Artificial Intelligence and Patents

Patents are only awarded for specific inventions, not for ideas, mathematical models, or algorithms. To qualify for a patent, no matter where in the world, an invention must be new, involve an inventive step, be tied to a particular subject matter, and have practical industrial use. An idea that lacks industrial applicability, is not completely original, doesn’t involve an inventive element, or isn’t linked to a particular use case cannot be patented. At the end of the day, Machine Learning (ML) or AI is just a “Software”, it is about writing software in a specific way. The problems with the software patenting regime are set to continue with AI software. There are two primary reasons for this: one is the policy decision of the state and the other is the legal uncertainty around software patents.

The patentability of AI technologies is a recurrent problem in the Indian Intellectual property arena. One of the evident issues in securing patents for AI technologies is its subject matter. As per section 3(k) of the Patents Act computer programs per se or algorithms are excluded from patent protection. This exclusion was made to encourage an open-source culture in these innovative fields. AI inventions are more or less considered to be algorithms that make use of pre-existing databases. With this view, patents for AI technologies are in clear violation of section 3(k). Additionally, these technologies also lack one of the basic tenets of patents, which is novelty. Many of these technologies are only a small addition to the existing algorithms and techniques. Hence the question of novelty and non-obviousness seems answered in negative. Just sprinkling the word “AI” over an existing technology cannot be a shortcut to gaining a new patent. But it is hardly surprising that many companies have succeeded in doing so.

Many scholars have openly expressed their disagreement with AI patents. Apart from the novelty argument, it is also observed that AI algorithms only list down a set of instructions to follow and therefore are void of any substance that can be patented. Moreover, such patents have enabled the tech giants to imprint their monopoly more evidently in the tech world. From a survey conducted in 2016, among the most important 1000 AI patents Google, IBM, and Microsoft own almost each of them. The small players are effectively restricted from adapting to this new technology and developing their own variants. In India, AI technologies should pass the Computer Related Inventions (CRI) test to get patented. However, there are challenges still in respect of inventorship, ownership, eligibility, inventive steps, and adequacy of disclosure.

The CRI guidelines of 2016 continue to be the basic principle that Indian patent office follows when they examine computer related inventions. When we look at the examination reports from 2019-20 there is a tendency among the observers to categorize these AI technologies as algorithms only, which are excluded under 3(K). But after Ferid Allani vs Union of India & Ors. there has been a comparatively relaxed approach towards software patents. The court observed that “innovation in the field of artificial intelligence, blockchain technologies, and other digital products would be based on computer programs, however, the same would not become non-patentable inventions – simply for that reason.” The change in the attitude of the IP office is evident through these words. AI inventions can be patented now by claiming a technical effect.

The issues of novelty and disclosure is however still present in most of these applications. Section 2(1) (ja) defines inventive step as "a feature of an invention that involves technical advance as compared to the existing knowledge or having economic significance or both and that makes the invention not obvious to a person skilled in the art." This existing knowledge base is larger than what we can imagine. So, the inventive step might also be very hard to establish in many cases. Also, the standard of “skilled person” becomes very complex in the case of AI as its growth is dynamic than ever. AI inventions are also reluctant to make proper disclosure of their algorithm and data sets as per section 10(4) of the Act. They are reluctant to reveal Data used because of the copyright issues associated with it. without the nature and size of the data, we have no way to know whether there is an innovative step in using that data. In short, beyond the software hurdle, AI patents have other issues to take care of.

When it comes to policy for patents, governments have encouraged the number of patents over the quality of patents. A case in point is the UGC's direction to many universities to set up IP cells and to encourage the filing of more patents. This has led to a huge increase in the patents filed including software patents. But more problematic is the patents filed by the big MNC which deprives other developers of not just the use of a specific code or set of code but also any code that achieves the same outcome. The patent application for software is usually generalized and widened in scope to cover a wide area beyond what the code implements. Now that AI/ML is the new kid in the block, there has been an increasing trend of filing for patents for AI software which is not allowed going by the letter of the law. This is complicated by the fact that the government is sending out contradictory signals on software patents. On the one hand, the government is encouraging the development of open-source software but, on the other hand, it is giving incentives to companies including startups to patent their software. The Scheme for Facilitating Startups Intellectual Property Protection (SIPP) encourages IP generation among software startups by offering a 50% rebate on filing fees for the patent. While it does not explicitly specify that it is encouraging software startups for patents, it is implicit that the government is facilitating patents for software through the enrolment of IP Mitras who can guide the process of patent grants. Granting patents to AI software would create monopolies in the industry and restrict the growth of many start-ups. This in turn would stifle innovation and creativity. Granting a patent for AI software would lead to increasing litigation. The continuing litigation will also mean that the company won't be able to use the software in question for a long time, potentially leading to a huge loss of business and perhaps even a shutdown of the company. The cash-starved start-ups cannot spend either time or money on it. Hence, it is important that the patent office put out clear guidelines on the patentability of AI software.

The intersection of Artificial Intelligence (AI) and intellectual property, particularly patents, poses complex challenges and raises several questions. While the current legal framework in India restricts patents for software and algorithms to promote an open-source culture, there is a growing trend of companies and innovators attempting to secure patents for AI-related technologies. This trend often leads to debates about novelty, inventiveness, and transparency. Despite court rulings like Ferid Allani vs. Union of India, which signaled a somewhat relaxed stance on software patenting by considering the technical effect of such inventions, many unresolved issues remain. The dual objectives of promoting open-source software while simultaneously encouraging IP protection through initiatives like the Startup India program can sometimes lead to mixed signals. For AI and software startups, this creates challenges as they navigate a landscape where the policy goals of openness and proprietary rights must be carefully balanced. Ultimately, a more refined and consistent approach is required to address these contradictions. This approach should consider the intent of the legislature, which did not want to grant a patent for any software subject. The approach should also consider the impact on innovation, competition, and the development of the AI ecosystem in India.