The EPO Board of Appeal decision in T 0687/22 confirms beyond doubt the relevance of G 2/21 to software inventions. The decision in T 0687/22 links the case law from G 1/19 and G 2/21 to highlight the importance of establishing a credible technical effect of software invention. The Patentee in the case made several attempts to formulate an objective technical problem solved by the invention based on several suggested technical effects. The Board of Appeal, citing both G 2/21 and G 1/19, was not convinced that the necessary criteria for relying on these purported technical effects were met.
Disclosure of a technical effect (G 2/21)
In G 2/21 the EBA found that post-filed evidence for technical effects can be used if "the skilled person, having the common general knowledge in mind, and based on the application as originally filed, would derive said effect as being encompassed by the technical teaching and embodied by the same originally disclosed invention." (Headnote II). Subsequent Board of Appeal decisions have found that, following G 2/21, it is not necessary for applications to explicitly mention the technical effect relied on for inventive step or to provide experimental evidence in support of a technical effect.
 |
| The broad applicability of G 2/21 |
According to T 0116/18, the technical effect relied on for inventive step only needs to be "conceptually comprised by the broadest technical teaching" and recognisable as "necessarily relevant to the claimed subject-matter." T 2465/19 found that “the technical effect of an invention over the closest prior art need not be explicitly stated in the application, as long as it is derivable from the original application, in particular since the closest prior art may not have been known to the applicant when drafting it”. T 1994/22 nonetheless clarified that vague statements like providing "a quality product" are insufficient to support specific undisclosed technical effects (IPKat).
The applicability of G 2/21 to software inventions
Many of the Board of Appeal decisions to apply G 2/21 have been in biotech and chemistry field, including cases relating to a second medical use (T 0258/21, IPKat), polymorphs (T 1994/22, IPKat), and cheese (T 0629/22, IPKat). However, G 2/21 has also been applied to other fields, including semiconductors (T 2465/19, IPKat). Indeed, G 2/21 is relevant in any case for which there is a question about the disclosure of a technical effect relied on for inventive step in the application as filed (IPKat).
There is a clear connection between G 2/21 and one of the central issues for software inventions, that of establishing features of the claimed invention that contribute a technical effect. According to the established case law in the software field, a feature may be only be relied on as contributing a technical solution to the objective problem to be solved in view of the closest prior art if the features provides a technical effect (see Case Law of the Boards of Appeal, I-D-9.2.8), see also the EBA decision in G 1/19). Given the importance of identifying a technical effect for the patentability of software inventions, the guidance from G 2/21 on the disclosure and evidence requirements for technical effects relied on for inventive step is highly relevant to the field.
Establishing a technical effect of a software invention (T 0687/22)
The patent at issue in T 0687/22 related to active occlusion cancellation in hearing devices (EP 3340653). T 0687/22 related to an appeal of the opposition division (OD) decision to maintain the patent in amended form. The Opponent appealed the OD decision, requesting that the patent be revoked.
The central issue in the case was whether the distinguishing features of the invention compared to the closest prior art provided a credible technical effect that solved an objective technical problem. Despite numerous attempts by the Patentee to identify a technical problem solved by the invention, including improved energy efficiency, circuit stability and flexibility, the Board of Appeal could not "identify any technical effect that can be credibly associated" with the distinguishing features (r. 2.3).
The Patentee submitted the alternative argument that the objective technical problem could be formulated as merely "providing for an alternative modification of the resulting signal based on the signal from the ear-canal microphone." However, the Board of Appeal rejected this effect for being too broad, noting that "a modification introduced solely for the sake of differentiation from the prior art simply means that, at most, the novelty requirement of Article 54 EPC is met" but does not guarantee compliance with Article 56 EPC (r. 2.3.4).
Given the absence of a credible technical effect that could be associated with the distinguishing features of the claimed invention, the Board of Appeal concluded that none of the requests were allowable under Article 56 EPC. As a result, the patent was revoked.
Relevance of G 2/21 to software inventions
Interestingly, in its reasoning in T 0687/22, the Board of Appeal made a direct reference to both G 2/21 and G 1/19:
[w]hen following the problem-solution approach to assess inventive step in the case in hand, a technical effect associated with the distinguishing feature(s) of the invention must be identified on the basis of the claim wording to establish a credible objective technical problem (see e.g. G 1/19, Reasons 82 and 124 as well as G 2/21, Reasons 25). If no such technical effect can be identified, there is not necessarily any technical problem to be solved (see e.g. G 1/19, Reasons 49).
