The patentability of AI-related inventions

AI has the power to revolutionise almost any industry and, without doubt, the life sciences can benefit enormously from the leaps and bounds being made in intelligent technology.

AI’s ability to analyse vast datasets, identify patterns and make predictions makes it well suited to drug discovery and development, and for understanding the role of complex genetic interactions in disease so that treatments can be tailored based on genomics. And its role in tech-based analysis and monitoring has the potential to enhance disease diagnosis and medical devices in a truly revolutionary way.

Patent law in the UK as we know it today dates back to the 1970s, when the idea of AI was the stuff of science fiction. As the legal framework for protecting innovation was established without AI in mind, it is inevitable that the law is having to play catch-up with the commercial reality of using AI to assist in innovation and with the development of AI technology itself. The increasing prevalence of AI is demanding a reassessment of fundamental patent law concepts and their application to new technology.

Conversations and consultations are going on worldwide. The UK Intellectual Property Office (UKIPO) sought views on AI and IP in 2020, following up with a further consultation specifically on whether the patent system is equipped to deal with AI in 2022.

It concluded that there was no evidence that UK patent law is currently inappropriate to protect inventions made using AI, but it remains open to the possibility that change may be needed in the future when there is a stronger technological case.

The UK Intellectual Property Office has published, and subsequently updated, guidelines for examination of patent applications relating to AI (at the time of writing, these guidelines have been temporarily suspended pending potential revision following judgment in the Emotional Perception case mentioned below).

So the position remains that patents may be granted for AI-related inventions, provided that the application satisfies the legal requirements set out in the Patents Act 1977. Specifically, the application must claim something that is novel, involves an inventive step, is capable of industrial application and does not fall within one of the statutory exclusions from patentability, which include computer programs, mathematical methods, business methods and presentation of information. The invention must also be disclosed clearly and completely enough for it to be performed by the skilled person.

In considering the issues that are likely to be encountered in patenting AI-related inventions, it is convenient to consider the different types of invention where AI may play a part.

First, there are inventions of AI technologies themselves, such as platforms for drug discovery.

Second, there are AI-based inventions, where AI is an element of the invention but a part of something else, like a medical diagnostic device.

Third, there are AI-assisted inventions, where AI is used in the inventive process but is not an element of the claims of the patent application, such as a new drug developed from a candidate identified by an AI tool. And finally, there are inventions which are generated by AI itself. Inventions within each of these categories encounter different issues with patentability—I will look at them in turn.

AI technologies

The pressure is on to develop ever more sophisticated AI systems, such as platforms that make recommendations by analysing vast datasets of molecular structures to predict drug candidates. AI is based on computational models and mathematical algorithms and consequently inventions in this space risk engaging the statutory exclusions from patentability, particularly "a program for a computer … as such" and mathematical models.

In the UK, in order to escape the statutory exclusions, the invention must amount to more than a computer program (or mathematical method)—it must provide a “technical effect”. Guidance from the UKIPO Manual of Patent Practice provides that a computer-related invention will not be excluded from patentability if it is directed to a specific technical process outside of a computer and contributes to a solution of a technical problem lying outside of the computer.

This requirement has recently been considered in detail by the UK High Court, in the case of Emotional Perception AI Ltd v Comptroller-General of Patents  [2023] EWHC 2948 (Ch). This was an appeal challenging a decision by the UKIPO to refuse to grant Emotional Perception’s patent. The application claimed an improved system for providing media file recommendations to an end user. This might be used, for example, by a music website where a user may be interested in receiving music similar to another track.

The High Court found that the artificial neural networks (ANNs) underlying the invention were not themselves a computer program, as they learn via a training process rather than by implementing code provided by a human. Neither was the internal training and subsequent operation of the trained ANN a computer program for the purposes of the exclusion. Although computer programming was involved in setting the training objectives, the patent does not claim that program and the program is subsidiary to the invention, so the exclusion is not invoked.

The High Court went on to consider the position if it was wrong about the claim not being a claim to a computer program, having particular regard to whether the computer program made a technical contribution outside itself. Significantly, the UKIPO had been influenced by the beneficial effect of the claimed invention (the recommendation of a music track) being of “a subjective and cognitive nature”.

The High Court disagreed that this was a determining factor, as the ANN has gone about its analysis and selection in a technical way. That is a technical effect outside the computer and the possible subjective effect should not disqualify it.

The court held that if the computer program was either the training program or the overall training activity, the resulting trained ANN may itself constitute a technical effect which prevents the exclusion applying. The AI system was therefore patentable.

