later in 2025. The guidelines may therefore be subject to further amendment, depending on the Supreme Court's final decision.[2] In the meantime, a closer look at Emotional Perception and the updated guidance can inform practitioners' drafting strategies for patenting AI inventions.

The Court of Appeal's Decision in Emotional Perception

A patent may be granted for an invention provided that it is not excluded from patentability. Exclusions from patentability are defined in the Patents Act 1977 and include mathematical methods and programs for a computer.[3]

The meaning of "a program for a computer" was considered in detail in Emotional Perception, a case which concerned the provision of media file recommendations, typically music tracks, to a user by implementing a trained ANN.

In deciding the appeal, the Court of Appeal firstly needed to determine what a computer program is. On this point it was found that a computer is a machine which processes information in a particular way, and a computer program is a set of instructions which cause the machine to process the information in that way.[4]

The Court of Appeal then sought to answer whether there is a computer program in an ANN. On this point it was held that, irrespective of how an ANN is implemented, the weights and biases of the ANN constitute a computer program as they act to instruct the machine to process information in a particular way.[5]

The Court of Appeal then applied these findings to the specific facts in Emotional Perception and, in doing so, found the ANN to be excluded from patentability as a program for a computer. The Court of Appeal, however, emphasized that this does not mean that all ANN-implemented inventions are unpatentable. To the contrary, the patentability of an ANN-implemented invention is to be treated like any computer implemented invention.[6]

The Test for Computer Implemented Inventions

A computer implemented invention is not excluded under the act if it makes a technical contribution to the state of the art. This is determined by applying the four-step test established in the Court of Appeal's 2006 ruling in Aerotel Ltd. v. Telco Holdings Ltd., which requires one to:

1. properly construe the claim
2. identify the actual contribution;
3. ask whether it falls solely within the excluded subject matter;
4. check whether the actual or alleged contribution is actually technical in nature.[7]

A set of signposts are also regularly applied by U.K. courts to specifically analyze whether a claimed invention has a technical effect. These signposts require one to consider:

• Whether the claimed technical effect has a technical effect on a process which is carried on outside the computer;
• Whether the claimed technical effect operates at the level of the architecture of the computer; that is to say whether the effect is produced irrespective of the data being processed or the application being run;
• Whether the claimed technical effect results in the computer being made to operate in a new way;
• Whether the program makes the computer a better computer in the sense of running more efficiently and effectively as a computer; and
• Whether the perceived problem is overcome by the claimed invention as opposed to merely being circumvented.[8]

The updated guidelines address each of these signposts as they relate to AI inventions.

Updated Guidance

Applied AI

The first signpost asks one to consider whether the AI invention produces an external technical effect. The new guidelines note that this may be evident in two different ways.[9]

One option is for the AI invention to perform or control a technical process which exists outside the computer on which an invention is implemented.

The guidelines cite Vicom in relation to this option, a 1986 case before the EPO technical board of appeal which concerned producing image convolutions with an increase in processing speed compared to known methods when executed on a conventional computer. This was found to make a technical contribution because it performed a specific technical image processing process lying outside the computer itself.[10]

The other option is for the AI invention to contribute to the solution of a technical problem lying outside the computer on which the invention is implemented.

The guidelines cite Halliburton in relation to this option, a 2011 case before the High Court which concerned modelling the design of drill bits to improve their drilling efficiency and operational life. This was found to make a technical contribution because designing drill bits is a technical process involving considerations about wear and rock cutting capability. Therefore, finding a better way of designing drill bits was deemed to constitute a general technical problem.[11]

However, the guidelines caution that not every process lying outside a computer will reveal a technical contribution. For example, in Emotional Perception the recommendation message was held to be non-technical because it was based on semantic qualities. These qualities were said to be subjective and cognitive in nature and thus not technical according to the Court of Appeal.[12]

Further, an effect which amounts to no more than one produced by merely running a program cannot count as a technical contribution. A patentable AI invention must therefore make a technical contribution beyond the mere manipulation, analysis, or the generation of textual data.[13]

Additional Signposts

The second, third and fourth signposts collectively ask whether the AI invention results in a "better computer" because the computer which implements the invention generates an internal technical effect. The guidelines note that this may be evident in two different ways.[14]

One option is for the AI invention to result in a better computer because the invention solves a technical problem lying within the computer.

The guidelines cite Symbian Ltd v Comptroller General of Patents in relation to this option, a 2008 case before the Court of Appeal which concerned the programming of a dynamic linked library for storing common functions. This was found to solve a technical problem lying within the computer because it had the resultant effect of the computer working better as a matter of practical reality.[15]

The other option is for the AI invention to result in a better computer because the invention defines a new way of operating the computer in a technical sense.

The guidelines cite Gale's Application in relation to this option, a 1991 case before the Court of Appeal that concerned a new way of calculating square roots which was sympathetic to the operation of the underlying architecture of the computer. This was found not to define a new way of operating a computer because the instructions did no more than prescribe a different set of calculations from those normally prescribed. The guidelines therefore emphasise that to meet the third signpost the AI invention must result in a new way of operating the computer in a relevant technical sense.[16]

Core AI

In contrast to AI inventions dealing with technical problems external to, or lying within, a computer, the guidelines also address so-called core AI inventions. These inventions involve advances made within the model, algorithm or mathematical method which forms the AI invention.