In reason 25 of G 2/21, cited by the Board of Appeal, the EBA noted:
The technical problem must be derived from effects directly and causally related to the technical features of the claimed invention. An effect could not be validly used in the formulation of the technical problem if the effect required additional information not at the disposal of the skilled person even after taking into account the content of the application in question.
T 0687/22 thus confirms the clear relevance of G 2/21 to software inventions, and specifically to the criteria for establishing a technical effect of the distinguishing features of the invention in the application as filed.
A possible avenue for the Patentee to rescue the case might have been to provide some post-published evidence of their purported technical effects. The Board of Appeal would then have had to assess whether the criteria of G 2/21 were met, namely whether the technical effect relied on was encompassed and embodied by the application as filed. However, no post-published evidence was submitted, and so this issue was not considered.
Final thoughts
While the focus of commentary about G 2/21 has been on its relevance to biotech and chemistry cases, its principles are equally applicable to software inventions. T 0687/22 confirms the relevance of G 2/21 to software inventions, and particularly the need to both establish a credible technical effect for your invention that is identifiable from the application as filed.
G 2/21 was not focused on the concept of plausibility, given that it was primarily concerned with the disclosure requirements for a technical effect relied on for inventive step. However, the focus of the decision in T 0687/22 on the credibility of the alleged effects is also a reminder that the concept of plausibility/credibility in inventive step is highly applicable to software inventions. More on this to follow!
Further reading
Reviewed by Dr Rose Hughes
on
Wednesday, March 26, 2025
Rating:
Follow the IPKat on LinkedIn 
![[GuestPost] G1/24: Tuning in! A take on the state of proceedings before oral proceedings](https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEjZhEivE5bp7QOwZsyZXAXbVNYSmLjUthkB2Q7fm1_dpB97u5lIQeyWT9ZadUTAH3Z-hXn13VpW4vBDRPx9emCnoDV6tbUTkyvfmqPv1nNInL8XMdrAtSZ2hcRQr2LjxKovC9wTk_XyZxQ0CtX1MUrO_Muz3OJ4ld8AftymsdUmKD7xNksYMwk6/s150/Picture%201.png)
![[Guest post] ‘Ghiblification’ and the Moral Wrongs of U.S. Copyright Law](https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEhxl1BQBAW3Y-asjb1xXB9eA4DYy77fky6WgR-prC-_6DeBbDqOgCUDWyiz0Q3B23MWWAXnkbS2H2js7OUwA0JQXAHmsyVFgGIHeJz7zJ791vTzOD-4SJqWFIuywFXQyd3ahybbdZd4e8IEVfcNqctvyR8lumv_Gix6Tsw5cSvbHpTI1nwvztDuAQ/s150/IMG_2179.HEIC)
![[Guest book review] The Handbook of Fashion Law (with a discount code)](https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEgB4h2AdqJKwq9O3Ft4Mb7C39tv_NeFpkzrOfvhIsuwAkM_ops2Hgj7fdwzq_TQsjQDvQrQa-yyC9Q9pNiugseXRlUaMdsr_cmYUbh9lH8HDECMCbsTuNboVgpafyEhkgDkVS6ruHkuz8Sx0QVGI_1S8R9kbsHdNIYrRjqhyphenhyphen010_txjJUYvlZOtWA/s150/Fashion%20Law%20Book%20Bicture.jpg)
Posts
The title of the blog entry appears somehow misleading.
ReplyDeleteIn T 687/22 one reads: “a technical effect associated with the distinguishing feature(s) of the invention must be identified on the basis of the claim wording to establish a credible objective technical problem (see e.g. G 1/19, Reasons 82 and 124 as well as G 2/21, Reasons 25). If no such technical effect can be identified, there is not necessarily any technical problem to be solved (see e.g. G 1/19, Reasons 49)”
G 2/21, Reasons 25 reads
The technical problem must be derived from effects directly and causally related to the technical features of the claimed invention. An effect could not be validly used in the formulation of the technical problem if the effect required additional information not at the disposal of the skilled person even after taking into account the content of the application in question (see CLB, 10th edition, I.D.4.1, and the decisions therein).
