As AI and machine learning advance, protecting innovations based on these technologies is essential to securing IP and capitalizing on efforts. For an AI invention to be patentable, it must have a “technical character,” meaning it must use technical means to solve a technical problem. A software module that applies a machine learning model is not inherently patentable unless it contributes to solving a technical problem in a novel and non-obvious way.
Table of contents:
- Patent law and developments
- Applied AI inventions
- Challenges in patenting core AI inventions
- Why protect your AI inventions?
- Contact the patent attorneys of V.O.
- Frequently asked questions about AI and patents
Patent law and developments
Legal frameworks around artificial intelligence (AI) and patents are evolving worldwide. As AI technology becomes embedded in various industries, patent offices are refining policies and criteria for intellectual property protections, which will likely continue to change over time.
European and U.S. patent frameworks
Both Europe and the U.S. require AI-related inventions to be directed at technical applications to be patentable. In Europe, AI itself (such as a new machine learning model) is considered a mathematical method and is generally excluded from patenting. However, if it solves a technical problem using technical means to provide a technical effect, the application of it may be patentable. Similarly, the U.S. Patent & Trademark Office (USPTO) evaluates AI inventions as computer-implemented inventions (software) and requires them to offer practical applications and technical improvements.
Despite these similarities, national and regional differences exist. For instance, while all countries require sufficient disclosure of the invention’s implementation, the European Patent Office (EPO) applies this criterion more strictly. Additionally, the U.S. tends to be more lenient in determining whether an AI-related invention falls within a sufficiently technical field.
Applied AI inventions
AI inventions must solve a technical problem using a technical solution. Non-technical applications, such as business methods or purely abstract ideas, do not qualify. Applied AI inventions use AI in specific technical applications, such as medical devices, 3D printing and autonomous vehicles. These inventions are more likely to be patentable as they solve concrete technical problems and produce tangible technical effects.
Fields generally considered technical include encryption technologies and telecommunication methods, while areas like data processing, administrative tasks, or information presentation methods are often considered non-technical. For instance, in Europe, using a large language model (LLM) for translation or a user interface enhancement is not considered technical, whereas in the U.S., such applications might be patentable under a broader interpretation of technical contribution.
To enhance patent eligibility, AI inventions must demonstrate a direct and clear contribution to a recognized field of technology. This not only increases the likelihood of securing patent protection but also strengthens the foundation for commercial use and enforcement against infringement.
Highlighting technical effect of AI inventions
Technical effect refers to identifiable improvements AI technology brings to existing technologies or processes. A strong technical effect must be clear, specific and result in tangible technological improvement. Examples include:
- AI optimizing battery efficiency in electric vehicles through optimized charging algorithms.
- AI reducing energy consumption in industrial processes.
- AI improving medical imaging accuracy by identifying patterns undetectable to humans.
Patent applications should clearly outline how an AI invention interacts with the physical world in an innovative way, ensuring the description aligns with patent requirements.
Challenges in patenting core AI inventions
Another class of AI related inventions is directed at improvements concerning the AI itself. Think of its internal working, the manner of training, feedback process or specifics of input or output signals. This class of inventions may be referred to as core AI inventions, and focuses on the fundamental aspects of artificial intelligence itself. These might include new algorithms, machine learning models, or unique computational processes.
These “core AI” inventions face greater challenges in obtaining patents. European patent law excludes most of these as mathematical in nature. In the U.S., improvements to AI algorithms are often classified as abstract ideas and thus not patentable. Abstract ideas are deemed to form one of “the basic tools of scientific and technological work,” and are thus excluded from patentability*.
When core AI inventions may be patentable
A core AI invention may be patentable if it demonstrates a direct technical improvement in a system’s functionality. For example, an AI algorithm that optimizes storage or processing power by adapting to a computer system’s architecture might qualify. In such cases, the patent application must detail how the interaction between the AI model and the system results in a technical improvement.
