UK IPO Publishes Detailed Guidance on Examination of Patent Applications Relating to AI Inventions

In the UK Government’s response to their call for views on artificial intelligence and intellectual property in March 2021, it was stated that the government would “publish enhanced IPO guidelines on patent exclusion practice for AI inventions and engage AI interested sectors, including SMEs, and the patent attorney profession to enhance understanding of UK patent exclusion practice and AI inventions.”  Some 18 months later, these guidelines have finally been published.

Importantly, the guidelines cannot change the underlying law relating to exclusions from patentability, nor the interpretation of this law as applied by the courts which is binding on the UK IPO, and a large portion of the guidelines is given over to recitation of these sources of law at both the generic and specific level as an explanation of how the law applies to AI-related inventions.

The guidelines break down potential AI inventions into: “applied AI” (which, as the name suggests, is the application of AI techniques to fields other than AI itself), subdivided into “performing processes or solving problems outside the computer” and “making computers work better”; “core AI” (advances in the field of AI itself); and “training and datasets”.  For each of these categories the guidelines summarise the applicable law and, where available, sample decisions, before providing a short conclusion on the types of invention which are likely to be patentable and those which are not.  None of these conclusions will come as a surprise to those familiar with this field (there are several mentions of “solve a technical problem”), but they provide a useful summary for inventors and applicants who are not familiar with the existing approach.

However, probably of much greater assistance to inventors and applicants is the set of 18 “scenarios” which is provided alongside the core guidelines and illustrates the UK IPO’s approach to examination of AI-related inventions across the identified categories, providing examples of inventions which are likely to be allowable and those which are likely to be refused.  Whilst it is unlikely that individual future applications will fall directly within the ambit of any of these scenarios (which are carefully defined in terms of example claims), they provide very useful indicators of the areas in which applications are likely to be considered patentable (subject to further examination as to novelty and inventive step) or likely to be refused.  The incorporation of these scenarios into the guidelines also means that being able to make a close analogy with a scenario which is indicated as not excluded from patentability is likely to be highly persuasive to a UK IPO examiner.

Examples of inventions which are indicated in the scenarios as not excluded from patentability include:
• automatic numberplate recognition;
• monitoring a gas supply system for faults;
• analysing and classifying movement from motion sensor data; detecting cavitation in a pumping system;
• controlling a fuel-injector in a combustion engine;
• measuring the percentage of blood leaving the head;
(all technical applications of AI programs/systems outside of the computer)
• cache management;
• continuous user authentication;
• virtual keyboard with predictive text entry;
(all applications of AI which “make computers work better”)
• processing a neural network on a heterogeneous computing platform;
• a special purpose processing unit for machine learning inventions; and
• a multiprocessor topology adapted for machine learning
(all new arrangements of “core AI” which improve the internal working of computers or provide for a new technical way of operating a computer).

Examples of inventions which are indicated in the scenarios as likely to be refused as unpatentable are:
• automated financial instrument trading;
• analysing patient health records;
• identifying junk e-mail using a trained AI classifier;
(generally these are viewed as applications of AI in inherently unpatentable fields)
• optimising a neural network;
• avoiding unnecessary processing using a neural network;
• active training of a neural network
(advances in “core AI” which do not solve a technical problem, do not represent an improvement to the computer itself and/or circumvent these problems rather than providing an improved technical effect).

The guidelines conclude with a short section on sufficiency, which is an important reminder that it is important to consider how much information an AI-related patent application requires in terms of specifying training data and the specific operations and optimisations applied.

It is also worth noting that most patent applications which result in granted patents in the UK are filed with the European Patent Office and so are not examined by the UK IPO at all. Whilst the EPO applies the same black letter law as the UK IPO regarding exclusions from patentability, it adopts a quite different approach to the assessment of patentability, with the assessment of excluded subject matter for most computer-implemented inventions, including AI-related inventions, forming part of the test for inventive step. As well as the advantage of securing wider geographical protection, in recent years, the EPO approach to examination of AI-related inventions has generally been viewed as more favourable to applicants than the UK IPO approach and it remains to be seen whether the new UK guidelines will change this significantly.