Artificial Intelligence (AI) has emerged as a cornerstone of modern technological innovation, with applications spanning healthcare, industrial control systems, transportation, and beyond. As AI becomes increasingly embedded in industry and society, ensuring adequate intellectual property protection for AI-based inventions is critical. Patents not only incentivize innovation but also provide a framework to protect technological advancements.
This article explores the distinct approaches taken by the European Patent Office (EPO) and the United States Patent and Trademark Office (USPTO) in assessing the patentability of AI-based inventions. Learn how to navigate these frameworks effectively and ensure successful patent applications for AI innovations. While these offices utilize different criteria and frameworks for examination, their outcomes often converge. By examining the methodologies of both offices, as well as specific examples from recent guidelines, this article provides a comprehensive overview of how to successfully navigate the patenting process for AI inventions.
What Are the EPO’s Requirements for AI Patentability?
The EPO’s assessment of patentability is governed by the basic requirements outlined in Article 52(1) of the European Patent Convention (EPC). According to Article 52(1) EPC, "European patents shall be granted for any inventions, in all fields of technology, provided that they are new, involve an inventive step and are susceptible of industrial application." The reference to "all fields of technology" was introduced in the course of the EPC's revision (EPC 2000) to align with Article 27 of the Agreement on Trade-Related Aspects of Intellectual Property Rights (TRIPS). This amendment clarifies that patent protection is reserved for creations in the technical field and that the claimed subject-matter must have a "technical character," or involve a "technical teaching," i.e., an instruction addressed to a skilled person on solving a particular technical problem using particular technical means (see summary given in par. 24ff of G 1/19 issued by the Enlarged Board of Appeal of the EPO). These requirements present a specific hurdle for AI inventions, as they typically qualify as mathematical models, algorithms, or computational methods. As a result, AI-based claims must demonstrate technical and practical implications and implementations beyond their abstract nature.
A key consideration for AI inventions is that they should serve a specific technical purpose. This could include applications in medical diagnostics, autonomous vehicles, or industrial process optimization, where the AI system addresses a concrete problem in a specific technical environment. However, merely mentioning the application of AI to a technical area may not be sufficient for patentability. The EPO applies a problem-solution approach, testing whether it is obvious to use AI in the given technical area. Normally, additional elements are required, such as an inventive AI implementation for processing technical data in the specific area or a novel generation of AI training data. Alternatively, AI inventions may demonstrate a low-level technical design, focusing on details of the AI model itself, such as new developments in neurons, layers, feedback mechanisms, or other architectural improvements e.g. considering hardware constraints of the system on which the AI is implemented.
Disclosure also plays a crucial role in demonstrating how an AI invention solves technical problems. While it is not always necessary to provide the specific training data used, applicants must describe the type of data and the methodology for training the AI model to ensure that the invention is reproducible by a skilled practitioner.
How Does the USPTO Apply the Alice/Mayo Framework to AI Inventions?
The USPTO evaluates the patent eligibility of AI inventions under the Alice/Mayo two-step framework. This structured approach ensures consistency in analyzing whether AI-related claims qualify for patent protection:
- Step 1: Determining whether the claim is directed to a statutory category, such as a process, machine, or manufacture.
- Step 2A (Prong 1): Evaluating whether the claim recites a judicial exception. Judicial exceptions include abstract ideas such as mathematical concepts, mental processes, or methods of organizing human activity. AI inventions are particularly likely to fall under this prong, given their reliance on abstract mathematical models and algorithms. If this prong is true, the analysis continues with Prong 2 to determine whether the exception is integrated into a practical application.
- Step 2A (Prong 2): Determining whether the judicial exception is integrated into a practical application. For AI inventions, this involves demonstrating how the invention improves computer functionality or a specific technical field. A key consideration is whether the claim provides a particular solution to a problem in a particular technical environment or specifies a concrete way to achieve a desired outcome. Regarding AI inventions, this is reminiscent of the EPO requirement of a specific technical purpose. Thus, it is not sufficient to merely mention the application of AI or simple instructions to implement a judicial exception on a computer. A claim that merely recites a judicial exception and fails to integrate it into a practical application is deemed “directed to” the exception, triggering further analysis under Step 2B.