The passage quoted in G 2/21, Reasons 25, deals with a key aspect of the problem-solution approach: the determination of the effect of the difference with the closest prior art, which ought to be obtainable over the whole scope claimed. The same applies to passages in G 1/19, Reasons 82, 124 and 49. But this is nothing new under the sun.
The quoted passages have nothing to do with what was at the basis of G 2/21: the possibility to bring in post-published evidence to justify a technical effect for inventive step if the skilled person, having the common general knowledge in mind, and based on the application as originally filed, would derive said effect as being encompassed by the technical teaching and embodied by the same originally disclosed invention..
It is difficult to see how post published evidence could be used to support a technical effect in inventions belonging to other parts of technique than chemistry or biotech, i.e. in electric and mechanic inventions, and especially software inventions.
A mechanical machine is a set of elementary mechanical parts and an electronic circuit is a set of elementary electronic components. Software inventions are merely a specific form of electronic circuitry.
Once put together the elementary features have a specific way of cooperating and it is difficult to get anything else out of them, unless you change the arrangement of the elementary features, which then ends with a different object.
In other words, if the effect requires additional information not at the disposal of the skilled person even after taking into account the content of the application in question, then it appears that there is not a problem of inventive step, but of sufficiency. In this respect, I refer to T 2001/12.
By the way, T 2803/18 does not even mention G 2/21, but only G 1/19!
Thank you for you comment - may I refer you also to T 2465/19 (https://ipkitten.blogspot.com/2023/10/reliance-on-silent-technical-effect.html). Any thoughts on this?
ReplyDeleteT 2803/18 predated G 2/21 and so obviously does not mention it. However, as explained in my commentary, the principles of G 2/21 are highly relevant to inventions in the ML field, in which data demonstrating advantages of a new algorithm over a (potentially previously unknown) prior art algorithm attempting to solve a similar problem, are often necessary (see https://ipkitten.blogspot.com/2023/08/the-relevance-of-g-221-to-machine.html)
ReplyDeleteThanks for taking time for your detailed reply.
ReplyDeleteThanks for drawing my attention to the fact that T 2803/18 predated G 2/21 and so obviously did not mention it. My apologies.
In all the decisions you quote, reference is made to the objective technical problem, but it is difficult to infer that post-published evidence could help to demonstrate inventive step.
In T 2465/19, relating to a semi-conductor device with very definite topology, I am at a loss to understand which type of post-published evidence could have been brought to the procedure.
There is no technical effect encompassed by the technical teaching and embodied by the same originally disclosed invention able to support inventive step.
T 2465/19, Reasons 5.3.1, end of the last §, reads:
“The Board notes that the technical effect of an invention over the closest prior art need not be explicitly stated in the application, as long as it is derivable from the original application, in particular since the closest prior art may not have been known to the applicant when drafting it (see also decision G 2/21, Headnote II.).
This is nothing new when assessing inventive step according to the problem-solution approach.
T 2803/22, Reasons 4.2, second §, reads:
“The accuracy 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 regression analysis, the resulting mathematical model and the threshold criteria applied in D2”.
When it comes to ML, which kind of data could demonstrate advantages of a new algorithm over a (potentially previously unknown) prior art algorithm attempting to solve a similar problem? The only possible data in ML appear to be further training data. Bringing post-published evidence in this respect, would not help much as it would indirectly show that the invention was not sufficiently disclosed.
In chemistry and biotech, G 2/21 could bring something, although this decision is not applied in a coherent and convergent manner by all boards. Just look at T 314/20 which even criticised T 116/18 following G 2/21!
You say that data demonstrating advantages of a new algorithm over a (potentially previously unknown) prior art algorithm attempting to solve a similar problem, are often necessary. This is not the case.
ReplyDeleteThe issue in T 2803/18 was not the lack of comparative data in the application. The problem was the claims were vague and the application lacked concrete implementation detail. No amount of comparative data would change that.
