Introduction

In the 2019 case of “Ferid Allani v Union of India”, the Delhi High Court clarified that patenting of artificial intelligence isn’t prohibited in India solely because it would fall under the ambit of computer programs of the Patents Act of 1970. It held that firstly, although the bar on patenting is in respect of “computer programs per se….” the bar is not applicable on all inventions based on computer programs, since this practice would disincentives innovation involving computer programs to a large extent and secondly that artificial intelligence systems must have a technical effect in order to be eligible for patent protection. Therefore, the primary question that arises is whether obtaining a patent would be the best method of protecting AI systems or whether there exist other methods of providing intellectual property benefits to products, processes, methods or systems of artificial intelligence.

Modus Operandi Of Ai Systems

In an AI system raw data is fed into the machine learning element which contains a set of algorithms that analyses data by learning from the data and using it to identify, discover, or carry out a particular function. One such example of a machine learning element is a neural network which contains multiple layers and is trained by inserting the input of a known function and teaching the machine learning element the correct output . This process of training is typically carried out by adjusting the coefficients of each of the nodes in a neural network to achieve a specific output (mathematical process).

Each of these specific functions are then carried out by different components of the system . However, the primary question that one must ask in regards to the patenting of AI systems is which component of the system would be eligible for patent protection. Generally patent applications attempt to patent the neural net/machine learning element, the entire AI system or the method through which raw data is pre-processed before its fed into the neural network. However, the issue that has been observed is that many artificial intelligence systems don’t satisfy the criteria laid down for patent applications such as novelty, enablement, etc.

Moreover, AI systems must also be able to demonstrate the tangible nature of an invention i.e., to be able to demonstrate that the software or system has a concrete output. For instance, the Indian patent application (IPA) 3323/CHENP/2012, for a system that monitors an energy load was granted only because it was able to demonstrate the tangible output of reducing the energy consumption of an air conditioning system .

Challenges In Obtaining Patents For Ai Systems

The challenges for obtaining patents for AI systems are five-fold: Firstly, one of the most basic requisites for obtaining patents for any invention is that it must be novel . However, this is a particularly difficult condition for AI systems to fulfil because most AI systems use a standard neural net which utilises standard configurations within its systems. Therefore, since most of the outputs generated or functions executed would be a product of a standard process, it doesn’t fulfil the condition of novelty.

Secondly, the requirement that an invention must not be obvious is a higher bar to adhere to in comparison to novelty for artificial intelligence systems. This is because the extent of AI’s abilities is capable of being ascertained to a large extent. Therefore, even though an idea such as assessing whether a particular tree possesses chestnut blight by running a picture of the tree’s bars through the neural net to carry out image recognition is novel, it is also one that is obvious since image recognition is a very common function of AI.

Thirdly, in regards to enablement i.e., the condition which stipulates that the inventor must be able to explain and describe the invention, its mode of functioning and its features for others to replicate it, in exchange for several years of monopoly over the invention, it is generally difficult to explain and describe an AI system. This is because a lot of what goes into an AI system is, in fact, hidden. It's hidden in the black box of the neural network making it difficult to discover from the outside which is why enablement rejections in artificial intelligence applications are not uncommon.

Moreover, due to the evolving nature of the machine learning element, it has been observed that the coefficients of the neural nodes wind up changing in order to adapt to increasing levels of sophisticated data which might fundamentally change the claims of the patent one initially filed for. Therefore, it becomes difficult to capture the specificity one claims in their invention and still try and leave it vague enough to allow room for growth.

Fourthly, as a patent holder it’s important to know whether a competitor is infringing your patent. However, due to the black box of the neural network it becomes difficult to know how a competing AI system is making its decisions or processing its data, and whether it’s following the same process or method for which one already possesses a patent.

Fifthly, in regards to the ownership of AI systems, it still remains unclear as to who the inventor would be i.e., whether it’s the researchers that created the neural network and fed it data or the neural network itself that would be the inventor . Additionally, due to the open-source nature of AI and the culture of sharing that permeates the community there is constant growth, development and modifications based on the existing modules or systems and methods of AI systems making it difficult to obtain patents for a single permutation. Therefore, by the time a particular application is approved various other forms of the same system would have already come into existence.

