What is AI, and can it be patented?

AI is a hot topic at present, both in the media and in the IP world. As is often the case with matters that suddenly pique the public's interest, it is frequently misunderstood and misrepresented. The intention of this article is, therefore, to address some of the prevalent misunderstandings and identify some developing themes and topics in this area.

Firstly, to debunk a couple of myths. I recently discussed an invention with a biochemist who remarked that "it's got an AI in it." This seems to be a commonplace way of referring to the use of AI to facilitate certain functionality within an invention. However, to be clear, "an AI" is not a new species of beast. Mentioning that the invention works "using an AI" is somewhat akin to telling a small child that the steam train is dragon-powered. It sounds impressive; the appearance is convincing, but dragons do not actually exist.

So, if "an AI" is not an awe-inspiring fire-breathing dragon sitting in the belly of your steam engine, what actually is it? Most so-called Artificial Intelligence is based on neural networks. The theoretical basis for neural networks was established before 1900 with reference to the interactions of large numbers of neurons in the brain. Artificial neural networks were researched enthusiastically and extensively through the 1940s, 1950s, and 1960s. This research did not translate to commercial reality at the time for want of processing power.

The recent rapid increase in commercial applicability of AI in recent years has been catalysed by the availability of both processing power and large data sets.

An artificial neural network would be better described as a tool or facilitator of desired functionality. As such, there is a parallel with a previous hot topic of patentability, namely software. As is by now reasonably well established, although the law in most jurisdictions precludes the patenting of "software as such", patents can be obtained for software-implemented inventions. For example, the EPO guidelines for examination state that the exclusion does not apply to computer programs with a technical character or produce a further technical effect when run on a computer. In other words, if the technical effect exists solely within the computer, then arguing patentability is going to be much harder than when the technical effect is achieved outside the computer.

I suggest that AI can be viewed in a similar manner. The same algorithm could be found patentable or not patentable depending on what it is doing, with what it is interacting and how it impacts the technical system in which it is included. AI will form part of many patentable inventions as the effects that can be created using it provide considerable improvements over the state of the art. However, AI does not provide a silver bullet to make an otherwise unpatentable invention patentable because it cannot provide the technical effect by itself. As has always been the case with computer-implemented inventions, there are some inventions that fall between a number of exclusions to patentability and innovation in these will not suddenly become patentable because they are facilitated by AI. For example, a software-implemented business method, a business method that enables an improved display of information, or an App presenting information are all unlikely to remain unprotectable by a patent despite including AI.

Although AI is being used to facilitate innovation in almost every area of technology, there are some key areas in which its use is particularly prevalent and in which Stratagem has already successfully drafted and prosecuted patent applications:

• Healthcare and bioinformatics – for example, clinical sample screening and diagnostics, and experimental design in multi-factorial regimens where no single solution exists.
• Automotive – both self-driving vehicle technologies and meta-scale fleet management requiring vehicle-to-vehicle real-time data processing and communication.

Another area which gives rise to IP related discussion is the foundation data on which AI is 'trained'. Large volumes of data are critical for training AI models, but these data sets can be problematic from an ethical perspective. Depending on the nature of the data and its source, uncomfortable questions about the author's rights, consent, and identity need to be considered.

Where these training data sets incorporate copyright works, there is clearly tension between AI developers and copyright holders. A code of practice was in the process of being prepared by the UK IPO in consultation with representatives from the technology, creative and research sectors, but in February 2024, it became clear that the working group was not able to agree on an effective voluntary code. As it stands, therefore, it remains the case that the use of copyright works as AI training data will infringe copyright unless permitted under licence or an exemption.

In summary, AI is just the latest facilitator of progress. It is neither a panacea nor a diabolical influence. The ethics around training data sets are murky and need to be considered deeply before placing undue reliance on AI-enabled offerings.

Our patent attorneys can help with all patent matters. Please get in touch with us at mail@stratagemipm.co.uk.