Answer first: As artificial intelligence and machine learning technologies advance rapidly, companies face increasing scrutiny over patent eligibility. This analysis reviews the European Patent Office's approach to AI patentability, emphasizing the importance of demonstrating a technical effect and solving a technical problem. It h…
Patents context for IP teams
The rapid evolution of artificial intelligence (AI), particularly machine learning (ML), has driven a surge in patent filings within this sector. From established technology giants to emerging startups, there is growing interest in understanding the patentability requirements and limitations specific to AI inventions. This article examines key concepts in AI, the historical context of software patenting, and recent guidance issued by the European Patent Office (EPO) concerning AI-related patents.
AI differs from traditional software in that it enables computers to perform tasks involving reasoning or adaptation, rather than merely executing instructions explicitly coded by humans. While it is theoretically possible to program every scenario manually, this approach is impractical. Instead, ML allows systems to learn behaviors from data, encompassing a broad spectrum of algorithmic complexity.
Key takeaways for AI patentability at European Patent Office
- Confirm how the development affects patents ownership, enforcement, licensing, or portfolio records.
- Separate confirmed facts from legal interpretation before advising business teams.
- Map deadlines, affected assets, contracts, and evidence files to the responsible internal owner.
- Use the issue as a prompt for monitoring, filing strategy, dispute preparation, or member education.
Practical analysis
Contrary to common misconceptions, software is not categorically excluded from patent protection. However, patent eligibility depends on jurisdictional nuances and typically requires that software be part of a specific application solving a concrete problem. AI patents are attracting particular attention because AI often constitutes a core inventive element rather than a mere tool.
Under Article 52 of the European Patent Convention, certain categories such as mathematical methods and software are excluded from patentability "as such." This means that while standalone software or algorithms are not patentable, they may contribute to a patentable invention if integrated into a technical solution.
Central to the EPO's assessment is the concept of "technical character." The EPO's guidelines and case law clarify that software or mathematical methods can be patentable if they contribute to the technical means of solving a problem. Examples include AI methods that process images, videos, or interact with physical devices, thereby producing a technical effect.
In a 2020 EPO conference, officials emphasized that AI inventions undergo the same scrutiny as other computer-implemented inventions. The EPO tends to grant patents for AI/ML inventions that address technical problems involving external factors, such as image or voice processing, rather than for algorithms that merely organize or tag data. Past EPO decisions have rejected ML applications deemed too obvious or lacking a technical contribution, such as billing generation or document classification systems.
At a 2018 conference, the EPO reiterated that AI/ML patent applications must demonstrate how the AI or ML steps contribute to the invention's technical character. Applicants are advised to provide detailed descriptions of the technical effects achieved to enhance the likelihood of patent grant.
Several granted EPO patents illustrate these principles. One patent describes a neural network used as part of a method for vehicle driving assistance, framing the AI as a means to solve a specific technical problem. Another patent claims a voice input processing system where the neural network is a component addressing the technical challenge of voice identification. These examples highlight successful strategies where the AI is integrated into a broader technical solution rather than claimed in isolation.
A third example employs a "black box" approach, referring to a "failure condition estimation unit" configured by a learned AI model without explicitly mentioning ML in the claims. This approach can protect proprietary AI architectures but may risk broad claims vulnerable to prior art challenges.
Typically, claims refer to modules performing automated analysis or tasks, with the detailed description acknowledging possible use of AI/ML methods. This strategy shifts focus from the AI itself to the overall technical problem solved, aligning with EPO expectations and protecting innovation without disclosing sensitive algorithmic details.
In summary, patenting AI and ML inventions at the EPO requires careful framing of the invention to emphasize technical character and problem-solving contributions. Intellectual property managers and AI developers should provide comprehensive technical effect disclosures and avoid claiming AI methods in isolation to navigate the evolving patent landscape effectively.
Related IIPLA reading
Navigating Patentability of AI and Machine Learning Technologies Under EPO Guidelines As artificial intelligence and machine learning technologies advance rapidly, companies face increasing scrutiny over patent eligibility. This analysis reviews the European Patent Office's approach to AI patentability,... Read the full IIPLA blog post: https://iipla.org/blog/navigating-patentability-of-ai-and-machine-learning-technologies-under-epo-guidelines