New research data has confirmed that innovation activity linked to machine learning (ML) is gathering pace in Europe, potentially allowing businesses to realise value from their investment in artificial intelligence (AI) technologies sooner rather than later.
Research conducted by European intellectual property firm, Withers & Rogers, has revealed that a record number of patent applications linked to machine learning were published by the European Patent Office (EPO) in 2019. The actual number of patent publications related to this field has increased from around 1,000 publications in 2016, to around 3,000 publications in 2019 – an increase of more than 200 percent in the space of three years.
This activity is being fuelled by growing cross-sector interest in the potential of AI technologies to drive value by improving business’ process and workflow efficiency.
For electronics manufacturers and other OEMs, the main area of potential lies in automation, which is already possible to a limited degree using traditional computers. Now, with the rise of AI and the application of machine learning, it has become possible to automate much more; generating far greater enterprise value. While much progress has been made to date in the development of algorithms and their application to solve real-world problems, many experts believe we are only beginning to see the impact of this field of science.
To help strengthen the industry’s understanding of the direction of ML-related innovation, Withers & Rogers has developed a means of categorising individual patent applications according to their chief area of focus. There are essentially five categories of ML-related innovation – data processing methods; underlying algorithms; training methods; computing platforms; and the field of application.
Close analysis of the data by category indicates that the largest number of patent applications filed at the EPO, are directly linked to real-world applications, where a ML invention has been developed for use in solving a specific problem. This is unsurprising as a critical test of patent eligibility is an invention’s technicality and its ability to solve a technical problem.
One of the biggest areas of growth for innovation activity linked to machine learning in recent years is ML-specific hardware. Over 13 percent of all ML-related patents published by the EPO over the last three years have been directed to the ‘computing platform’ category. Whilst the majority of these patent applications have been made by big names such as Intel, Nvidia, and Qualcomm, a significant proportion are from newer players such as Graphcore, Kalray, and Cerebras. This high level of interest and investment in hardware, which is purpose built for machine learning approaches such as deep learning, is expected to continue.
Patent applications
In the case of the underlying algorithms, many of these are inventive in their own right and it is good to see that over 16 percent of all ML-related patent applications published by the EPO over the last three years fall into the ‘algorithm’ category. Importantly, these applications are not just being filed, the data shows that many of these applications are also being granted. This means it is possible for companies to obtain patent protection for innovative ML algorithms, provided the applications have been well drafted, address technical problems and provide technical details of the algorithm’s implementation.
Greater knowledge of the success of such patent applications should help to spur on innovation activity. New guidelines published by the European Patent Office (EPO) on 1st November 2019 will also help to smooth the way for innovators to secure patent protection for AI and ML inventions. The new ‘Guidelines for Examination’ contain specific advice about how AI and ML-related patent applications should be assessed by the EPO. This guidance has introduced greater clarity and will make it easier for innovators and their IP advisors to secure the commercial protection required when preparing to bring inventions to market.
In the past, a lack of certainty surrounding the EPO’s approach to the examination of AI and ML-related applications may have discouraged some innovators from seeking patent protection.
Based on the new guidance however, it is now clear that as long as innovators can demonstrate that their invention provides a technical solution to a technical problem, it will be assessed and examined in the same way as other inventions. This means that the valuable commercial protection that accompanies a granted patent is within their grasp.
One of the areas where the guidance has brought greater clarity is the description of how an AI or ML-related invention might make a technical contribution. They can achieve this either through their end use or by demonstrating that the underlying technology makes a technical contribution in its own right. The guidance takes the example of a neural network, which might qualify for patent protection when used to operate a heart-monitoring device capable of detecting an irregular heartbeat as this use demonstrates a technical contribution. Alternatively, a neural network could also achieve patent protection without reference to any real-world application, as long as its architecture is considered inventive and solves a technical problem.
Whilst the increased clarity surrounding the EPO’s approach to assessment will certainly help, the success or failure of individual patent applications in this field of R&D is likely to be determined by the way in which they have been drafted. For this reason, innovators should continue to take care to provide detailed evidence of their invention’s technical contribution. A good test to apply to every patent application is ‘is it inventive?’ Or in other words, ‘is this more than a routine development of the sector?’ Ultimately, the decision to grant the patent or not will rest on its ability to answer this question.
When preparing a patent application for a new algorithm for example, it is important to describe it in sufficient detail, so that a skilled person could implement it basedon the information provided in the patent specification and standard sector knowledge otherwise known as ‘common general knowledge’. Meeting this ‘sufficiency’ test is obviously more challenging when talking about algorithms and computational methods and it is important for the innovator to invest time in ensuring their patent attorney fully understands the technology at the outset.
Innovators should also consider the best approach to securing commercial protection. A granted patent automatically gives them a 20 year period of exclusivity in which to leverage their invention to the full and build up a market for their product. Other strategies are sometimes considered e.g. keeping the invention a trade secret. However, it is important that innovators understand the commercial risks they are taking.
A ‘trade secret’ is not an enforceable right in the same way as a patent, as no monopoly is provided, and a business would not be able to sue a competitor for infringement if the invention leaked to the marketplace. It is also entirely possible that a competitor develops the same invention at the same time and if they apply for patent protection first, this could block your route to market altogether.
In summary, there is a tremendous opportunity for innovators operating in the field of machine learning and the full potential of these inventions is only just being realised. With the right intellectual property strategy and commercial understanding, innovators can secure a stake in this rapidly evolving marketplace.
Author details: Harry Strange (left) and Karl Barnfather (right) are attorneys at European intellectual property firm, Withers & Rogers.
