The Enlarged Board of Appeal (EBA) decision in G 2/21 related to the evidence requirement for a purported technical effect relied on for inventive step. G 2/21 arose from a case in the biotech field. The referring decision and much of the surrounding commentary to G 2/21 also focused on the terminology of plausibility, which is also often used in the biotech field. However, the decision of the EBA in G 2/21, and how it is eventually interpreted by the Boards of Appeal, may effect patents in any technical field in which experimental evidence is required. The Board of Appeal in T 2803/18, in particular, highlights how G 2/21 may be relevant to inventions in the field of artificial intelligence and machine learning.
Experimental evidence and the problem solution approach
The referral in G 2/21 related to the evidence requirements for inventive step. The problem solution approach to inventive step adopted by the EPO relies on identification of the “closest prior art” with respect to the claimed invention. This creates two problems for applicants in cases where a purported technical effect requires experimental evidence:
First, the document representing the “closest prior art” may be obscure and not be known to an applicant at the filing date. The applicant may thus not include comparative data in the application as filed demonstrating the superiority of the invention over the “closest prior art” later identified by the patent office or an opponent.
Second, negative data concerning the technical effect may arise after the filing date. For example, an opponent may submit new data showing that the invention does not have the purported technical effect over the full scope of the claim. The existence of such data may require the patentee to amend their inventive step argument to one reliant on a narrower technical effect, for which additional evidence may be required. This was the case in G 2/21 (EP 2484209).
In both of these situations, the problem faced by the patentee is a need for experimental evidence that only becomes apparent after the filing date of the application.
G 2/21 and its interpretation: Beyond biotech?
A critical question in European patent law is therefore the evidence and disclosure requirement for a purported technical effect. Is it fair to require the applicant to include comparative data in the application as filed with respect to the many and varied disclosures that may be later considered “the closest prior art”? Considering these issues in G 2/21 the EBA found that a purported technical effect relied on for inventive step must be “encompassed and embodied” by the application as filed. We await the written decision of the referring Board of Appeal in this case on how this general guidance should be interpreted.
The “purported technical effect” problem encapsulated by G 2/21 is particularly acute in fields where it is necessary to provide more experimental evidence of a purported technical effect, i.e. in technical areas where there is a higher level of doubt with respect to a purported technical. Inventions in the biotech field fall into this category. However, G 2/21 should not be considered restricted to the biotech field. Another field in which evidence of a purported technical effect is required is that of machine learning. In fact, G 2/21 is relevant for any field in which experimental evidence of a technical effect, particularly experimental evidence for technical effect over the whole scope of the claim, may be required.
The evidence requirement for machine learning inventions
In the field of machine learning, inventive step may be reliant on evidence that demonstrates the superiority of the claimed method over the methods disclosed in the prior art. For example, it may be necessary to demonstrate that the claimed method is faster or more efficient than the method disclosed in the prior art. Given that a claim to a machine learning method is likely to include some broadly claimed features, the EPO will also need to be persuaded that there is no verifiable doubt that the claimed method has the purported technical effect over the whole of the claim.
However, there is currently a dearth of Board of Appeal case law relating to the evidence requirement for machine learning inventions. This may be because of the current lack of litigation and/or enforcement of these types of invention. It is also possible that many machine learning inventions don’t reach the appeal stage because an objection of “lack of inventive step over the full scope of the claim” is often overcome by limiting the claim to include more technical features, as is the case in the biotech field. However, for the few cases that have reached the Board of Appeal, the similarity of the objections to those encountered in the biotech field is striking. One such case is the Board of Appeal decision in T 2803/18.
Providing evidence of a technical effect in ML (T 2803/18)
The patent in T 2803/18 related to a method for automatically detecting incontinence (EP2582341B1). Claim 1 specified a method for processing signals relating to the wetness of an absorbent article (e.g. clothing). The signals were automatically processed and represented using a machine learning algorithm. The Board of Appeal found the claim to lack inventive step in view of the closest prior art, US 2007/0270774. The Board of Appeal identified the sole distinguishing feature between the claimed invention and the closest prior art as being the mathematical method steps. The Patentee argued that these steps contributed to a technical effect and had therefore to be taken into consideration when examining for inventive step (G 1/19). The Patentee particularly argued that the mathematical methods steps specified by claim 1 had the technical effect of increasing the accuracy of the estimation of wetness in an absorbent article.
However, the Board of Appeal was not convinced that the technical effect of “increased accuracy” had in fact been achieved. In language that will be very familiar to a biotech patent attorney, the Board of Appeal highlighted the lack of direct comparative data demonstrating the superiority of the claimed method in view of the closest prior art:
“The [purported technical effect] would depend on many factors (size of training sets, number and type of elements/variables constituting the representative vectors, etc.), none of which are defined in claim 1, so that the results obtained by the claimed method are not necessarily more accurate than the results obtained by [the closest prior art]. The patent in suit does also not support such an alleged benefit by comparative data.” (r. 4.2, emphasis added)
In the absence of comparative data demonstrating the superiority of the technical effect of the method relied on inventive step, the Board of Appeal identified the problem solved by the claimed methods as merely the provision of an alternative method of processing sensor signals related to wetness of absorbent articles (r. 4.3). The Board of Appeal went on to find the claimed invention to lack inventive step, given that, according to the Board, the machine learning methods used in the method formed part of the common general knowledge and it would therefore have been obvious for a skilled person to use them in an alternative method of processing wetness signals. The Board of Appeal noted the lack of any “unexpected technical effect” associated with the claimed method (r. 4.3.1).
Prior art enablement
In another interesting similarity with the biotech field, a further contentious issue in T 2803/18 was whether or not the method disclosed in the prior art was enabled with respect to certain method steps of the claimed invention. According to established case law, only enabled disclosures are considered prior art (EPO Guidelines for Examination, G-IV-4). In T 2803/18, the patentee argued that the mathematical model provided in the prior art provided far too little technical detail in regard to processing and analysing sensor signals required for one particular step of the claimed method. The Board of Appeal rejected this argument on grounds that all these signal processing requirements needed to implement the method were based on well known principles of differential and integral calculus (r. 3.2.2). The claimed feature was thus found to be present in the prior art document.
Final thoughts
Reports from the recent oral proceedings in the referring case to G 2/21, indicate that the Board of Appeal ruled in favour of the patentee. Discussion at oral proceedings reportedly moved away from the terminology of “plausibility” and instead focused on whether the purported technical effect of the claims was “encompassed and embodied” by the application as filed. We now await the minutes of oral proceedings and the written decision of the Board. The decision in T 2803/18 highlights that it is not just those working in the biotech field that should be interested in the outcome. G 2/21 and its interpretation may have consequences in any technical field in which it is necessary to provide experimental evidence of a technical effect.
The EPO is not alone in connecting software inventions with the evidence standard for patentability. The overlap between the US enablement requirement and subject matter eligibility in the software field was discussed by Denis Crouch this week over on Patentlyo.
This post is based on a previous article on IPKat.
