In the rapidly evolving world of artificial intelligence (AI), securing patents for innovative technologies has become a critical concern for companies and inventors alike. As AI continues to revolutionize industries across the board, understanding the intricacies of patent law and how it applies to AI inventions is more important than ever. This blog post dives into the challenges and strategies for creating patentable AI inventions according to the standards set by the European Patent Office (EPO), based on insights from Robert Klinski, founder of Patentship and is a summary of this paper:

Robert Klinski; Creating patentable AI inventions according to EPO standards; THE PATENT LAWYER Issue 68

The AI Revolution and Its Impact on Patents

The disruptive nature of AI technology has sent shockwaves through the tech industry and beyond. The release of ChatGPT, a pre-trained AI language model, demonstrated the real-world impact of AI, reversing a multi-month bear market and propelling the technology sector into high gear. This surge in AI development has caught not only R&D departments but also patent offices off guard, raising crucial questions about how to handle AI-related technologies in the context of patent law.

Divergent Approaches Among Patent Offices

One of the key challenges in patenting AI inventions is the lack of a unified global approach. Different patent offices have adopted varying stances on the patentability of AI technologies:

  • The United States Patent and Trademark Office (USPTO) and the Japan Patent Office (JPO) have taken a less restrictive approach, already granting patents for AI core inventions.
  • In contrast, the European Patent Office (EPO) maintains a more restrictive stance, particularly when it comes to AI-implemented inventions and AI core inventions.

This disparity can lead to situations where an AI invention might be patented in the US and Japan but rejected by the EPO, complicating international patent strategies for innovators.

Graphics Top Companies by Generative AI Patents by Visual Capitalist https://www.visualcapitalist.com/ranked-top-companies-by-generative-ai-patents/

The EPO’s Approach to AI Patents

The EPO’s stringent criteria for patentability pose unique challenges for AI inventors. To be considered patentable by the EPO, an invention must make a reproducible contribution to the technical solution of a technical problem. This requirement has several implications for AI inventions:

  1. Technical Purpose: AI core inventions that address AI structures as such, without a clear technical purpose, are generally regarded as non-patentable mathematical methods by the EPO.
  2. Reproducibility: The EPO’s requirement for reproducibility effectively prevents the patenting of AI “black box” implementations.
  3. Inventive Step: Perhaps the most significant hurdle for AI patents at the EPO is the inventive step requirement. The EPO employs a unique problem-solution approach, demanding that an invention involving an inventive step must technically solve a technical problem.

The COMVIK Decision and Its Impact

The EPO’s assessment of inventive step in software inventions, including AI, is heavily influenced by the landmark decision T 0641/00, known as the COMVIK decision. This decision states that for inventions consisting of a mixture of technical and non-technical features, only those features contributing to the technical character of the invention are considered when assessing inventive step.

This approach can have unexpected consequences for AI inventions. The EPO may ignore or “blend out” particular claimed features that do not appear to contribute to a technical solution, even if these features are essential to defining the invention. This can potentially render the remaining claim as common knowledge, failing to meet the inventive step requirement.

Strategies for Patenting AI Inventions at the EPO

Despite the challenges, AI inventions can be patented at the EPO with the right approach. Here are some key strategies to consider:

  • Separate Training and Deployment Claims
    AI applications typically involve two phases: training the AI model and deploying the trained model. When possible, it’s advisable to claim these phases separately with distinct sets of claims. This approach can increase the chances of obtaining valuable AI patents.
  • Focus on Real-World Data for Training
    The EPO makes a distinction between AI models trained on simulated data and those trained on real-world data. Training an AI model with specific data for a specific technical task is more likely to be considered patentable. Therefore, when describing the training process, emphasize the use of data originating from physical measurements or real-world interactions rather than simulated environments.
  • Highlight Technical Adaptations
    AI inventions that are specifically adapted to solve technical problems in the context of technical applications are more likely to be patentable at the EPO. For example, a neural network used in a heart monitoring apparatus for identifying irregular heartbeats would be considered a patentable AI application.
  • Demonstrate Physical Interaction
    Inventions that involve physical interaction with the “real world” are more likely to be considered patentable by the EPO. When describing AI systems, emphasize how they interact with and respond to physical environments or processes.

Case Study: Patenting AI for Autonomous Industrial Trucks

To illustrate these strategies, let’s consider a hypothetical IP portfolio for AI-controlled autonomous industrial trucks in a closed indoor environment, such as an industrial storage area.