AI-based inventions

AI-based inventions are computer implemented inventions, where AI is an element of the invention. Such inventions often fall foul of the disclosure requirements (resulting in a finding of insufficiency) and/or lack inventive step.

This was exemplified by ARC Siebersdorf’s application (EPO case T161/18, May 2020) which claimed a method for determining aortic blood pressure based on blood pressure measurements taken at the periphery of the body. The application was refused on grounds of lack of sufficiency as there was insufficient disclosure of how to obtain the trained AI needed to work the invention—training data was mentioned only at a very high level.

The system also lacked inventive step because the use of AI neural networks was already known for the transformation of the blood pressure curve measured at the periphery into the equivalent aortic pressure.

An AI-based invention that has been granted a patent by the European Patent Office (EPO) followed an application by Oxipit (resulting in EP 3 651 117 B1) for a system for detecting irregularities in medical images by means of a machine learning model. In that case, the application gave sufficient detail of how to train a machine learning model using a dataset of images with irregularities in defined segments.

Plausibility may also be an issue for AI-based inventions. The specification must show how the AI is adapted to solve a technical problem to show advantages across the claim scope. This may require the inclusion of experimental data that demonstrates advantages over the prior art and an indication of how this can be extrapolated across the entire scope of claims.

So the message is clear as for all patents, the specification must not be an invitation to a research programme, but must enable the invention so that the skilled person (applying their common general knowledge) can carry out the invention across its scope. This will involve including detailed information about, for example, the training data and training methods, and validation data. Perhaps in time a data bank will be needed to support patents, akin to the existing patent depositories for biomaterials.

AI-assisted inventions

AI-assisted inventions are where AI has been used as a tool in the inventive process, but is not an element of the claims. Sophisticated drug discovery platforms are now available which use AI to analyse biomedical information for target identification, molecular design and precision medicine. The main issue here is with inventive step—is there the necessary degree of invention if an AI system just replaces a manual task?

Patents have been granted for compounds discovered using such platforms (see, for example, BenevolentAI’s patent on SSAO inhibitors, EP 3 286 189 B1). However, this is likely to be an area that sees an increase in issues relating to patentability over time.

Many questions and challenges arise. Inventiveness has to be assessed by considering AI as it was at the priority date, ignoring capabilities developed later. However, pinpointing this in an area as fast-moving as AI is likely to be a challenge.

Inventiveness may become an increasingly difficult hurdle as AI platforms become ever more sophisticated and widely used, making many elements of research and development routine and leading to an increase in the threshold for obviousness.

There is no obligation to disclose in the specification that AI was involved in the inventive process, so will all applications have to be examined on the assumption that it was? And are we going to need the concept of “the skilled AI” alongside the person skilled in the art in order to consider what is obvious?

Inventions generated by AI

The Artificial Inventor Project sought to claim IP rights for inventions generated by AI in the absence of human creative input. Dr Thaler created an AI system known as DABUS (Device for the Autonomous Bootstrapping of Unified Sentience), which autonomously invented a food container and an emergency warning light. Dr Thaler sought patent protection for these products, naming DABUS as the inventor.

The question as to whether an AI system can be named as inventor on a patent has now been considered in numerous jurisdictions worldwide, and most recently by the UK Supreme Court which handed down the final UK judgment in December 2023. The decision has been unanimous in all jurisdictions which examine patents on more than formalities—a human being must be named as inventor on a patent.

So the AI itself must take a backseat when it comes to attribution of the invention. This would seem to be right, given that AI systems have no legal personality. And, on the positive side, the Thaler case has revealed an apparent acceptance by patent offices and courts around the world that AI systems are capable of creating things that are patentable inventions.

Conclusion

Patent applicants, attorneys and lawyers are hungry for patentability guidance and will eagerly await further decisions from the world’s patent offices and courts. But in an area as fast moving and complex as AI, these decisions are unlikely to provide a complete answer and it is conceivable that an overhaul of the legislation will be needed before too long.

It is also likely that challenges in patent protection will fuel increasing interest in trade secrets, which may offer effective protection for AI inventions given that reverse engineering is often impossible. But will the world be poorer from the resulting reduction in the sharing of knowledge?

Patentability of AI-related inventions is likely to be replete with questions and challenges for many years to come. We need to ensure that the answers that develop do not stifle innovation, so that the life sciences industry can reap the full benefit of modern technology.

Rebecca Lawrence is a partner at DLA Piper UK.