The UK Intellectual Property Office (UKIPO) published their Enhanced Guidance for examining patent applications relating to artificial intelligence (AI) inventions, informed by wide ranging consultations that have taken place over the last 18 months. Although this guidance cannot change the underlying laws regarding exclusions to patentability, or how these laws are interpreted by the courts, the guidance should provide clarity on how the law is applied through generic explanation and specific examples of how the law applies to AI-related inventions. Before this guidance was published, we reviewed the UK law on AI inventions and how the law was being applied by the UKIPO, alongside the corresponding practices of the EPO; our article discussing this can be found here. We have now considered how the guidance may impact UK practice, and this article sets out our preliminary thoughts.
The new guidance separates potential AI inventions into three main categories: applied AI for the purposes of performing processes or solving problems lying outside a computer, applied AI for making computers work better, and core AI. An invention is considered an “applied AI” invention if the invention applies AI techniques to a technical field other than AI. In contrast, a “core AI” invention does not specify any application or use-case for its AI features but instead advances the field of AI itself. The guidance also briefly discusses inventions relating to training an AI and the datasets used.
For each category, the guidance provides a summary of the law, including various specific examples, before providing a brief conclusion on which types of invention are likely to be considered patentable subject matter and which are not. The guidance also provides a short summary of the requirements for sufficiency of an application, noting that “the disclosure of an AI invention relying upon relevant features of a training dataset should teach those details in a manner that enables the invention to be worked across its scope without undue burden”. Applicants should therefore be aware that, for all AI inventions, and particularly those that rely on relevant features of a training data set, the training data should be included in an application to that AI invention wherever possible to reduce the risk of sufficiency objections.
The guidance also provides a brief discussion of “hardware-only” implementations of AI inventions. In particular, it states that merely drafting a claim to cover both software and hardware implementations is not, of itself, sufficient to avoid an invention being considered excluded subject matter. Applicants should therefore be wary of attempting to circumvent excluded subject matter provisions through introducing generic hardware into their claims, as if the software implementation covered by the claim is found to be excluded subject matter as such then the claim as a whole may be considered unpatentable.
Although the conclusions provided in this newly issued guidance do not indicate a significant shift in the position the UKIPO has historically taken on AI inventions, the specific examples of “scenarios” that are provided in the guidance may be of considerable use to inventors, applicants, and foreign patent professionals looking to patent an AI invention in the UK. The scenarios provided in the guidance are intended to demonstrate which inventions are likely to be found to be patentable subject matter and which are likely to be refused across each of the identified categories. These scenarios therefore provide useful indicators as to whether the subject matter of an application will be considered to be patentable subject matter; being able to refer to an analogous scenario presented in the guidance during prosecution may help to persuade a UKIPO examiner that an invention falls within an allowable subject area.
Below we have provided a list of the scenarios presented in the guidance, using the same numbering as used in the guidance. Should any of these particular scenarios be of interest, searching the guidance for the relevant scenario number should direct you to the relevant portion of the document, which discusses the scenario and its relation to the relevant law.
Scenarios which are indicated to not be excluded from patentability include:
Scenario 1 – ANPR system for recognising a vehicle registration number (Applied AI)
Scenario 2 – Monitoring a gas supply system for faults (Applied AI)
Scenario 3 – Analysing and classifying movement from motion sensor data (Applied AI)
Scenario 4 – Detecting cavitation in a pumping system (Applied AI)
Scenario 5 – Controlling a fuel injector in a combustion engine (Applied AI)
Scenario 6 – Measuring percentage of blood leaving a heart (Applied AI)
Scenario 10 – Cache management using a neural network (Applied AI – making computers work better)
Scenario 11 – Continuous user authentication (Applied AI – making computers work better)
Scenario 12 – Virtual keyboard with predictive text entry (Applied AI – making computers work better)
Scenario 16 – Processing neural network on a heterogeneous computing platform (Core AI)
Scenario 17 – Special purpose processing unit for machine learning computations (Core AI)
Scenario 18 – A multiprocessor topology adapted for machine learning (Core AI)
Scenarios which are indicated to be excluded from patentability include:
Scenario 7 – Automated financial instrument trading (Applied AI)
Scenario 8 – Analysing patient health records (Applied AI)
Scenario 9 – Identifying junk e-mail using a trained AI classifier (Applied AI)
Scenario 13 – Optimising a neural network (Core AI)
Scenario 14 – Avoiding unnecessary processing using a neural network (Core AI)
Scenario 15 – Active training of a neural network (Core AI)
From the above, it can be seen that the scenarios that have been presented as allowable are very much in line with the common understanding of existing UKIPO practice. However, inventions relating to improving neural network technologies themselves (either by optimising the network, reducing the amount of unnecessary processing, or improving the training process of the network) may be considered excluded subject matter. This may be an issue for those seeking to patent developments in the field of AI itself, rather than using AI to support developments in other areas; if those developments cannot be considered to perform external processes and/or solve problems external to a computer, then the invention may be considered to fall foul of excluded subject matter provisions. If this understanding is applied strictly, there is a risk that the UK patent system may exclude a burgeoning field of innovation entirely; although copyright is intended as an alternative to patent protection for these inventions, copyright is unlikely to provide a viable substitution for the rigorous protection a patent could provide.
It is also worth noting that protection in the UK can also be achieved through a European patent application filed with the European Patent Office (EPO). The harmonization of patent laws across contacting states to the European Patent Convention means that the EPO and UKIPO each start from the same list of categories of subject matter that are excluded from patent protection. However, the EPO and UKIPO may not apply the same approach when considering whether an AI application falls within a category of excluded subject matter. Should you wish to learn more about the differences between EPO and UKIPO practice (prior to the publication of the UKIPO AI guidance), our analysis of this can be found in our previous article (linked above).
Although the present guidance does not alter the underlying laws regarding exclusions to patentability in the UK, the scenarios presented therein may result in changes in how the UKIPO interpret different areas of AI technology. We will be revisiting UKIPO practice regarding AI inventions in a year or so, once we have had an opportunity to see how UK IPO Examiners are implementing the new supplementary guidance, in order to fully evaluate the scope of any changes to UK practice for AI inventions.