Author Patent Consultant Maria Laini.
In the recently published Patent Index 2024, the EPO has shown that computer technology became, for the first time, the leading field for filing of patent applications (1). Certainly, this does not come as a surprise, as the latest years have seen a staggering worldwide growth of AI innovation.
Less recent, but still relevant data comes from the USPTO in the published AI trends of 2022 (2), which illustrated a rapidly growing volume of published US patent applications in various fields of AI.
What is AI technology, and in which fields are patents being filed?
AI technology is an umbrella term covering an extremely large variety of applications, which attempt to mimic human intelligence in fields such as knowledge processing, language processing, planning and control.
In 2020, the greatest volume of published patent applications in the US came from the fields of knowledge processing, planning and control. In particular, the number of published patent applications in these fields rose from 40,000 to 50,000 between 2018 and 2020.
The field of knowledge processing aims at creating artificial intelligence solutions that can answer questions intelligently, come up with hypotheses and make deductions. AI planning and control aims at creating an artificial intelligence that can formulate strategies and action sequences. These two fields go hand-in-hand in many hot spots of contemporary research, such as the quest to a completely autonomous vehicle, or a surgical robot.
These data show how competition in the field of software solutions, and in particular in the AI field, is surging at an impressive rate. Businesses are working hard to secure their intellectual property rights, and their first-row seat in the AI innovation arena.
In this space, it is crucially important for businesses to think carefully about their strategy to protect their innovation. Starting a new business and putting your innovation out there can feel like trying to put fences around your newly built house in a field that is already rather crowded with fenced-houses and other people trying to build their own fenced-house. As it is not advisable to step into someone’s garden, or try to build a fence around someone else’s property, it is vital to know where you stand in the innovation arena, where property is intangible but equally real.
Can AI inventions be protected?
Contrary to a popular legend, computer-implemented inventions can be, in principle, eligible for patent protection. This is demonstrated by the Patent Index 2024 (1), which shows that many businesses are building their patent portfolio in the computer technology sector.
However, once established that a computer-implemented invention, and thus an AI-based solution, can be patentable, it is still very important to be aware that there are a few dangers to look out for. These dangers stem from two non-negotiables of patent eligibility:
- the presence of a non-obvious technical effect,
- and the need for sufficiency of disclosure.
The need for a non-obvious technical effect
The legal definition of “technical effect” is not trivial. Patent Offices around the world have built their own portfolio of case law and guidelines to define what is technical, and what is not, according to local Intellectual Property law.
The EPO, for example, has published detailed guidelines on what would or would not be considered a feature with a technical character (3). As can be seen in said guidelines and examples, extra care needs to be taken with computer-implemented inventions.
For example, mental acts and mathematical methods are excluded from patentability at the EPO (so as in other jurisdictions, such as the US). Because AI mimics human actions and thoughts, this exclusion becomes particularly relevant in this sector. Claims need to be drafted in such a way that they do not merely recite steps that can be performed by a human. There needs to be some “hardware” involved, to avoid an outright exclusion from patent protection.
Moreover, even if standard functioning of computer hardware needed to run a computer program is per se technical, it would not be considered inventive. However, a novel way to control technical processes within computer hardware components may be inventive.
On the software side, the EPO Guidelines illustrate that the mere activity of coding is not patentable as such. However, software-implemented steps which control the computer internal processes in a novel way may be technical, if it can be demonstrated that they produce a technical advantage. Of course, producing a software that controls the operation of a machine which performs a technical operation (e.g., a software to control a robot for planting seeds) would be considered technical.
The USPTO has developed their own guidelines and case law to determine whether an invention, and in particular a software-implemented invention, is patentable. These are in some respects similar to the EPO guidelines, but divergence in practice is not uncommon.
The need for clear and “sufficient” instructions
Patent offices around the world agree that a patent application should disclose the invention sufficiently well. From a European perspective, this point is formalized in Article 83 EPC, but every patent office has an equivalent article.
In layman terms, a patent awards the owner with the benefit to prevent a third party from copying their invention, as protected by the claims. This comes at the price of revealing good and clear instructions on how to make the invention itself. You may think of it as disclosing the recipe for a cake: you may be able to keep for yourself some details about the icing, but you cannot omit the use of flour and eggs (I am not an expert of vegan cakes, so I leave them out of the analogy).
When it comes to AI, it cannot be ignored that the requirement for sufficient disclosure demanded by patent offices conflicts with the desire or need to keep the AI-tool as a black box.
Take on messages for effective patent protection
In order to secure patents for AI-solutions, it is important to keep the focus on two main points.
- It is important to be specific and clear on the technical advantage given by the AI-solution. It is unlikely that a patent can be obtained by merely claiming steps done by a generic AI-trained model, so as it is unlikely that patent protection can be obtained on the basis of an AI-model producing parameters and data, without claiming how these parameters and data will be used to achieve a specific, clear and non-obvious technical effect.
- Care needs to be taken when thinking about sufficiency. For example, if an application concerns an invention that uses an artificial neural network (ANN), solely claiming a generic ANN without any information regarding the choice of model and the learning algorithm is most likely inadequate, from a sufficiency perspective. Think again about the (non-vegan) cake recipe: you may be able to omit listing all the different types of chocolate or coconut sprinkles, but you probably cannot keep to yourself the need for a certain ratio between eggs and flour.
Trade secrets: why not?
Drafting and prosecuting patent applications requires effort and requires giving away the privilege to keep your invention secret. And in the competitive and crowded field of AI, this may feel unsafe.
It may feel safer and more cost-effective to treat AI-based inventions as trade secrets. Trade secrets have the advantage of offering potentially unlimited protection, as long as they remain undisclosed. Additionally, the initial costs to set them up can be minimal, compared to patents. If an AI-based invention can be easily kept as a black-box, which is sufficiently difficult to reverse-engineer, a trade secret may be seen as a cost-effective and totally adequate option for its protection.
However, trade secrets suffer from serious limitations. Firstly, it has to be stressed that a trade secret cannot offer any protection against successful reverse-engineering. Once someone finds the way to crack your code, your code is not protected by the trade secret anymore. A patent, on the other hand, can be used to protect your business from this risk. This is a very important point to consider, especially in the field of AI, where competition is stiff, and research evolves rapidly.
It can also be hard and costly to go after people who breached confidentiality of the trade secret, because breach of confidentiality can be extremely hard to prove. It is generally easier to prove that a patent has been infringed, as there is a proven, public and official record of what has been protected, from which date and where. This means that, although the initial costs for establishing a trade secret are typically quite small compared to patent applications, the opposite is generally true in a contentious context.
Trade secrets and patents: can they coexist?
There is not a one-size-fit-all approach, when it comes to intellectual property, and both trade secrets and patents have their pros and cons.
However, one important consideration is that it is not always necessary to choose only one or the other. Trade secrets and patents can coexist in the same portfolio. And this is where the role of a patent attorney is really crucial. A patent attorney can help businesses, advising when and how these two rights can coexist effectively, so that advantages of both patents and trade secrets can be leveraged for maximum protection.
Think of the cake analogy for one last time. Even though you will most likely need to disclose your flour-to-egg ratio, you may be able to keep a few tricks secret. A patent attorney has a trained eye to advise on this.
Bibliography
2) Artificial Intelligence (AI) trends in U.S. patents, June 29, 2022, Pairolero Nicholas A., USPTO.
3) EPO Guidelines on examples of technical effects of computer-implemented inventions
