The EU’s AI Act: Where Are We Now And What Could Its Impact Be On EU Companies?
6th July 2023
A fine wine requires a great grape and time to mature while stored in a stable environment. The controlled maturation of wine improves its quality and value which separates wine from most consumable goods. The process of making wine and building an artificial intelligence (AI) patent portfolio has some interesting similarities. A valuable (AI) patent portfolio requires insightful algorithms and a strategic long term vision. It requires an executive understanding that AI patent assets developed now may require time before widespread market adoption, similar to the aging process of fine wine.
When building a portfolio of AI patent assets, one should take a strategic approach with the vision that current legislation in the various jurisdictions may change. Further patent case law may evolve to catch up with the AI revolution, where technological advances are often years ahead of legislative change. A long term patent strategy in the AI field must take account of how the IP5 patent offices (USA, EPO, China, Korea, Japan) are currently examining AI patent applications and, more importantly, consider how they may examine AI patent applications in the future. Whether a patent strategy is focused on building a portfolio of AI patent assets for future cross-licensing deals or a bunker for defensive assets; the same approach applies. It is imperative that AI patent applications are carefully drafted to allow different generations of claimed AI inventions to be harvested from a single application when the time is right. The following discussion considers how such a strategy could be implemented at the EPO.
The current approach that the EPO uses to examine AI and machine learning patent applications follows a two stage process. The patent claims need to pass both stages before the EPO will allow an application to proceed to grant.
The first stage examines the claims for eligibility in view of Article 52(2)(c) and 3 EPC. Article 52(2) EPC specifies that “programs for computers” shall not be regarded as inventions. However, Article 52(3) introduces a grey area by limiting the exclusion of “computer programs” using the phrase “as such”. The phrase “as such” has been interpreted to mean that computer related inventions are considered patentable provided they have technical character. In essence the EPO requires that the claims include technical subject matter. Thus the first stage simply requires that the claimed subject matter includes technical subject matter. It is straightforward to meet the requirements of the first stage. One approach is to introduce the technical nature of the data such that the data represents real world parameters such as current, voltage, temperature, humidity, etc. These real world parameters have inherent technical character which are not abstract in nature or purely mathematical and as a consequence will automatically pass the first stage of the EPO examination process. However, it’s even easier to pass the first stage by simply incorporating any technical means in the claims. This can be achieved by framing the claim to recite “A computer implemented method…”
The second stage examines the claims with regard to novelty and inventive step. Traditionally, this stage often proved problematic for computer related inventions because the claims were categorised as mixed-type claims comprising technical features and so-called non-technical features. Non-technical features could not be used to support an inventive step and EPO Examiners often dismissed many claimed features for being non-technical which resulted in claims being refused for being obvious. The EPO has significantly changed its mind-set when examining computer related inventions, which is reflected in the revised Guidelines for Examination. The revised Guidelines were published in November 2018 and include a section relating to artificial intelligence and machine learning G-II 3.3.1. AI and machine learning inventions are based on algorithms and computer models which are of an abstract mathematical nature and therefore traditionally struggled to pass the eligibility test of the first stage. The revised Guidelines suggest that AI and machine learning inventions would be eligible for patentability if their application in a particular field contributes to the technical character of the claimed invention. Further these technical contributions should be considered by Examiners when assessing novelty and inventive step during the second stage. The following examples are referenced in the Guidelines G-II 3.3.1 “the use of a neural network in a heart-monitoring apparatus for the purpose of identifying irregular heartbeats makes a technical contribution. The classification of digital images, videos, audio or speech signals based on low-level features (e.g. edges or pixel attributes for images) are further typical technical applications of classification algorithms”. This mind-set change is a significant leap for the EPO, and was required in order to allow for the patenting of AI and machine learning inventions at the EPO. The EPO’s approach to examining mixed-type claims is documented in the Guidelines G-VII 5.4 which describes the problem and solution approach for claims comprising technical and non-technical features.
When drafting patent applications to be examined by the EPO, it is important that features which may first appear as being non-technical for relating to AI mathematical steps are described with concrete technical effects so that they can be considered as contributing to the technical character of the claimed invention. The contribution to technical character can be achieved in two ways. Firstly, the AI feature can be described in terms of how it is adapted to a specific technical implementation. In other words, the AI algorithm is adapted to a specific implementation motivated by technical considerations of the internal functions of the computer system that the AI algorithm is being executed on. It is important that the AI algorithm is directed to a specific implementation as the revised Guidelines make it clear that generic technical implementations will not be sufficient. Secondly, the AI feature can contribute to the technical character of the invention if it serves a specific technical purpose by technical application. In other words, the AI algorithm solves a technical problem in a specific technical field.
AI Divisional Filing Strategy
The EPO’s approach has materially changed G-II 3.3.1 in how Article 52(2) and 3 EPC are interpreted to allow AI and machine learning inventions to be patented at the EPO. It is clear based on the current approach that claims directed to AI algorithms should be 1) described with a specific implementation and 2) serve a specific technical purpose. The EPO has found a way to allow AI inventions to be patented without modifying the European Patent Convention. When drafting AI patent applications at least one of these specific criteria should be met in order to pass the second stage. The pendulum has swung far from the historical days when EPO examiners considered many computer related inventions as non-technical without considering all the steps of the algorithm. In present day the EPO has helpfully modified its Examination Guidelines to give practical advice detailing how AI and machine learning inventions can be patented. This change has no doubt been welcomed by innovators in the AI space, and hopefully this positive momentum will continue. Strategic innovators should draft patent applications that both meet the current specific requirements at the EPO while also having sufficient and flexibility to allow for future changes in the law.
The life of an EPO patent is 20 years from its filing date and over the life time substantial changes may occur to the environment perceived validity or enforceability of patents. Therefore, it is advisable that companies have a long term AI divisional filing strategy which would focus on protecting the generic AI algorithm implementation and its generic purpose. This approach requires the patent drafter to, on one hand; follow the EPO revised guidelines to allow a parent European Application to have its claims directed to the specific implementation and to serve a specific technical purpose. The patent drafter should also have an eye to the future by also providing a disclosure of a generic implementation and describe a generic technical purpose. This enables later filing of European divisional applications directed at broader protection than is currently being allowed at the EPO.
The scope of a European divisional application cannot extend beyond the content of the parent application as originally filed in accordance with Article 76 EPC. Therefore it is important that the subject matter of future divisional AI applications with broader scope than is currently allowed at the EPO is fully described within the parent application. This can be achieved by including a generic description of the AI algorithm which can provide ingredients for later filed divisional applications if the threshold to meet the second stage becomes more permissive. These broader-scoped divisional applications may provide targeted arrows against potential infringers in the future.
While the current approach by the EPO to look more favourably on AI and machine learning applications is welcomed, it does have certain limitations for enforcement as the specific requirements make it difficult to determine potential infringing algorithms in competitor products. Pursuing a strategy of broader-scoped divisional applications may also provide companies with valuable assets from a defensive perspective.