<iframe src="//www.googletagmanager.com/ns.html?id=GTM-5T7PGR" height="0" width="0" style="display:none;visibility:hidden">

FROM THE BLOG

How to Protect your AI Innovations with a Patent: Updated EPO Guidelines

Posted by on 10 October 2019

Software has become ubiquitous: it has reached almost all areas of industry and commerce. Artificial intelligence and machine learning in particular are game-changers – fueled by more data and faster hardware they show no signs of slowing down. Ever increasingly it is the software component of products and services which plays the decisive role in determining their success. In fact, a strong integration of software into your products can end up changing your business model.

The purpose of this blog post is to go through the part of the updated Guidelines of the European Patent Office (EPO) which relate to artificial intelligence and machine learning. The updated Guidelines will come into effect on the 1 November 2019 and they will be used by the EPO during the examination of patent applications. It is also our aim to bring across the basics of patenting AI inventions with some practical tips and strategies.

"The New Electricity"How-to-protect-your-AI-innovations-Kerpen-Rentschpartner-Froriep

"Just as electricity transformed almost everything 100 years ago, today I actually have a hard time thinking of an industry that I don’t think AI will transform in the next several years,"Andrew Ng, one of the foremost experts on AI.

Artificial intelligence (AI) and machine learning (ML) are here to stay. Besting humans in complex games such as Go and Poker was just the beginning. Today, companies in fields as diverse as life and medical sciences, telecommunications, energy management, security, and manufacturing are seeing the benefits that artificial intelligence and machine learning can bring. This impact is reflected both in the number of scientific publications in the field (over 1.6 million and counting) and also in the number of patent filings (nearly 340'000 worldwide to date). The pace shows no signs of slowing down: patent filings in deep learning (an area of AI) experienced an average annual growth rate of 175% between 2013 and 2016. Computer vision, biometrics, natural language and speech processing, robotics, and predictive analytics are some of the technological areas which have seen the largest number of patent filings. WIPO Technology Trends 2019: Artificial Intelligence

Is AI not an unpatentable abstract mathematical idea?

The abstract mathematical nature of the computational methods and algorithms behind artificial intelligence and machine learning does not preclude them from patentability in Europe altogether. Although the computational methods and the algorithms taken by themselves are abstract mathematical ideas and therefore cannot be patented according to European patent law, technical solutions to technical problems are patentable. Therefore, if the AI component (e.g. a neural network) of an invention is directed towards achieving a technical effect, it can make a technical contribution and make the patent allowable.

In light of this, the EPO has specifically mentioned artificial intelligence and machine learning in their updated guidelines for examination of patent applications. The updated guidelines confirm that the EPO is treating artificial intelligence and machine learning in line with the settled case law and general practice of the EPO regarding software patents, which is permissive to patents on software directed towards achieving a technical effect.

What needs to be considered when assessing an AI invention?

To help clarify whether the AI technology your company has developed may be susceptible to patentability, we recommend asking: what does the AI do or enable? It may be that the AI component is able to make sense of data in a novel way, for example finding previously hidden correlations and relationships, or it may be that the AI component provides better robustness or accuracy than previous methods. It may also be that the AI component increases the speed of a calculation and therefore achieves computational efficiency, or that it allows a calculation to be carried out on a piece of low-power hardware such as an IoT device or a smartphone.

It is also important to understand what kind of data is used and produced by the AI: what data is input into the AI and what is the data output? If the input data is of a technical nature, such as physical quantities (i.e. measurable by a device, such as temperature, noise, position), then it is more likely that the AI-component makes a technical contribution and the invention could be susceptible to patentability. Likewise, if the output of the AI-component is used to achieve a technical effect beyond the normal interaction that occurs between software and the processor, then chances are good that the invention could be patentable. If neither of these is the case, then the invention needs to be looked at more closely, as it is often the subtleties of a particular invention that become decisive.

How does this work in practice?

