The phrase “Artificial Intelligence” (AI) was coined soon after the mathematician-cryptologist Alan Turing devised the eponymous Turing Test to measure machine intelligence, about seven decades ago. The efforts of these pioneers are paying off now, as AI stands to transform how we engage with the world, revolutionizing solutions applied in every industry, from transport to healthcare.

Correspondingly, there has been an exponential increase in innovations around AI technology. But the protection of intellectual property (IP) of AI inventions has challenged well-established practices followed across the world. This article gives a perspective of the patentability of AI inventions in India.

AI systems use complex systems of mathematical models and/or algorithms to achieve their intended results and are effectively classified as Computer Related Inventions (CRIs) in India and treated as such by the Indian Patent Office (IPO). The IPO’s Guidelines for Examination of Computer-Related Inventions (CRIs) offer some direction to examiners in handling AI inventions, although some incongruities remain. Like any other invention, a CRI must fulfil the requirements of novelty, inventive step, and industrial application to be patentable under the Indian Patents Act, 1970. In India, the hurdles most likely to arise for AI inventions relate to subject matter patentability and inventive step, as well as sufficiency of disclosure.

Patentability

Most concerns pivot on and follow from the interpretation of Section 3(k) of the Patents Act, 1970, which excludes the patentability of computer programmes per se.

The Delhi High Court, in its 2019 decision in Ferid Allani v. Union of India and Ors, acknowledged that the future of innovation lay in AI, blockchain and related technologies, all of which are based on computer programmes, but they “would not become non-patentable inventions – simply for that reason.” It held that “technical effect” or “technical contribution” must be determined to decide patentability of CRIs, and that an invention merely being implemented using a computer program is no bar to its patentability. While the Guidelines do not mention the term, the decision states “the meaning of ‘technical effect’ is no longer in dispute owing to the development of judicial precedents and patent office practices internationally and in India”. The fact that AI inventions are not excluded from patentability if they have a technical effect or a technical contribution is in line with the position in the United Kingdom and the European Union.

A review of recent decisions shows that Controllers are now examining CRI applications for technical effect or contribution. In most cases, this can be demonstrated if the invention solves a technical problem in a field of technology and provides a technical solution to a technical problem (when the invention is directed to a specific technical implementation of AI).

For example, Indian Patent Application (IPA) 3323/CHENP/2012, for a system and method to model and monitor an energy load, was granted upon a demonstration of the technical effect of reducing the energy consumption of an air conditioning system. Similarly, an AI system for targeted content on display devices in real time, in IPA 202041003260, was granted because it was shown to be a technological advancement solving conventional technical problems. An AI-based assessment and tutoring method in 60/CHE/2010 providing contextual audio and video assistance to help users solve problems was granted because it used emotional elements, such as encouraging tones, to enforce learning.

In contrast, a CRI in IPA 8383/DELNP/2009 for a machine-to-machine communication platform that transfers medical data, was rejected. The IPO was not convinced by the technical effect provided by applicant, who claimed that it was an efficient and secure method of aggregating electronic patient records.

Inventive Step

The inventive step requirement in India is similar to that in other jurisdictions, and is determined by prior art: the patent examiner must be convinced that the invention is not obvious from the cited prior art.

Here, AI inventions are likely to encounter some hurdles. If a problem was solved previously by simple mathematical analysis, using an AI system instead may not satisfactorily demonstrate inventive step. AI inventions must include significantly more than mere AI implementation of a previously known concept or analysis. The Japanese Patent Office (JPO) AI guidelines provide illustrations to understand this better. Let us suppose that prevalent science allows the detection of cancer levels by measuring the levels of two markers, A and B in cells. Merely training and using an AI system to do the same would not amount to an inventive step. But, if a new component besides the markers were added to the training data, for example, cell size, this could clear the hurdle, as this was not part of the previous measurement criteria.

Sufficiency of disclosure

AI inventions are also likely to receive objections related to insufficient disclosure of the invention in the specification. Like its peers, the IPO also requires that an invention be described such that a person skilled in the art can replicate the invention. AI inventions are inherently complex and relevant pieces of information are more likely to be left out. To pre-empt objections, the specification must describe the invention’s essential features in as much detail as necessary. For instance, terms like ‘neural network’, ‘reasoning engine’, and ‘support vector machine’ may be considered as abstract models or algorithms; therefore it is necessary to define these terms in the specification, in relation to their specific technical use in an invention.

In other cases, explaining the algorithm in pseudocode may demonstrate how the algorithm initiates the intended result. This is particularly useful when differentiating an invention from prior art or showing linkages between the mathematics and the technical effect. Another useful practice is to include details of training data used to train the algorithm and specify the correlation between multiple types of data in the specification.

Clearly, there is no absolute bar on the patentability of AI inventions in India. Although the IPO is yet to provide specific guidelines on the issue, it has started examining AI-related inventions based on the “technical effect” caused by such inventions, following the Delhi High Court’s 2019 decision in Ferid Allani. In anticipation and to avoid objections related to subject matter patentability, applications for AI inventions must clearly indicate the technical effect or technical contribution of the invention, defining general terms in relation to specific technical effect, and preferably distinguishing the invention from the cited prior art. The domain is still growing in India, though, and it is likely that practice and procedure at the IPO will also evolve with time.

Neha Arora is a Senior Associate and Dr. Joyita Deb is an Associate at Obhan & Associates.