Here the guidelines caution that models and algorithms are inherently mathematical and abstract in nature and any claim directed purely to a core AI model is liable to fall within the ambit of the mathematical method exclusion. Further, to the extent that a core AI invention is limited to a practical implementation, it should be reviewed to determine if it makes a technical contribution falling outside of the exclusions.

The guidelines cite several EPO technical board of appeal decisions to support this position.[17]

An appeal by Google, for example, involved using a large neural network as a reference for training a smaller distilled neural network. The board found last year that while the smaller model had reduced memory requirements, this was exclusively due to the model being smaller and having an associated lower accuracy. The board therefore concluded that this was inadequate to establish the presence of a technical effect.[18]

Machine Learning

AI inventions often rely on models which need to be trained before they fulfil their intended purpose. The guidelines consider that these machine learning methods are analogous to calibration techniques found in other fields, and that a computer implemented method of calibration could be found to be a patentable invention under the Aerotel approach.

The guidelines therefore conclude that a method of training may meet one or more of the signposts. However, where an AI invention is trained for a nontechnical purpose, and there is no more to it, then the invention is not patentable.[19]

Training Data

Training data for AI models may be claimed in various ways, either independently or in combination with other features.

Where the training data is claimed as a constituent part of a wider training method, the guidelines note that whether the training data is excluded from patentability or not will be determined by whether the training method as a whole makes a technical contribution.

Where the training data is claimed as a method of generating or improving a dataset, the guidelines note that whether the method is a patentable invention will be determined by whether the method makes a technical contribution.

However, where the training data is itself claimed independently as a data set characterised by its content, the guidelines note that it is unlikely that the claim will be found to be a patentable invention. The data set is instead likely to be excluded under the presentation of information exclusion.[20]

Takeaways

Patents for AI inventions are obtainable in all fields of technology and will be assessed in the same way as other computer implemented inventions according to the well-established Aerotel approach. When the process performed by an AI invention makes a technical contribution to the existing state of the art, the invention is patentable.

An AI invention will be found to make a technical contribution if it performs or contributes toward solving a technical process or problem outside of a computer, or if its implementation in a computer results in a new way of operating the computer.

These conditions apply equally to inventions involving the training of AI, and to software implementations of AI. Patent protection may also be available to training data sets if they are used as part of an invention which otherwise makes a technical contribution.

The presence of a technical contribution is paramount to the UKIPO's patentability assessment of AI inventions, which European practitioners will recognize is similar to the European Patent Office's approach.

Accordingly, similar drafting strategies may be adopted as would be adopted when drafting European patent applications for AI inventions. In particular, care should be taken to ensure that the technical contribution made by the AI invention is clearly explained within the application.

This explanation should focus on describing how the technical contribution is achieved. Statements alleging a technical effect without explanation of how the effect is achieved are unlikely to be successful.

Instead, an explanation of the causal relationship between specific claim features and the alleged technical effect needs to be provided. Practitioners are strongly encouraged to describe, where possible, both the technical effects falling outside of the computer, and any technical effects arising from the specific implementation of the AI invention in a computer.

This approach improves the applicant's options for demonstrating that the AI invention provides a technical contribution during examination of the application before the UKIPO.

Endnotes

[1] Emotional Perception AI Ltd v. Comptroller-General of Patents, Designs and Trade Marks  [2023] EWHC 2948 (Ch) and Comptroller-General of Patents, Designs and Trade Marks v. Emotional Perception AI Ltd.  [2024] EWCA Civ 825.

[2] "Guidelines for examining patent applications relating to artificial intelligence (AI)" and "Scenarios applying the guidelines for examining patent applications for AI," both updated January 30, 2025.

[3] The Patents Act 1977, Section 1.

[4] See Emotional Perception EWCA, paragraph 61.

[5] See Emotional Perception EWCA, paragraph 68.

[6] See Emotional Perception EWCA, paragraphs 70, 71 and 83.

[7] Aerotel Ltd v. Telco Holdings Ltd. & Ors  [2006] EWCA Civ 1371.

[8] AT&T Knowledge Ventures LP [2009] EWHC 343 (Pat) and HTC Europe Co Ltd v. Apple Inc.  [2013] EWCA Civ 451.

[9] See guidelines, paragraph 5.2.

[10] T208/84.

[11] Re Halliburton Energy Services Inc. [2011] EWHC 2508 (Pat).

[12] See guidelines, paragraphs 5.36 and 5.37.

[13] See guidelines, paragraphs 5.41 and 5.44.

[14] See guidelines, paragraph 6.2.

[15] Symbian Ltd v. Comptroller General of Patents  [2008] EWCA Civ 1066.

[16] Gale's Application [1991] RPC 305.

[17] See guidelines, paragraph 7.10.

[18] Google, T1425/21.

[19] See guidelines, paragraphs 8.2 and 8.5.

[20] See guidelines, paragraphs 8.9 and 8.10.