Alternative Forms Of Intellectual Property Protection For Ai Systems

As per recent developments many inventors are shifting their efforts from obtaining patents for their AI systems or processes to protecting the data they possess exclusively as trade secrets. Due to the immensely valuable nature of data, early leadership in a market will facilitate its access to extremely sensitive consumer data and data sets which can be used to develop its own systems or offer better services while gaining a competitive advantage over other players.

This competitive advantage was largely evinced in the case of the GAFA tech giants (Google, Amazon, Facebook and Apple) wherein these companies were able to aggregate rich data sets due to its early access in its respective markets, attain a dominant position and use the data to control key channels of distribution, carry out killer acquisitions, proliferate its own content to promote its businesses, etc . Therefore, by concealing the data that one has access to, the competitive and economic benefits gained may be able to compensate for the lack of intellectual property protection.

The second alternative that exists in terms of providing intellectual property protection to artificial intelligence processes, systems or methods is obtaining a license for the research code under copyright licenses. For instance, if an individual were to create an app to perform a particular function, it doesn’t automatically guarantee the issuance of a patent since the app is likely to use fairly routine features that perform in a computing environment. This is why various artificial intelligence inventors opt for licensing the codes through equity licences that implement machine learning techniques and AI, even if they were to forego protection of the underlying model.

Conclusion

Although in India, the patentability of AI inventions is not entirely prohibited, the above-mentioned challenges pose a detriment in obtaining patent protection for AI systems. However, in response to the Delhi High Court's Ferid Allani judgement, the IPO has begun reviewing Artificial Intelligence related innovations based on the “technical effect” created by such inventions, even though it has not yet provided clear instructions on the matter. The applications must explicitly state the technical contribution of the invention, and differentiate the invention from the cited prior art. This is done to avoid objections related to the possibility of patenting the subject matter in question. However, the area is still expanding in India and around the world, and it is possible that IPOs practise and methods may change over time as well.

Author: Silvia Tomy Simon,  A Student of Symbiosis International University, SLSH, in case of any query, contact us at Global Patent Filing or write back us via email at support@globalpatentfiling.com.

REFRENCES

Sas.com. (2022). Artificial Intelligence (AI) – What it is and why it matters. [online] Available at: https://www.sas.com/en_in/insights/analytics/what-is-artificial-intelligence [Accessed 31 Oct. 2022].

Builtin.com. (2021). What Is Artificial Intelligence (AI)? How Does AI Work? | Built In. [online] Available at: https://builtin.com/artificial-intelligence [Accessed 31 Oct. 2022].

Arora, N. and Dr. Joyita Deb (2022). [The Viewpoint] A future-proof Indian Patent Office? Patenting AI inventions in India. [online] Bar and Bench - Indian Legal news. Available at: https://www.barandbench.com/view-point/a-future-proof-ipo-patenting-ai-inventions-in-india [Accessed 31 Oct. 2022].

THE PATENTS ACT, 1970 § 2(1).

THE PATENTS ACT, 1970 § 2(ja).

TRADE-RELATED ASPECTS OF INTELLECTUAL PROPERTY RIGHTS, § 29(1).

Essenese Obhan (2020). Patenting AI Inventions: Best Practices from Around the World. [online] Mondaq.com. Available at: https://www.mondaq.com/india/patent/890652/patenting-ai-inventions-best-practices-from-around-the-world [Accessed 31 Oct. 2022].

Rutkin, A., Doty, J., Advisor, S., Hiller, A., Williams, J., Greengrass, D., Counsel, S., Reddick-Smith, S., Schwarz, D., Hariharan, A., Presley, J., Sharma, M., Robinson, M. and Hamedi, K. (2020). Investigation of competition in digital markets. [online] Available at: https://judiciary.house.gov/uploadedfiles/competition_in_digital_markets.pdf?utm_campaign=4493-519