In this section, we provide more patent specific details and recommendations regarding European patent practice.

In general, an AI invention can be placed in one of the following three categories:

  1. AI algorithms as such: these are currently not patentable as they are not considered inventions under Art. 52 EPC. For example, this could apply when a patent is sought for an AI algorithm on its own - separate from any technical context.
  2. Generic trained models: a generic trained model which can be applied to many technical problems may be difficult to patent – providing sufficient comparative examples and parameter ranges may be needed. For example, if an AI invention is capable of both detecting anomalies in MRI images and determining the wear of truck tires, we would recommend providing sufficient detail for both cases – or filing separate patent applications, each directed to one particular technical problem (and application).
  3. AI applied to a particular technical problem – here the AI component is typically part of an encompassing claim which is directed towards a concrete technical problem, and therefore is patent eligible.

Read the whole EPO Conference Summary: Patenting Artificial Intelligence.

When filing AI patent applications, we recommend including in the specification as much information as possible regarding the technical effect produced or enabled by the AI component. Further, defining the AI component (e.g. by specifying its data input/output and network architecture) is recommended for purposes of sufficient disclosure and clarity. Jargon should be avoided, for example even if a new type of neural network or layer has a name that has been established in a number of scientific publications, it is advantageous to define exactly what is meant.

The claims of the patent application define the subject-matter for which protection is sought. Broad claim types include method and device. For AI inventions we recommend claiming both the method itself, as carried out by a computerised device (defining a step-by-step process of how the invention is carried out), as well as the computerised device itself, configured to implement the method. Further, the steps of generating the training set and training the AI component may also contribute to the technical character and can also be claimed.

How do I protect my AI invention worldwide?

The realities of a globalised world demand that patent protection is sought in multiple countries, depending on which your most important markets are and where your competitors are located. As different national patent offices have slightly different criteria when assessing patent applications, it is vital to select the right national patent office for filing your first application. A good option here is the EPO, as applications which satisfy the European requirements are usually unproblematic when it comes to subsequent filings, for example in the US. In our experience, the converse is not always the case as the US drafting style, in particular regarding software inventions, can lead to objections when it comes to a subsequent European filing.

Within one year of filing the first application, subsequent applications can be filed in other jurisdictions for the same invention. If the number of jurisdictions sought is high, it can make sense to file an international PCT application (Patent Cooperation Treaty), which also has the advantage of giving the applicant another 18 months to decide in which jurisdictions to ultimately file subsequent applications.

Conclusion

The EPO has recognised the importance of artificial intelligence and machine learning in inventions and its updated guidelines underscore that inventions based on artificial intelligence and machine learning will be examined in the manner established for other software inventions. If your company uses, or intends to use, AI in its products or services, we suggest assessing how to best protect and generate value from this intellectual property.

This is another guest blogpost written by our cooperation partner Rentsch Partner AG.

If you liked this article you may also be interested in reading about:

Guest blog post written by Alena Bach and Demian Stauber on "Patents for Tech Startups". 

 

Stay on top of the latest legal topics and subscribe to our blog here:

SUBSCRIBE TO OUR BLOG

 

Topics: Intellectual Property

  
Name 13

Dany Vogel | Rentsch Partner

Dany Vogel works as a patent attorney focusing on electrical, computer and information technologies. He has particular expertise in computer software which covers a broad spectrum including telecommunications, digital signal, image and information processing, medical and control engineering, cryptography, Blockchain, Fintech, and computer-implemented business methods.

Connect with me:
Name 13

Philip Kerpen | Rentsch Partner

Philip Kerpen works as a patent engineer focusing on physics, semi-conductor technology and computer and information technology, especially the areas of measurement and control engineering, cryptography, artificial intelligence, machine learning, neural networks and blockchain.

Connect with me:
https://blog.froriep.com/hubfs/IMAGES_BLOGPOST_850x850/How-to-protect-your-AI-innovations-Kerpen-Rentschpartner-Froriep.jpg