Patenting Artificial Intelligence in a Complex World
This is a summary of the lecture by Yanan Huang, IP Manager at TRUMPF China about the patentability of AI from the joint CEIPI European Patent👉 A legal right granting exclusive control over an invention for a limited time. Office Diplom Universitaire IP Business Administration
Artificial intelligence (AI) is rapidly transforming industries, driving innovation👉 Practical application of new ideas to create value., and reshaping how we interact with technology. As AI’s influence grows, so does the importance of protecting AI-related inventions through patents. However, the unique nature of AI presents significant challenges to traditional patent frameworks. This post summarizes a lecture by Yanan Huang, IP manager at TRUMPF China on the patentability of AI. It delves into the key considerations for securing patents for AI-based innovations, drawing insights from patent practices in the US, China, Europe, and Japan.
This lecture is part of the certificate course IP in digital technologies
https://ipbusinessacademy.org/certified-university-course-ip-in-digital-technologies
and part of the university diploma (distance learning) IP Business Administration
https://ipbusinessacademy.org/ceipi-epo-university-diploma-in-ip-business-administration-du-ipba
In this lecture, IP experts Robert Matthezing, Bettina de Jong, and Johan Krebbers delved into the intricacies of IP management👉 Strategic and operative handling of IP to maximize value. for CPS, offering valuable insights into patenting strategies, trade secret protection, data management, and the evolving landscape of digital innovation. This post summarizes the key takeaways from their lecture, providing a roadmap for navigating the IP maze in the age of CPS.
Defining the AI Landscape: Components and Categories
Huang begins by defining AI as software and hardware capable of learning, solving complex problems, making predictions, and performing tasks that require human-like sensing. She emphasizes the surge in AI patent applications, particularly from China and the US, highlighting the strategic importance of patenting in this field. AI innovations can be broadly categorized based on their components and applications:
AI Components
AI components are the foundational building blocks that enable artificial intelligence systems to function effectively. These include knowledge processing, which utilizes predefined knowledge databanks to simulate reasoning and decision-making processes. Speech techniques, such as those used in intelligent assistants like Siri or Alexa, enable machines to process and respond to human language audibly. Natural language processing (NLP) focuses on analysing and interpreting vast datasets of human language to facilitate understanding and communication between humans and machines. The hardware components of AI provide the necessary computing power to process data-intensive tasks, often leveraging advanced GPUs or specialized chips. Evolutionary computation mimics Darwinian evolutionary principles to optimize solutions iteratively, making it particularly useful in solving complex problems. Finally, **machine learning** involves training models on data to recognize patterns, make predictions, and improve performance over time without explicit programming.
AI Categories
Class 1: Core AI Inventions
Core AI inventions focus on enhancing the fundamental functionality of AI systems themselves. These innovations often involve advancements in **mathematical or statistical information processing**, such as breakthroughs in deep learning algorithms or reinforcement learning techniques. The goal is to improve the efficiency, accuracy, or adaptability of AI models, making them more capable of solving complex problems. For example, a new neural network architecture that reduces computational requirements while maintaining high accuracy would fall into this category. Such inventions are critical for pushing the boundaries of AI’s capabilities and enabling more sophisticated applications across industries.
Class 2: AI Utilization in Technical Fields
This category involves applying AI as a tool to achieve specific technical effects in various domains. Examples include using AI for **image recognition**, where machine learning models analyse visual data to identify objects or patterns with high precision. Similarly, AI can enhance **speech processing**, enabling more accurate transcription or voice commands in real-time applications. In **natural language processing**, AI is used to translate languages, summarize text, or detect sentiment in written communication. Robotics also benefits from AI by integrating intelligent decision-making into autonomous systems, such as self-driving cars or industrial robots. These applications demonstrate how AI serves as a powerful enabler for innovation across diverse technical fields.
Class 3: AI as “Inventor”
This emerging category refers to inventions created autonomously by AI systems without direct human intervention. For instance, an AI algorithm might independently design a new chemical compound or generate a novel engineering solution based on its training data. While such inventions showcase the immense potential of AI, they raise significant legal and ethical challenges regarding inventorship. Current patent laws generally require inventors to be human beings, leaving inventions generated solely by AI in a legal gray area. This issue has sparked debates about whether and how patent systems should adapt to accommodate non-human inventors while maintaining accountability and fairness in intellectual property👉 Creations of the mind protected by legal rights. rights.
