What do weaving, glass-making and artificial intelligence all have in common? At first glance, perhaps not much. They all have, however, been impacted by patents and the laws that protect innovation.

The use of patents stretches back hundreds of years, and in England the concept was written into law in 1623. Bruce Dearling, a patent attorney and partner at Hepworth Browne, explains that the Statute of Monopolies was introduced when the government realised that innovation should be rewarded. Subsequently, inventors have been granted monopoly rights for up to 20 years in return for sharing their knowledge to advance the nation’s industry. This worked well for the weavers and glassblowers of days past and the arrangement advanced England’s status as an industrial powerhouse.

12th Jan 2024

Landmark Ruling in the High Court Opens Doors for Artificial Intelligence (AI) Entrepreneurs and Investors Seeking Patent Protection in the UK

Patenting the Future:

TMC2’s Legal Triumph Paves Way for AI Innovations

The law evolved along with technological innovation, but until recently artificial intelligence hasn’t been given the same kind of patent protection. That is set to change, however, now that the High Court has ruled that artificial intelligence - particularly artificial neural networks - are worthy of patent protection. The seminal case, which Dearling brought to the court on behalf of TMC2’s subsidiary innovation incubator Emotional Perception AI, marks a watershed moment in which more AI entrepreneurs - and their investors - can get their innovations patented in the UK.

Why are patents important?

Patents reward the brilliance of inventors and give them rights over their inventions so they can be sold, or licensed, as a commercial product. It also stops others from impinging on their work and gives them certainty that their research and development will be rewarded in the long run. For investors in AI start-ups, patents make it more likely they will get their money back.

“There is a direct correlation between company values and patent rights,” says Dearling, who points to academic research that indicates that each national patent can add a seven-figure euro value to a valuation prior to a company’s initial public offering (IPO).

According to research from the EU Intellectual Property Office and the European Patent Office that was published in October 2023, companies at the early growth stage were 6.4 times more likely to get venture capital funding if they had filed patents. And investors were twice as likely to have a successful exit if the company had patents.

“It is good for the industry and in particular good for our AI ecosystem. I believe we are still in the very early stage of this technology"

Prof Andy Pardoe | Senior Executive & AI Leader

AI: just a computer program?

Certain computer programs have been excluded from patent law because they do not technically contribute to basic human endeavour. They have been given copyright protection making them more akin to literary works rather than technical inventions. This is where the recent High Court case that TMC2 won comes into play, because a key legal principle was at stake: whether aspects of artificial intelligence - particularly artificial neural networks - are substantially different from computer programs and therefore worthy of patent protection.

Professor Andy Pardoe, a leading expert in the field of AI, believes they are. A traditional computer program has an input, a ‘black box’ of code, and an output. The output is predictable, i.e. a known output or expected condition and will be the same each time. By contrast, an artificial neural network (ANN) – as a subset of machine learning - is set up differently. The ANN has neural nodes that replicate how the human brain processes information, and it is trained on data so it can teach itself to solve problems. The more it does - always giving itself feedback - the more accurate it becomes. The ANN can be used for problems that cannot be simply coded and where the output or indeed even the input conditions are unknown.

“The whole mechanism of AI is very different to traditional writing a computer program. It is not static - it changes all the time what we do with it,” says Pardoe, who is also an advisor to TMC2.

This was part of the argument that Emotional Perception - a company under TMC2’s umbrella - made to the High Court: the creation of the ANN is setting the stage for the machine to work out its own solution, and this is different from a straightforward computer program. And, given the time and resources that go into training such neural networks, it is only fair that innovators are rewarded with patent protection for their efforts.

The law catches up

Patent law has evolved from the days of weavers and glassblowers, but the UK’s current Patents Act was established in 1977 - a time when computing was very different. Dearling explains, “The legal system generally lags in time behind technical innovations. This is certainly the case for AI technologies and how they must be treated under patent statute written 50 years ago and interpreted over that time by a handful of legal cases.”

The High Court ruling from November 2023 is likely to set a precedent that the training of artificial neural networks, and the networks themselves can be patented, a ruling that could have a far-reaching impact for the AI industry in the UK.

Closing the semantic gap

The case centred on a patent application that Emotional Perception filed in 2019 for a technique that trains an ANN to perceive content in a similar way to a human. The invention had three principal inventors: Dr. Joe Lyske (Chief Technology Officer of TMC2) and Dr. Nadine Kroher (Co-Head of Research at TMC2) and Professor Aggelos Pikrakis (Co-Head of Research at TMC2).