Meeting patent disclosure requirements
A fundamental patent requirement is that an invention must be disclosed in enough detail for a person skilled in the art to be able to reproduce it. For AI-related inventions, this means providing:
- The type of machine learning model used.
- The training method and required data characteristics.
- Input and output signal specifications.
- The technical architecture of the invention.
Since machine learning models function as “black boxes,” simply describing the model is not enough. Patent applications must specify how the model is trained and applied, ensuring the AI system achieves the intended technical effect. It is not required to provide the training data or ground truth data itself, nor samples thereof.
Why protect your AI inventions?
Securing patents for AI innovations offers significant advantages. For one, patents provide exclusivity by protecting your technology from competitors and preventing imitation or unauthorized market entry. This legal safeguard also creates opportunities for monetization, allowing you to generate revenue through licensing deals and royalty agreements.
A strong patent portfolio enhances your company’s appeal to investors by adding a valuable, legally protected asset to your balance sheet, ultimately boosting your overall valuation. Moreover, these patents establish a robust market position by defending your technology against infringement. Think of them as a protective barrier that not only keeps competitors at a safe distance but also strengthens your footing in any potential intellectual property disputes.
Contact the patent attorneys of V.O.
For personalized advice or to discuss specific aspects of your AI invention, consulting with a patent attorney who specializes in AI technology can provide tailored guidance to navigate you through the patenting process. The patent attorneys of V.O. Patents & Trademarks have been working at the forefront of patents for AI inventions and have cemented themselves as experts in the field. Contact us for more information.
Frequently asked questions about AI and patents
The description of an AI invention in a patent application is crucial. It must be detailed enough to enable a person skilled in the art to replicate the invention. This includes a thorough description of the machine learning model, the method of training, requirements to the data (not the data itself), the architectural framework, any relevant hardware implementations, and the technical problem it solves. Additionally, due to the opaque nature of some AI systems, providing evidence like experimental data or simulations that demonstrate the technical effect may be advantageous.
A “technical effect” refers to a clear and specific improvement made by an AI invention to existing technologies or processes. This effect should be directly attributable to the AI technology and not just a byproduct of general computer operation. Examples include optimizing energy consumption in industrial processes, improving the precision of robotic systems, digital audio, image, or video enhancement, efficient data compression algorithms, etc.
For an AI invention to be eligible for patent protection, it must exhibit a “technical character” by solving a technical problem with a technical solution. This means the invention should use technical means to achieve a measurable improvement over existing technology, and not merely automate an abstract idea or business process.
AI inventions that automate non-technical processes, such as business methods or purely abstract ideas, generally do not qualify for patent protection unless they involve a novel technical implementation that solves a technical problem. The key is whether the AI contributes to the technical character of the solution, not just the automation of non-technical activities.
Yes, patent requirements can vary between jurisdictions. For example, the European Patent Office (EPO) and the United States Patent and Trademark Office (USPTO) have different criteria for what constitutes a patentable AI invention. Although both jurisdictions require the AI invention to relate to a technical application of AI, there are differences in what is considered technical and what not. Also, the inventive step is differently assessed in both countries. It’s important to consult with a specialized patent attorney familiar with the specific requirements of the target jurisdiction(s).
No, the data itself does not have to be disclosed, but it is required to disclose the data requirements and the training method that is needed to train the machine learning model. In many cases, it is also advisable to describe what kind of machine learning model needs to be applied. If this is absent, the model itself may be considered to be insufficiently disclosed, which could result in a refusal of the application.
In Europe: No. In the US, may be. Under European standards, this type of applications is not considered to be of technical character, and therefore typically cannot be patented. In the U.S., it’s less clear-cut and falls into a bit of a gray area. American patent guidelines are more flexible. Even though the invention might initially seem like it’s just about organizing human activity or mental processes (which are usually not patentable), it could still be eligible for a patent if it adds enough unique technical elements that go beyond these simple processes. For instance, if the task it performs couldn’t be done by a human alone because of its complexity. However, trying to get such a patent in the U.S. can be complicated and expensive.
