Issues surrounding patenting of inventions relating to artificial intelligence in the US and Europe


Financier Worldwide Magazine

March 2019 Issue

Arguably, artificial intelligence (AI) – that is, the use of tools to perform tasks previously done by the human brain – is as old as the abacus, or perhaps even older. However, it is only with the dramatic increase of computer computational power after 1990 that large-scale AI applications took off. For the most part, these applications employed two basic approaches: first, looking through large quantities of data for correlations where none had been previously detected, and second, ‘playing through’ all possible steps in a particular scenario to see which resulted in the best outcome. The first approach has been important in developing marketing techniques, such as, if there is a correlation between the purchase of three different types of product, to advertise the third of these products to those who have bought the other two and medical diagnosis – is there a hitherto unobserved correlation between particular symptoms and an underlying cause? The second approach led to computers being able to beat chess and Go world champions in their respective games, but also led computers to control self-driving cars and telecommunications systems.

Today, we are moving on beyond these basics, which could have been performed by humans if only we had big enough brains, to techniques such as machine learning, wherein combinations of algorithms and statistical analysis allow a computer to draw inferences without being given explicit instructions. Commonly, such algorithms are linked together in an artificial neural network. Such techniques have solved, and have the potential to solve not only problems that we are aware of, for example by providing voice recognition, computer vision and machine translations, but also to identify and address problems that we are unaware of as more data becomes subject to computer scrutiny.

How is the patent system to cope with this? As described in a January 2019 Technology Trends publication by the World Intellectual Property Organisation, a boom in the filing of patent applications for inventions related to AI commenced in about 2012 with IBM, Microsoft, Toshiba, Samsung and several Chinese institutions at the forefront. Such filings fall into two main categories: AI techniques themselves, primarily in the field of machine learning, but also in logic programming and fuzzy logic techniques, and applications of AI to real world functions.

Recent guidance from both the US Patent and Trademark Office (PTO) and the European Patent Office has addressed how patent applications in this field are to be examined going forward.

United States

On its face, US patent law should present no problem for patenting any type of invention relating to AI. It simply states: “Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.”

Although the courts had held that natural phenomena, laws of nature and abstract ideas were judicial exceptions to the broad generality of the definition, and there was some hesitation initially about the patenting of software-related inventions, until 2010, the courts tended to give statutory language a broad interpretation. Since then, in an overreaction to the grant of a large number of patents of uncertain merit during the dotcom boom, the limitation on patents for ‘abstract ideas’ has developed a life of its own, with the Supreme Court devising a two-part test to determine whether or not an invention was patent eligible in which one first had to determine whether the invention was ‘directed to’ a potentially excepted idea and then, if it fell into such a category, determine whether it could nevertheless be patented as being an application of the idea that was more than performance of “well understood, routine, [and] conventional activities previously known to the industry”. Unfortunately, the Supreme Court gave no clear guidance as to what types of ideas it regarded as ‘abstract’ and left this to the lower courts to work out. Over time, the Court of Appeals for the Federal Circuit has done this, leading the US PTO to conclude that it is possible to provide more useful guidance. Such guidance was issued on 7 January 2019, stating that normally ideas that were subject to the Supreme Court’s two-step test were limited to: (i) mathematical concepts, including mathematical relationships, mathematical formulas or equations, and mathematical calculations; (ii) certain methods of organising human activity, such as fundamental economic processes, commercial or legal interactions, managing personal behaviour or relationships or interactions between people; and (iii) mental processes, including concepts performed in the human mind, such as observations, evaluations, judgments and opinions.

If the invention relates to any of these, it must be analysed to determine whether the idea is integrated into a practical application of that idea that goes beyond simply saying ‘apply it’, but there is no need to consider whether the application is a conventional application of that type of idea. The guidance goes on to state that elements improving the functioning of a computer or other technology are to be regarded as patent eligible.

This guidance will be applicable to inventions relating to AI. There should be no problem with inventions relating to the application of AI to real world issues, except perhaps if they simply relate to applications of the results of correlation analysis to human activities. Difficulties may arise, however, with inventions relating to improved methods of machine learning or of implementing fuzzy logic in situations divorced from any specific application where courts will be left to consider whether such inventions are more analogous to ‘mental processes’ or the ‘improved functioning of a computer or other technology’. The best advice, therefore, is to try to make sure that any patent application where such issues may arise clearly sets out the real world, potential technical applications of the improvement being patented.

Machine learning is now moving into a realm where it is ‘unsupervised’ and could start to define new problems to be solved that have not been considered before. Questions then arise as to whether an invention wholly conceived of by a computer qualifies for protection. The statute provides that ‘whoever’ makes a patentable invention is entitled to a patent. The question therefore arises whether the word ‘whoever’ is necessarily restricted to humans. It has been argued that the ‘monkey-selfie’ case, relating to copyright, which held that a monkey has no standing to bring a claim of copyright infringement even if the monkey took the photograph in which copyright was claimed, indicates that a computer cannot be an inventor. That case, however, turned on previous decisions relating to the rights of animals and so may not be determinative in the present situation.

European Patent Office

The European Patent Convention, which provides the legal basis for the examination of patent applications by the European Patent Office for 38 European countries, both inside and outside the European Union, provides that “European patents shall be granted for any inventions, in all fields of technology, provided that they are new, involve an inventive step and are susceptible of industrial application”.

However, there are specific exclusions for certain types of inventions, including mathematical methods, methods for performing mental acts or doing business programmes for computers and presentations of information, although inventions relating to the application of any of these are patentable, as long as the invention provides a technical solution to a technical problem.

Late in 2018, the European Patent Office amended its examination guidelines to include a specific section on how these principles apply to inventions relating to AI.

The guidance notes that a mathematical method may contribute to the technical character of an invention, and contribute to a technical effect serving a technical purpose, by its application to a field of technology or by being adapted to a specific technical implementation. Similar to the situation in the US, the guidance notes that computational models and algorithms per se are of an abstract mathematical nature, irrespective of whether they can be ‘trained’, based on training data. However, where such training serves a technical purpose, the steps of generating the training set and training the classifier may contribute to the technical character of the invention. Similarly, the guidance notes that the application of computational models and algorithms to technical problems, such as the use of neural networks to identify irregular heartbeats or use of classification algorithms to classify aspects of data having technical applications, such as digital images or speech signals, can have a sufficient technical character to be patentable.

When the European Patent Convention was adopted, the countries forming the organisation kept to themselves the question of who should be regarded as an inventor. This issue remains one of national law. No European country has yet adopted a law that addresses the question of whether a computer can be an inventor. However, in the field of copyright, the UK’s Copyright, Designs and Patents Act of 1988 provides that for “a literary, dramatic, musical or artistic work which is computer-generated, the author shall be taken to be the person by whom the arrangements necessary for the creation of the work were undertaken”.


On both sides of the Atlantic, the use of AI to solve technical problems is accepted as being patentable. In Europe, there are restrictions on the patentability of the use of AI to solve non-technical problems and of development of new methods of machine learning or analysis divorced from a technical problem. Similar reservations exist in the US, but the case law providing for patent protection for elements resulting in improved functioning of a computer provides a basis for patenting, for example improved methods of organising neural networks or of effecting machine learning.


John Richards is of counsel at Ladas & Parry LLP. He can be contacted on +1 (212) 708 1915 or by email:

© Financier Worldwide


John Richards

Ladas & Parry LLP

©2001-2019 Financier Worldwide Ltd. All rights reserved.