Thursday - 28 February 2019

Considering Differing National Protection Regimes For AI-Related Inventions

The difficulties in obtaining protection for artificial intelligence (AI) and machine learning (ML)-related inventions, and the differences between the various approaches adopted by the national and regional offices, has been the focus of much discussion in recent years. However, as a European patent attorney sharing a stage with attorneys from around the world in a recent seminar series on the topic in Japan, what struck me was that despite the differences in legal framework, the resulting approaches to patent examination in this area are more aligned than we often think.

Focusing for example on the US where, unlike Europe, there is no statutory provision listing what is not considered to be an invention, the examination of AI and ML inventions is, nevertheless, rooted in subject matter eligibility requirements and what is, and is not, considered ‘abstract’. What is important for demonstrating that an AI or ML invention is significantly more than an abstract idea is the nature of the technological ‘improvement’ that the invention provides. However, it is not sufficient simply to explain the ‘improvement’ in the description – to obtain protection the technological ‘improvement’ should be recited in the claims.

In Europe, November 2018 saw the introduction of a new section to the EPO’s Guidelines for examination, providing guidance on when inventions related to AI and ML may be considered to have sufficient ‘technical effect’ to avoid being considered ‘abstract’. The new section treats AI- and ML-based computational models and algorithms as being per se of an abstract mathematical nature and hence as a sub-class of the statutory exclusion ‘mathematical methods’. Nevertheless, the application of these models to various fields of technology, such as image recognition or medical monitoring, can provide sufficient technical effect to avoid the exclusion. In essence the guidelines extend existing guidance that the fact that a ‘mathematical method’ serves a technical purpose is not, in itself, sufficient – to obtain protection the claim must be functionally limited to a specific technical purpose.

What was evident from the comments of the US contributors was that reciting the specific technical purpose in the European sense would, in general, meet these requirements in the US. It was also apparent that a similar approach could be applied in Japan and China. Accordingly, when preparing claims suitable for protecting AI and ML inventions throughout the world, while it is of course, important to take account of specific national differences in claim drafting practice, the overall approach to avoiding subject matter eligibility type issues is, nevertheless, similar.

What was also apparent, was that drafting AI- and ML-related applications with an eye for which part of the relevant patent office will likely be examining the application is a strategy that is common to both sides of the Atlantic.

From discussions with my US colleagues, for example, it is clear that the ‘Technology Center’ at the USPTO that is allocated a particular AI or ML-related patent application for examination, can have a significant impact both on the likelihood of receiving rejections of the claimed subject matter as being patent ineligible and on the overall allowance rate. The way in which a patent claim is drafted – even for the same subject matter – can affect the probability of being allocated to a particular Technology Center and hence the chances of having to respond to unfavourable patent eligibility rejections. For example, the rate of patent eligibility rejections could vary from less than 3% (with an 83% allowance rate) for one Technical Center to a patent eligibility rejection rate of around 25% (with a 68% allowance rate) for another.

Ironically, some US attorneys use a specialist AI tool to analyse prospective AI- or ML-related claim language before filing to provide a comparative indication of the relative probabilities of that claim language resulting in an application being allocated to different USPTO Technology Centers. The language can then be adjusted iteratively to maximise the probability of being allocated to the more favourable Technology Centers. This also has resonance for attorneys involved with protecting inventions at the borders of the statutory exclusions before the EPO where the subject matter classification attributed to a patent application can, anecdotally, also affect the prospects for successful grant. For patent applications falling in one particular subject matter classification, for example, an allowance rate of less than 10% has been reported.

Notwithstanding the care that is always necessary when drafting patent applications related to AI and ML, with the rapid growth of AI and ML technologies and their importance to society it seems likely that, over time, patent offices and legislators will develop a pragmatic approach to ensuring that those developing AI and ML-related innovations will be able to obtain fair protection for their efforts with a view to encouraging investment.

This article first appeared in the March 2019 edition of Intellectual Property Magazine. To read the full article, click here.

About the author

Jeremy Smith


Jeremy entered the patent profession from an industry background in semiconductor device design and development. During his time in industry, Jeremy was involved with thin film transistor technologies, analogue process and device development, and transient suppressor design. Jeremy is also a named inventor on a number of semiconductor device related patents.