In this series of articles, we will review some past EPO decisions on AI patents in which sufficiency of disclosure played a role. Commentary will be published every Wednesday over the next weeks.

 

Sufficiency

 

Before we begin discussing the decisions, we briefly recall the general rules on sufficiency. As for all other inventions, sufficiency of disclosure is regulated by Art. 83 EPC that requires

 

“The European patent application shall disclose the invention in a manner sufficiently clear and complete for it to be carried out by a person skilled in the art.”

 

In the field of AI, the skilled person can be assumed to be a software engineer with experience in the implementation of AI. In particular, the skilled person can implement an AI using the known libraries, such as PyTorch or TensorFlow. The skilled person is also aware of the publications and picks up suggestions from the well-known publications, which are found in the documentation, and mostly on arXiv when it comes to core AI, and more in the corresponding scientific journals when it comes to applications of AI.

 

As always, it is the invention as defined in the claims (Art. 84 1^st sentence EPC) that sets the bar and the application in toto that is measured by it – the invention must be disclosed by the application.

 

In the case of a claim to a method of applying an artificial neural network, this means that the application must at least describe how to obtain a suitable neural network in the first place, as far as this is not known to the skilled person anyway. Since the term “artificial neural network” implies that implies that its weights are determined by learning (T 161/18 No 3.5), the application must indicate what data can be used for training. However, it has been subject to debate in some decisions what kind of disclosure is needed – disclosure may be implicit or explicit and must always be seen from the point of view of the skilled person whose technical knowledge may help fill in the missing parts.

 

What is to be learned from previous decisions

 

As repeatedly decided by the Boards of Appeal, the EPO has no doctrine of binding precedent (J 9/97 No 6, T 1212/97 No 10, see also R 14/11 No 2.9.1). To add to the irony of citing precedent to prove the inapplicability of precedent, it should be remarked that the EPO has titled its White Book the “Case Law of the Boards of Appeal” only in English, but „Rechtsprechung der Beschwerdekammern“ and « La Jurisprudence des Chambres de recours » in its other two official languages. This being said, the EPO appeal decisions are not the law of the land and should therefore be seen as an instruction on how the Boards' members think, with the general caveats that can be summarized as follows:

 

AI is a fast-developing field.

A decision (like T 161/18) that relates to an invention with priority in November 2005 may still be interesting, but only when taking into account that the general technical knowledge of the skilled person in the field of AI today has very little resemblance to that of the skilled person back then.

 

Obiter Dicta.

If the Boards were operating in a fully efficient way when refusing an application, they would cite only the best ground for refusal without commenting on others. Further reasoning is required only as far as the parties have submitted arguments, in view of their right to be heard. However, this does not mean that the Board must give a decisive opinion on sufficiency if it can refuse the application based on lack of inventive step. Lack of decision on the sufficiency question is not a fundamental procedural violation if the decision is correct for other reasons, see the White Book, Section  III.B 2.3.1. However, decisions are full of obiter dicta, which are not always fully thought through, in particular in those cases where they do not address any criticism by a party to the proceedings. Examples are cases where an application is refused both for lack of inventive step and sufficiency although one ground of refusal would have been enough.

 

Requirements for sufficiency.

If a Board denies sufficiency of disclosure by determining that none of a plurality of aspects of the invention is described (e.g. neither the structure nor the training data nor the training method), this does not necessarily mean that every application must contain all those things. What follows is that it must describe at least one of the aspects, and possibly also other aspects. The Board would perhaps not have made the effort to check all of them if they could have refused the application with just one of them.

 

Requests, Facts, and Arguments.

Reasons for a decision must also be seen in the context of the application documents and the requests and arguments filed by the parties. The Boards only need to decide on the requests as filed, and some objections made, or reasons given, only make sense when seen in context. Moreover, a decision in ex-parte proceedings is usually based on a discussion between the applicant, which argues for grant of the patent, and the EPO, which has to make a neutral decision in the end but tends to find arguments against the patent as it represents the public.