Skip to main content

A guide to protecting AI and machine learning inventions

(Image credit: Image Credit: PHOTOCREO Michal Bednarek / Shutterstock)

Securing patents for inventions that use artificial intelligence (AI) and machine learning can be challenging for innovators of these ground-breaking technologies, which attempt to use the processing power of computers to replicate the intelligence and learning capabilities of humans. Without patent or other intellectual property protection, they may be unable to commercialise their inventions, which could undermine investment in this dynamic field of research and development.

To clear the way for innovators, the European Patent Office has recently amended its ‘Guidelines for Examination’ by including a new section containing advice about how patents related to AI and machine learning technologies should be assessed. The guidance clarifies that whilst algorithms are regarded as ‘computational’ and abstract in nature, which means they are not patentable per se, once applied to a technical problem they may become eligible for patent protection. Beneficially, the approach outlined in the guidance is similar to that currently used to assess the patentability of computer-implemented inventions.

To clarify, one of the keys to patentability lies in an invention’s ‘technical effect’. If an AI or machine learning invention is shown to have an effect in a real-world application, it is likely to be deemed patentable under the European Patent Convention. The recently issued guidelines provide a useful example to help illustrate this. The example given is that of a neural network, which in itself may not be patentable as it would be deemed a mathematical method and would therefore be excluded from patentability. However, when the same neural network is embedded in heart monitoring equipment used to detect an irregular heartbeat, it is then considered to make a technical contribution and so could then meet the eligibility criteria.

For innovators, the publication of these new guidelines is both helpful and timely. First and foremost, they confirm that it is possible to secure patent protection - giving the patent owner a 20-year period of exclusivity during which to exploit their invention commercially. In fast-developing markets where innovation activity is rife, IP ownership could prove incredibly valuable if the technology becomes a standard feature, which other innovators wish to build on or develop further. In these circumstances, the technology could be licensed by the IP owner to third parties in exchange for royalty payments. From an industry perspective, the rapid growth in AI and machine learning innovation also means inventors are becoming more aware of the commercial value of their inventions and the benefits that patent protection can bring.

A recent analysis of global patent filings in the field of AI and ML, carried out by Withers & Rogers, confirms that innovation activity is expanding rapidly, and this rate of expansion is expected to continue. Between 2008 and 2018, the number of patent publications which are directed to AI or machine learning related subject matter has risen over 250 per cent. Furthermore, the number of AI or machine learning patents filed internationally has increased almost six-fold between 2006 and 2016, the latest year from which reliable patent filing numbers can be obtained.

Patent protection

As well as there being evidence of more global patent filings, the AI and machine learning technologies featured in these applications have become increasingly sophisticated. To illustrate this, a recently published patent highlights technology capable of generating audio using a convolutional neural network. Preliminary examination of the patent carried out by the EPO indicates that the invention does indeed meet the described patentability criteria. The 'technical effect' produced by the patented invention is considered to be a reduction in the computational requirement to generate waveform data compared to existing methods. This should be reassuring for both innovators and practitioners, as it clearly shows that the guidelines are being followed by the examiners at the EPO.

Further indication that proper consideration is being given to the new guidelines is evident with a recently granted European patent highlighting the use of deep learning for bone segmentation and removal in Computed Tomography Angiography imaging. Despite the patent being initially considered not to meet the requirements for patentability by the EPO, the applicant was able to overturn this opinion by successfully arguing that the invention provided a method for reliable and precise bone removal in a 3D medical image.

Despite growing interest in intellectual property rights, there are some pockets of innovation activity involved in the application of machine learning specifically, where innovators may be reluctant to pursue patent protection. For example, the owner of key data sets used for training a machine learning model may be unwilling to make these public. Instead, it may still be possible to protect these intellectual property assets as ‘trade secrets’ and advice should be sought about how to achieve this.

As these fields of research mature and attract new entrants, a patented technology could have considerable commercial value, particularly if it leads to acquisition or licensing deals with other innovators further down the line. As an example, Vertex.AI (formerly 1026 labs), a company focussed on enabling deep learning architectures to be deployed on different platforms, had a small but strong patent portfolio covering their core innovations. In 2018, Vertex.AI was acquired by Intel Labs, the latter citing that IP rights owned by Vertex.AI were a key benefit of the acquisition.

More licensing deals and acquisitions will no doubt be possible in the future as the renaissance in the application of machine learning continues. In the 1980s, when neural networks were first conceived, it was difficult to train and use them, due to limited computing power. Nowadays, things have become much easier thanks in part to the increased use of Graphical Processing Units (GPUs) which are capable of training models faster than traditional CPU-based architectures.

From an intellectual property perspective, there is still a significant opportunity for technology businesses to develop AI and machine learning inventions that will shape our way of life in the future. For these innovators, understanding how and when to seek patent protection and the geographies where this might be required, could be critical to secure their long-term profitability.

Stuart Latham, partner and patent attorney, Withers & Rogers (opens in new tab)
Harry Strange, Withers & Rogers

Image Credit: PHOTOCREO Michal Bednarek / Shutterstock

Harry Strange is a patent attorney at intellectual property firm, Withers & Rogers. The firm’s Electronics, Computing and Physics Group is one of the largest practice groups of its kind in Europe.