Part II-Artificial Neural Networks: Are They Mathematical Methods or Computer Programmes, or Does it Even Matter?

Continuing the discussion on patentability of Artificial Neural Networks (ANN), in Part II of his two part guest post, Bharathwaj Ramakrishnan explores the situation in India vis a vis ANNs and application of Section 3(k). Bharathwaj is a student at the Rajiv Gandhi School of Intellectual Property Law, IIT Kharagpur and loves reading books and IP law. His previous posts can be accessed here, here and here.

Part II-Artificial Neural Networks: Are They Mathematical Methods or Computer Programmes, or Does it Even Matter?

By Bharathwaj Ramakrishnan

Section 3(k) and ANN, a Source of Confusion?

In the previous part we saw the technical definition of ANN and also the two cases one from EPO and other from UK Court of Appeal on patentability of ANN. It is only a matter of time that similar questions arise in India. When it does arise in the Indian context, it would involve an interpretation of Section 3(k) of the Patent Act (see this post, which covers and provides links to the extensive discussions on Section 3(k) that has happened in this blog). In the previous part, we saw there are good reasons for classifying ANNs as both mathematical methods and computer programmes. In India, this interpretive issue is further compounded by the existence of a class called algorithms, which is absent in Article 52 of the EPC and Section 1(2) (c) of the UK Patent Act. Thus, in this post, I wish to examine the possible paths available for the Indian courts in understanding ANN in the context of Section 3(k) and show post-Ericsson this classification issue might not be significant if all judicial classes receive per-se treatment. I will rely on two cases, Open TV vs Controller of Patents (see here and here) and Ericsson vs Lava and 2017 CRI guidelines, as they hold the clues as to how ANN might be understood regarding Section 3(k).

CRI Guidelines and ANNs, the Issues of Classification 

CRI guidelines 2017 have been a source of interpretation of Section 3(k) for the courts and hence it is a good starting point to see how ANNs might be understood under Section 3(k) (see Ericsson vs Lava where the Delhi High Court has used the CRI guidelines 2017 as a source of definition of terms in Section 3(k); but there have been issues raised regarding the binding nature of these guidelines; see here). CRI guidelines 2017 define Mathematical methods as follows “Mathematical methods like method of calculation, formulation of equations, finding square roots, cube roots and all other similar acts of mental skill are therefore, not patentable. Similarly, mere manipulations of abstract idea or solving purely mathematical problem/equations without specifying a practical application also attract the exclusion under this category.” Algorithms are defined as “a set of rules that must be followed when solving a particular problem.” For computer programmes, the definition provided in the Copyright Act under Section 2 (ffc) has been adopted. Based on the definitions provided in the CRI guidelines, and the technical definition provided in the previous post since the nodes in an ANN perform a series of mathematical operations, they could be classified as mathematical methods. Yet it must also be kept in mind that an ML engineer, when designing ANNs for practical purposes, would implement it on a computer with the help of a software program (Para 16 of Emotional Perception). Thus, these nodes can be virtual entities implemented on a computer and hence it must be clear that nothing is stopping the Indian courts from going the UK route. Likewise, one can also wonder whether an ANN would fall under algorithms. As the CRI guidelines concede (Para 4.4.4), it’s conceptually difficult to distinguish between algorithms and Computer programmes, and Delhi HC in Blackberry vs Controller grappled with such an issue recently. Yet the Court’s understanding that the technical aspect or technical feature of the invention was present in a set of instructions embodying if-then logic is slightly unsatisfactory as computer programmes, too, have if-then logic. But when an ANN is implemented, it must be a software ANN (thus written in code) or a Hardware ANN (“a physical box with electronics in it”, as the UK court put it). Thus, it is unlikely the courts will see them as algorithms as algorithms are a mere series of steps, and an invention incorporating an ANN would be implemented in hardware or software and would not fall under algorithms. Thus, it is safe to conclude that the Indian courts have to follow either the EPO route or the UK route. 

Are the Indian Courts Moving Towards an EPO-Style Jurisprudence?

 In Open TV vs Controller, the Delhi High Court observed that “The qualifier ‘as such’ thus applies to all categories of excluded inventions, including business methods in both the U.K. and Europe. Thus, the bar is not absolute, and patenting could be permissible if there is something more than the business method itself. However, in India, the phrase ‘per se’ does not qualify business methods.” Even though the reference is here to business methods, this absolute bar also applies to mathematical methods due to how Section 3(k) itself has been drafted. Thus, the Court has recognised that Section 3(k) has a relative bar in the context of computer programmes, meaning that an invention being classified as a computer programme itself would not make it ineligible subject matter. But then it is also reasonable to conclude on a plain reading of Section 3(k), along with the Court’s comparative analysis, that if an invention is classified as a mathematical method or an algorithm, it would mean it is barred by Section 3(k). 

Now, In Ericsson vs Lava, the Court made a very interesting observation which deserves to be quoted in full: “An invention that merely incorporates algorithms, sets of instructions, mathematical or business methods within a method or system, and satisfies all the criteria for patentability, is not inherently non-patentable.Therefore, what has to be seen is that if the algorithms are directed at enhancing the functionality of a system or a hardware component, the effect or the functionality derived by the system or the hardware component is a patentable subject matter. However, the algorithm itself is not a patentable subject matter.” 

It has already been noted here that there is a lack of clarity regarding the specific test that needs to be applied in Section 3(k) (see here). However, one can see a slight but significant shift in jurisprudence from Open TV to Ericsson. In Open TV, the Court recognised the difference between Article 52 of EPC and Section 3(k) and denied patents to business methods, but the Court in Ericsson made a slight and critical shift. It stated that if the algorithm or a mathematical method is directed at enhancing the functionality of a system or hardware and if the invention merely incorporates an algorithm or a mathematical method, then the inventions would be patentable. To make this shift clearer, the Court in Ericsson gave an example of a thermostat with an algorithm that allows it to adjust room temperature and how such an invention would be patentable. If there were an absolute bar, the fact that the thermostat used an algorithm to help with temperature regulation and the main technical contribution that arose from an algorithm would have barred the patent under section 3(k). Yet post-Ericsson that is not the case. Thus, by judicial interpretation, the Court has extended the relative bar to all the judicial classes of section 3(k) and has adopted an EPO-style reasoning.  

What Does This All Mean for ANNs Under Indian Patent Law

The Indian courts have good reasons to classify ANNs as either mathematical methods or computer programmes. Earlier classifying ANNs as a mathematical method might mean that there is an absolute bar, but post-Ericsson, it does not really matter as the Court, by interpretation, has extended a per-se treatment to all judicial classes in section 3(k). So, if ANNs are classified as Mathematical methods, the applicant can always argue that it improves the functionality of the system, and the claim itself is not directed at a mathematical method. Thus, the applicant can always argue that his system claim incorporates an ML system which uses an ANN which is implemented on a computer and written in a particular program and can provide better song recommendation to the user or can detect irregular heartbeats might clear the patentability bar set under Section 3(k).