Anyone who has worked as a patent attorney and has drafted several applications would have come across this kind of question. I have drafted many software and technology-related inventions over time, and a similar difficulty is often noticed when it comes to AI-based inventions.
It is not always clear how novelty should be properly shown. It also becomes challenging to explain how the invention is not obvious. This is because many AI systems are built using existing models, known techniques, and already available data. In such situations, it often feels like everything already exists in some form.
Because of this, while drafting, it becomes difficult to clearly identify what exactly is new and what makes the invention different from what is already known. This issue is not limited to one or two cases it is something that is commonly faced by many drafters.
With this in mind, a simple but important question arises:
If it is so difficult to explain novelty and inventiveness in AI inventions, can such inventions really be patented?
This question becomes even more relevant for startup founders who are actively working on AI-based products. Many of them are already developing valuable solutions, but are often unsure whether their work can be protected, or how to go about it.
This article is the beginning of a 10-day series where these questions will be discussed in a simple and practical manner from understanding what an AI invention really is, to how it can be structured and protected under patent law.
For anyone building in AI, the idea is not just to innovate, but also to secure that innovation properly. Even if an AI system is already being developed or deployed, there may still be ways to identify what is protectable and how it can be claimed.
This series is intended to give a clear starting point.
Software Patents and its Patentability
Before moving to AI, it helps to look at how software has been treated under patent law.
Under the Indian Patents Act, 1970, especially Section 3(k), computer programs “per se” are not considered patentable. At first, this may appear as a complete restriction. However, in practice, the position is more balanced.
It has been understood over time, including from the approach of the Indian Patent Office, that:
- Software is not completely excluded
- Only software in isolation is not allowed
- If a software-based invention shows a technical effect or technical contribution to the technical problem, it may be considered for patent protection
In simple terms, it is not the code alone that matters, but what the code does in a practical, technical sense and how it is solving the technical problems.
Software Patents to AI Patents
AI can be seen as an extension of software, but with added complexity. Most AI inventions involve:
- Software components;
- Mathematical models; or
- Algorithms that learn from data.
Because of this, the same concerns that existed for software patents are now being raised again for AI. In fact, the concern is even stronger.
If software needed to show a technical effect to be patentable, the same requirement will naturally apply to AI inventions as well.
Finding the Invention in AI inventions
In traditional inventions, it is usually easier to identify the invention by identifying that machine has clear parts or that a process has defined steps; or a system has identifiable components. However, in AI-based inventions, this clarity is often missing.
While drafting, it may not be immediately clear whether the invention lies in the model or in the way the model is trained or in the data used or in the final application of the model.
This creates confusion not only for the drafter but also for the examiners.
Technical Effect Doctrine
To address the above issue, the idea of “technical effect” becomes important. It has been observed in practice, and also supported by decisions such as Ferid Allani v. Union of India that inventions involving software or digital technologies should not be rejected only because they are based on algorithms.
Instead, what is looked for is:
- Whether a technical problem is being solved; or
- Whether there is an improvement in performance or efficiency; or
- Whether the invention produces a practical, real-world result.
If such a technical effect can be shown, the chances of patentability improve.
Complexity in AI-related Inventions
Even with the above understanding of technical effect docyrine, AI introduces additional challenges such as:
- The system may generate outputs on its own
- The working of the model may not be fully explainable
- The results may depend heavily on the data used
- The core of the invention may still be based on mathematical methods
Because of these factors, it becomes more difficult to clearly present the invention in a way that fits within patent law requirements.
What Question Should We Ask?
Taking all of the above together, the main issue can be framed simply:
Can AI inventions be patented, or are they only advanced forms of algorithms and mathematical methods?
At present, there is no absolute bar against AI inventions. However, they cannot be claimed in an abstract manner. If the invention is presented only as a model or algorithm, it may not be allowed. If it is presented as a solution to a technical problem with a clear practical effect, it may be considered.
Therefore:
AI is not entirely new, and yet it does not fit neatly into old categories.
It stands somewhere between software and innovation, between logic and learning.
And so, the question remains open:
When an invention is built on learning systems rather than fixed instructions, how should patent law treat it?
With these questions still open, the next part of this series will move one step deeper into the subject. The focus will be on understanding the structure of an AI invention by breaking it down into its basic elements such as data, model, training, and application.
This is important because, in the context of AI patentability, protection does not begin with drafting it begins with clarity. Once it is properly understood what exactly forms the core of the invention, it becomes much easier to identify what can be protected and how it should be approached in a patent application.
*The author is a practicing advocate and IP attorney who regularly advises on intellectual property and technology matters. The views expressed in this article are personal and are intended for informational purposes only, and should not be construed as legal advice.