Key Considerations:

  1. Data Collection: Instead of relying on simulated data from a digital twin, focus on collecting data during the actual operation of the autonomous trucks. This approach reflects dynamically changing environments and is more likely to be considered patentable.
  2. Real-Time Adaptation: Emphasize the AI model’s ability to dynamically train and adapt to the changing environment based on real-time position information from the trucks and storage goods.
  3. Technical Problem Solving: Highlight how the AI system efficiently determines truck movements within the storage area based on actual position information, solving the technical problem of collision avoidance and optimizing routes in rapidly changing conditions.
  4. Network Integration: Consider incorporating a specific network architecture, such as a 5G network slice, to provide fast communications for the AI control system. This could potentially be patented as a service for dynamically controlling autonomous trucks in indoor environments.

Overcoming EPO Hurdles: Key Takeaways

  • Expert Claim Drafting: Given the EPO’s stringent requirements, expert claim drafting is essential for obtaining AI patents. Claims should be carefully constructed to emphasize technical contributions and real-world applications.
  • Focus on Technical Solutions: Always frame AI inventions in terms of how they solve specific technical problems. Avoid claims that could be interpreted as purely mathematical methods or abstract ideas.
  • Demonstrate Reproducibility: Ensure that your AI invention can be reproduced and that the patent application provides sufficient technical details to enable this reproducibility.
  • Highlight Real-World Data and Interactions: Whenever possible, emphasize the use of real-world data and physical interactions in both the training and deployment phases of AI systems.
  • Consider Multiple Jurisdictions: Given the varying approaches to AI patents across different patent offices, consider filing in multiple jurisdictions to maximize protection for your AI innovations.

The Future of AI Patents

As AI technology continues to advance at a rapid pace, patent offices worldwide will likely continue to refine their approaches to AI-related inventions. The current restrictive stance of the EPO may evolve as the field matures and the technical nature of AI applications becomes more apparent.

For inventors and companies working in the AI space, staying informed about the latest developments in patent law and maintaining a flexible approach to patent strategy will be crucial. By understanding the nuances of different patent offices’ requirements and tailoring applications accordingly, innovators can build robust IP portfolios that protect their AI inventions across multiple jurisdictions.

Invention harvesting and AI Patents

Invention harvesting is a powerful tool for companies developing AI solutions. It’s a systematic approach to identifying and documenting potential inventions, and it offers several key advantages when it comes to protecting AI innovations with patents.

Uncovering Hidden Value

AI solutions are often complex, involving intricate algorithms and data structures. Invention harvesting helps to unearth valuable innovations that might otherwise be overlooked. By systematically exploring the AI development process, it can reveal hidden gems and capture cross-disciplinary insights that arise from the intersection of AI with various fields.

Navigating the Challenges of AI Patenting

The fast-paced evolution of AI technology presents unique challenges for patent protection. Invention harvesting helps companies stay ahead of the curve by identifying patentable innovations early on. It also assists in overcoming patent eligibility hurdles by facilitating structured discussions and documentation that clearly articulate the invention and its applications.

Cultivating a Culture of Innovation

Invention harvesting fosters a collaborative environment where diverse teams can come together, spark new ideas, and share knowledge. This cross-pollination between AI researchers and domain experts encourages innovation. Moreover, regular harvesting sessions raise IP awareness among AI developers, educating them about the importance of protecting their work and recognizing potentially patentable inventions.

Building a Strong AI Patent Portfolio

AI technology has broad applications across various industries. Invention harvesting enables companies to identify strategic patent opportunities and build a comprehensive portfolio that covers core technologies and potential uses. This proactive approach can secure patent protection before competitors, potentially blocking them from key technological areas and enhancing the company’s competitive advantage.

Conclusion

Patenting AI inventions at the EPO presents unique challenges, but with the right approach, it is certainly achievable. By focusing on the technical aspects of AI implementations, emphasizing real-world applications and data, and carefully crafting patent claims, inventors can navigate the complex landscape of AI patents successfully.

As the field of AI continues to evolve, so too will the strategies for protecting AI innovations. Staying abreast of these changes and working with experienced patent professionals will be key to building a strong and sustainable AI patent portfolio in Europe and beyond.

The insights provided by Robert Klinski and the strategies outlined in this article offer a valuable blueprint for systematically generating valuable AI inventions that are potentially patentable at the EPO, despite its currently restrictive stance. By applying these principles, innovators can design and patent sustainable IP portfolios that cover future AI applications, ensuring their place at the forefront of this transformative technology.