Computer technologies such as Artificial Intelligence (AI) and machine learning are being adopted in fields far from their origins in computer science. In fact, AI is helping to drive forward innovation in almost every industry – from self-driving cars to personal assistants such as Siri and Alexa. The pace of innovation is astounding yet unsurprising given the positive impact this technology can have on our daily lives. As the amount of investment funneled into AI increases, developers are unsurprisingly keen to protect, not only their financial outlay, but their intellectual property. However, patenting AI hasn’t always been a smooth road.
Patenting to protect innovation
Let’s begin with the importance of patenting. Rather than being a necessary evil, the patent system incentivizes companies and individuals to invest in research and development. It promotes a culture of innovation by allowing businesses to reap the rewards of their investment – both monetary and intellectual. Being granted a patent gives the innovator a 20-year period in which to benefit from their revolutionary ideas. Whether that be by bringing a product to market, by earning royalties through licensing or even selling the patent on to an organization in a better position to take it forward.
The pharmaceutical industry is a prime example of this in practice: a drug can be sold by the developer for 20 years before it’s available on the market generically. The patent can be licensed to third parties who can sell or manufacture the product at a cost. Or, as often happens, the intellectual property is bought by a global player with the scale and supply chain needed to rapidly, profitably and cost-effectively roll it out to the widest possible number of people.
In comparison to the pharmaceutical sector, it can be tempting to think the short lifecycles of most tech products might limit a patent’s value. Yet, owning a standard essential patent can transform a business’s commercial potential. Just look at Nokia, which made over $1billion in licensing last year, much of it from the 3,800 Standard Essential Patents it holds related to mobile phones.
The bottom line is, without patent protection, many businesses simply wouldn’t invest in creating new products as they could be reverse engineered and reused by competitors as soon as they’re in the public domain.
The challenges of patenting AI
There are two main conditions for patenting any invention: it must be new and it must be inventive (i.e., not an obvious modification to an existing invention). But, innovations in AI are even more challenging to patent at the European Patent Office (EPO) because the EPO views AI as a type of mathematical method. European patent law prohibits patenting subject-matter that relates to mathematical methods ‘as such’ and so the EPO will ignore features of a patent application that relate only to mathematical methods, such as AI, when they assess novelty and inventiveness.
But, this extra obstacle can be overcome by showing that an invention as a whole is more than just a mathematical method.
Breaking down the barriers
There are two ways to show that an AI invention is more than just a mathematical method: if the AI is applied to a different field of technology and the use of AI produces an effect in that field, or if the AI itself solves a 'technical problem'.
The first assesses whether the use of the AI results in a real-world effect beyond the algorithm. For example, EPO patent EP3073098B1 specifies operating an engine using AI. This causes the engine to be operated differently and so the AI has had an effect in the field of 'engines', which is beyond the field of 'AI'. This means that the EPO can take the use of AI into account when assessing novelty and inventiveness. However, the use of AI must still actually be inventive – patenting an obvious or basic use of an off-the-shelf AI tool is difficult. In EP3073098B1, the patent specifies how the AI is trained with particular engine-related data and these extra details seem to have been enough to convince the EPO of inventiveness. The different field does not need to be a traditional engineering field such as mechanical engineering. For example, the EPO generally accepts classifying data to be a technical field in which AI can make a technical contribution and so the use of AI to classify data can be patentable.
The second is to show that the AI solves a technical problem by itself because it results in an efficient hardware implementation. The EPO has to be satisfied that the AI considers the internal functioning of the computer on which it operates. An AI that is particularly efficient may offer advantages when implemented on a computer, such as a reduced number of operations, but the EPO will consider such an AI to be an improved mathematical method unless the AI is designed according to the internal functioning of the computer. EP3446260B1 is an example of a patent that was granted by the European Patent Office and based on an AI considering the internal functioning of a computer, as the AI was designed to use a computer's memory more efficiently.
When is patenting AI not desirable?
A granted patent can prevent others from using an invention without permission. But, applications for patents are published and will tell the public how to implement an invention, regardless of whether the application is eventually granted.
An alternative to seeking patent protection is to keep the invention secret. If third parties don't know the invention then they are also kept from using it. This option might be appropriate for some AI inventions, such as AI implemented in secret on a private server or if an AI-containing product is hard to reverse engineer. But, keeping it secret will not stop someone else using the invention if they obtain it independently or through reverse engineering. And things get complicated if someone else is able to patent the invention later.
To decide between patenting and secrecy for an AI invention it is important to consider if a patent might actually be granted. For a European patent application this includes considering whether the EPO might see the AI invention as more than just a mathematical method. Proper consideration of whether patent protection is suitable for an AI-based innovation could be vital to recouping investment in research and development. If the invention can be shown to be novel and inventive and be more than just a mathematical method, there is nothing stopping developers from protecting their innovative creations.
Christopher Smith, Senior Associate, Reddie & Grose LLP