The innovation focuses on closing the semantic gap, which is the difference between reality and what the ANN suggests is reality. Closing the semantic gap has often been described as the ‘holy grail’ in AI technology.

Kroher talks through what this semantic gap means for this particular invention. Imagine you have a song in mind - say ACDC’s ‘Highway to Hell’ - and you want to find something similar that is royalty free. In the past, such technology would have classified music by genres and given each track a tag. With ACDC, it would be classified as ‘hard rock’, and the tool would give you another song that is also hard rock. There are problems with this approach, however. “There is the problem of granularity - someone needs to define a taxonomy,” says Kroher. If the classification is too general, like ‘rock’ then the bucket is too big and so many things could be included, which renders the description meaningless. And if the buckets are too narrow, it can be too difficult to find the right one. Kroher asks: “What bucket would you put Queen’s Bohemian Rhapsody in?” And what songs are similar to Bohemian Rhapsody? These are tricky questions that traditional methods would have struggled with. A neural network that has been trained in the right way, however, knows exactly how to go about finding similarities.

Getting rid of buckets

Kroher explains that their technology aimed to get around the traditional classification and they asked, “Can we get rid of buckets?” The aim was to measure two different songs in the same way that a human would perceive them. Instead of a tag with a classification, the output would be a vector - a dot that floats along many dimensions - and the closer the two dots are together, the closer a human would perceive them.

You can train a neural network with other types of content, like poems or paintings, and it maps them to a space and shows things that a human would perceive to be close together. Such technology could show clouds of vectors that represent trends in social culture as they emerge, such as videos of hamsters eating food.

The ANN was trained using vast troves of textual descriptions of songs, which already exist in song libraries, and it learned how this compares with how a human perceives a song. A track carries all sorts of information, such as how fast paced it is, how traditional, how hard - all things that can be measured by a machine. Just because they have the same measurement, however, does not mean they are perceived in the same way. The way that humans process the information - after a lifetime of listening to lots of music - is a lot more complex. Kroher explains that closing that gap is taking something that a computer can measure to a higher level of abstraction that is closer to how a piece of content is actually perceived.

Kroher is keen to point out that this is not the first attempt at closing the semantic gap. “The way that we do it is unique - that is the bit that is patented,” she says.

Defining ‘technical function’

The Patent Office - known formally as the Intellectual Property Office of the United Kingdom (UKIPO) - rejected the patent application in part because it deemed the innovation no different from a computer program, which are usually excluded from patent law.

Dearling explains that the law states a patent must have a ‘technical contribution’ and it must also add to human endeavour. The UKIPO didn’t believe there was a ‘technical contribution’, although what counts as having this is not defined in law, says Dearling.

At the time of writing (December 2023), the UKIPO still had the option of lodging an appeal. This means that the case could then be heard at the Court of Appeal, or it could be leapfrogged straight to the Supreme Court - the final destination for cases in England and Wales - if it is decided that the case is in the public interest or contains a significant point of law that needs to be decided.

Impact of the ruling

Emotional Perception has complementary patent rights for its inventions in the United States, and with the High Court ruling in their favour, it is well positioned to develop world-leading concepts that use these monopoly rights. The case, however, wasn’t just about acquiring another patent, as Lyske commented at the time of the High Court judgement: “It’s a game changer for the whole AI industry.”

Dearling also describes the impact of the ruling: “It puts the UK front and centre in protecting AI innovation, possibly even now ahead of the USA which has historically had a more liberal understanding,” he says.

There are other applications for this kind of technology, such as searching for video files or text, and for searching and correlating medical records, for example. “The invention underscores a powerful next generation search tool,” says Dearling.

Pardoe says of the High Court ruling, “It is good for the industry and in particular good for our AI ecosystem. I believe we are still in the very early stage of this technology - there is a lot more innovation and ideas to come in 10 years’ and 20 years’ time.”

Editor's note, 15 January 2024 - Since this article was written, the Comptroller General of Patents has been given permission to appeal the High Court decision to the Court of Appeal. The Patent Office’s appeal will argue, again, that the invention is a program for a computer but will also now allege that the invention is excluded for being a mathematical method. Both issues will be defended by Emotional Perception AI arguing that the initial decision was correct regardless of the approach taken.