It would not even be possible to produce comparative data for the claims in T 2803/18. The claims are so broad that if you build a working embodiment to produce comparative data, the embodiment you build will necessarily be much narrower than the claims and so the data would be useless. The fundamental problem in T 2803/18 was the claims encompassed embodiments that provided an improvement but also embodiments that certainly did not provide an improvement. Again, no amount of comparative data would change that.
This is a general problem with trying to apply G2/21 in software/ML cases.
In response to: "When it comes to ML, which kind of data could demonstrate advantages of a new algorithm over a (potentially previously unknown) prior art algorithm attempting to solve a similar problem? The only possible data in ML appear to be further training data. Bringing post-published evidence in this respect, would not help much as it would indirectly show that the invention was not sufficiently disclosed."
DeleteAccording to Mr PatKat (my AI advisor): Almost every single machine learning paper and model is justified by demonstrating its advantages over prior art algorithms attempting to solve similar problems. The basic approach is to try the machine learning algorithm on the problem of interest, and see how well it does. For instance, if the algorithm is supposed to classify cats vs dogs, then you check how accurately it does that (i.e. does it classify an image of a cat as a cat? or does it classify an image of a cat as a dog?). You do this using a dataset that was not used during training, known as a "test dataset", and then compare its performance to previous algorithms (e.g. prior art algorithms). In fact, advances in "benchmarking" and "evals" (used to demonstrate one algorithm is better than another) are widely acknowledged to have driven much of the recent progress in machine learning (e.g. see https://medium.com/data-science/the-olympics-of-ai-benchmarking-machine-learning-systems-c4b2051fbd2b)
That you need a “test data set” to see whether the training has been correct does not appear comparable with bringing in post published evidence in the meaning of G 2/21. The “test data set” allows to check that the training was correct, as well as a comparison with state of the art algorithms, but nothing more. This is not what is at the core of G 2/21.
DeleteI have difficulties in seeing how a “test data set” can be used to define a technical effect, not directly disclosed in the application as filed but nevertheless encompassed by the technical teaching and embodied by the same originally disclosed invention.
In a presentation on the Case Law incorporated in the 2025 Guidelines, one decision, T 601/22, Reasons 8.2.5, was mentioned in which a post-published document (D7) in the meaning of G 2/21 was used by the applicant.
The board was of the opinion that the question whether the requirements of point 2. of the order of G 2/21 were met could be left unanswered. The board held that the alleged technical effect, even if it were evidenced by D7 and "encompassed by the technical teaching and embodied by the same originally disclosed invention", is not achieved over the whole scope of claim 1. Even there, G 2/21 was not helpful for the applicant.
I thus still have strong doubts that G 2/21 will be of much use in other areas than chemistry and biotech.
This is a really interesting discussion, and my thoughts are below.
DeleteWith AI/ML, any comparative data is tied to a specific set of weights and balances for a specific neural network. If a detailed method of training a neural network to obtain those is not disclosed on filing, the application is insufficient. Without that inventive step can’t even be established since the claim needs to be limited, at least indirectly, to the method of training the neural network.
Only and only then, might comparative data become relevant for inventive step.
I can only but agree with Anonymous of Wednesday, 2 April 2025 at 15:11:00 GMT+1.
DeleteIn order to be sufficiently disclosed a minimum of training data has to be disclosed when filing the application. Without this minimum, which may vary, sufficiency is indeed not given.
T 161/18 and T 1169/21 are prime examples of the lack of sufficiency in the absence of training data.
As data are very valuable goods, it cannot be expected that a great number of applications in AI/ML will be filed.
Any AI/ML system for which a patent is applied for, has to solve a technical problem by producing a technical effect going beyond the simulation's implementation on a computer, see headnote 1 of G 1/19.
A welcome section of decision T 687/22 is as follows :
ReplyDelete"The Patentee submitted the alternative argument that the objective technical problem could be formulated as merely "providing for an alternative modification of the resulting signal based on the signal from the ear-canal microphone." However, the Board of Appeal rejected this effect for being too broad, noting that "a modification introduced solely for the sake of differentiation from the prior art simply means that, at most, the novelty requirement of Article 54 EPC is met" but does not guarantee compliance with Article 56 EPC (r. 2.3.4). "
An "objective technical problem" has nothing technical if it is defined as the provision of an alternative solution. Such as a definition is only a negative, it is technically meaningless.