Protecting AI and machine-learning inventions

“Artificial Intelligence is a new digital frontier that will have a profound impact on the world, transforming the way we live and work” – Francis Gurry, WIPO director general

Technological advances in smart machines and computers are having a huge impact on the banking, business, communication, defence, education, internet, medical and transport sectors. They are becoming dependent on AI technology to collect and analyse historical data. Further, neural networks are being developed to use the processing power of computers to replicate the intelligence and learning capabilities of the human brain. Examples of these processes are self-driving vehicles, product recommendations on e-commerce websites and fraud detection.

Growing competition to develop AI has forced major jurisdictions to amend their patent laws – for example, it led India to provide statutory protection to software through the Copyright Act.

AI is the future

According to WIPO, machine learning is the most dominant feature of AI and is mentioned in more than 40% of all AI-related patents filed worldwide, with a very high growth rate of 28% between 2013 and 2016. Further, use of the term ‘neural network’ grew at a rate of 46% over the same period. The top three fields in which machine learning-related patent applications were filed were telecommunications (more than 51, 273), transport (50,861) and life and medical sciences fields (40,758). This shows that there is a future in AI and the protection of AI-based inventions is therefore of utmost importance to inventors and innovators around the globe.

‘Computer program per se’ versus protecting AI inventions

‘Computer program per se’ means a computer program without hardware implementation, which is considered to be a mathematical model, business method, presentation of information or a scheme. As AI falls into this category, it is therefore deemed unpatentable in all major jurisdictions. However, these inventions can be protected by linking the computer program to a hardware or computer network since it may include certain other things, which are ancillary to or developed through the program. Therefore, when drafting AI or machine learning-based inventions, it is worth showing real-world application rather than an abstract idea.

For instance, a machine learning model may be deemed a mathematical model or abstract idea and is therefore unpatentable. However, the model embedded in a self-driving vehicle for automatic detection of a route can be considered to provide a technical enhancement to self-driving vehicles and thus meets the patentability criteria. Inventors around the globe are encouraged to link AI with practical applications and innovate AI-assisted technology.

Since AI-based inventions can be categorised as abstract ideas, a solution-based approach should be kept in mind when drafting patent applications. Here are some tips:

• Link the solution to a practical application.
• Include a system architecture, which illustrates that all hardware elements are connected via a network, which can provide additional support for any objection on unpatentability during prosecution stages. Inclusion of the system architecture proves the hardware and/or the computer network link, thus making it patentable.
• Draft a system claim showing that limitations to hardware provide additional proof of the hardware limitations of the AI-based invention. The system claim may include a memory, an interface and a processor configured to implement an algorithm stored in the memory.

Advantages of patenting AI inventions rather than protection under copyrights

Patenting can be expensive and, while it has its advantages over copyright protection, AI-based computer programs can be protected under copyright law as they can be considered as literary works. Patenting an AI-based invention provides a broader scope of protection and covers the logic of the invention, while copyright merely protects the inventor against an entity copying the literary work (computer program). A patented technology is considered to have commercial value, particularly if it leads to acquisition or licensing deals (eg, Vertex.AI (which had a strong AI-related patent portfolio)), and was later acquired by Intel Labs, citing key benefits as IP rights owned by Vertex.AI.

Patenting activity for AI and machine learning-based applications has steadily increased in the last few years. In fact, the number of AI-based patent publications nearly doubled in 2018 and 2019 as compared with the previous years, as Table 1 illustrates.