- Step 2B: This step assesses whether the claim includes additional elements that amount to "significantly more" than the judicial exception itself, providing a concrete advancement beyond the abstract idea.
The USPTO’s subject matter eligibility guidance, particularly Examples 47 and 48, illustrates these principles. For instance:
- A claim for detecting malicious network packets is considered patent-eligible because it addresses a technical problem in network security through AI-based methods.
- A claim for separating background noise from speech qualifies because it demonstrates how AI improves audio processing technology.
AI inventions that fail to move beyond abstract ideas, such as generic data analysis without specific technical implementations, remain ineligible under the USPTO framework. Such claims may also be deemed ineligible unless it is demonstrated under Step 2B that the claim elements provide significantly more than common sense implementation details and approaches. This could include inventive features or specific improvements that go beyond routine or conventional techniques, thus establishing a meaningful advancement over the abstract idea. By explicitly linking the invention to technical improvements or specific applications, applicants can strengthen the patentability of their claims.
How Do the EPO and USPTO Approaches to AI Patents Compare?
While the EPO and USPTO apply distinct frameworks for evaluating the patentability of AI-based inventions, their results frequently align. The EPO emphasizes demonstrating a specific technical purpose and technical effect, while the USPTO focuses on integrating abstract ideas into practical applications under the Alice/Mayo framework.
For example, the USPTO Example 47 (Anomaly Detection) discusses claims that involve the use of artificial neural networks (ANNs). One claim mentions an ANN for detecting anomalies without specifying a particular technical environment or concrete implementation. Such a claim would be ineligible under both the USPTO and EPO approaches because it fails to establish a specific technical purpose or integrate the abstract idea into a practical application.
Conversely, another claim in Example 47 describes an ANN specifically applied to detect malicious network packets, addressing a technical problem in network security. This claim qualifies as patent-eligible under both systems, demonstrating alignment despite differences in analysis methods.
Similarly, in Example 48 (Speech Separation), a claim describing a general AI method for separating speech signals without further technical implementation details is ineligible. However, a claim that specifies how the AI system achieves improvements in audio processing, such as enhancing speech-to-text transcription or isolating speech sources in complex environments, is eligible under both systems.
The key takeaway is that, regardless of jurisdiction, AI inventions must move beyond abstract ideas and demonstrate tangible technical advancements or applications. Both offices emphasize the importance of technical specificity and practical integration to meet patent eligibility criteria.
What Are the Key Takeaways for AI Patentability at the EPO and USPTO?
The patentability of AI-based inventions is governed by distinct frameworks at the EPO and USPTO, but their underlying principles and outcomes often converge. Both offices require AI inventions to transcend abstract ideas and exhibit concrete technical advancements. While the EPO mandates a specific technical purpose and technical effect, the USPTO emphasizes integrating judicial exceptions into practical applications under the Alice/Mayo framework.
For inventors, this means ensuring claims clearly address problems in specific technical environments or demonstrate improvements to existing technologies. Abstract or generic claims, such as those merely mentioning using AI models are unlikely to succeed under either framework. Low-level design details serve as an alternative to applying AI to certain technological areas, as they can also qualify for patent eligibility.
Ultimately, the patenting process for AI inventions underscores the importance of bridging the gap between abstract concepts and technical realities. By aligning claims with the rigorous standards of both the EPO and USPTO, inventors can navigate the complexities of patent eligibility and protect their innovations effectively.
Take the Next Step
For expert advice on drafting AI patent applications that meet the requirements of both the EPO and USPTO, contact our specialists today. Let us help you turn your innovative ideas into protected assets.
Talk to us and contact us for further information at:
Phone: +49 (0)611 / 341568-0
Fax: +49 (0)611 / 341568-11
E-mail: Send email to us
PGP-Key
Author: Dr. Michael Schmid