The Patentability Puzzle: Eligibility, Inventiveness, and Disclosure
Huang highlights three major hurdles in patenting AI inventions: eligible subject matter, inventive step standards, and sufficient disclosure requirements.
Eligible Subject Matter: What Can Be Patented?
The core question here is whether an AI invention👉 A novel method, process or product that is original and useful. qualifies as patentable subject matter under the laws of different jurisdictions. Huang notes similarities in the approaches of the US, Chinese, and European Patent Offices (EPO) in requiring technical characteristics and technical effects. They generally exclude AI subject matter that is considered “well-understood, routine, or conventional activity.”
- United States: The US patent law accommodates all three classes of AI inventions, without raising unique eligibility concerns for any of them.
- China: China adopts a more restrictive approach, requiring human involvement in defining the technical problem. Only inventions where a human raises the problem and AI either verifies the solution or independently solves the problem are eligible. Class 3 inventions (AI as sole inventor👉 A person who creates new devices, methods, or processes.) are excluded.
- Europe: The EPO excludes mathematical methods claimed “as such” (Art. 52(2) and 52(3) EPC). However, AI inventions can be patentable as computer-implemented inventions (CII) if they involve a technical element, contributing to a technical character and effect. Germany even allows AI inventions to be registered as utility models for faster protection, without substantive examination.
- Japan: Japan adopts a less stringent stance, deeming inventions “specifically implemented by using hardware resources” as eligible subject matter.
Inventive Step: Is It Non-Obvious?
Demonstrating an inventive step, or non-obviousness, is crucial for securing AI patents. Huang points out that the USPTO, EPO, and Japan Patent Office (JPO) generally apply existing standards for computer technology to AI inventions, rather than creating entirely new ones.
- United States: The invention must be new and involve an inventive step, meaning it is not obvious to a person skilled in the art (POSITA).
- Europe: The EPO uses a two-hurdle approach. The lower-level hurdle requires claims to be drafted with “computer-implemented” or “processing hardware.” The higher-level hurdle demands a “technical effect serving a technical purpose.”
- Japan: The JPO requires novel input and output data, and the process must be implemented by hardware resources. The JPO considers “mere systemization using AI” or “mere modifications of methods” as lacking an inventive step. Selecting training data with significant effort or preprocessing data to get a pattern right, on the other hand, may indicate an inventive step.
Sufficient Disclosure: Can It Be Replicated?
Sufficiently disclosing the AI invention is another critical requirement. The standard requirement is that an application should disclose the invention in a manner clear and complete enough for a person skilled in the art to carry it out. This includes the algorithm, the data used, and how it performs the claimed function.
- However, AI’s “black box” nature makes it challenging to meet this requirement. Knowing the input, output, image class, and decision process might not guarantee the same result when replicated by others.
- IP5 offices (the five largest IP offices in the world) have set stricter disclosure requirements, emphasizing the need for clear, sufficient, and complete disclosure to enable implementation by those skilled in the art.
- Japan: JPO guidelines emphasize the condition where a certain relation (correlation) among the multiple types of data can be recognized based on the disclosure in the description, or presumed in view of common general technical knowledge.
- Europe: The EPO requires disclosure of the training data used or a dataset enabling the technical problem to be solved.
- United States: USPTO guidelines require sufficient detail regarding the hardware and software, including detailed steps, procedures, formulas, diagrams, and flowcharts that perform the claimed function.
AI vs. Human Innovation: Key Differences
Huang highlights the fundamental differences between human and AI-driven innovation:
- Human Innovation: Relies on an unstructured database, knowledge in the brain, an access protocol, and, crucially, imagination.
- AI Innovation: Relies on machine intelligence, a structured algorithm, and a structured database. AI lacks imagination; its ideas come from extracted algorithms and databases.
While mainstream patent offices are granting AI-related patents, the predictability of patentability remains uncertain due to the challenges in eligibility, inventive step, and disclosure.
Future Challenges and Considerations
Looking ahead, the patenting of AI presents novel challenges that require careful consideration and adaptation of existing legal frameworks. Huang addresses these emerging issues, focusing on the potential impact of AI on the standard of non-obviousness and the question of AI inventorship.
- The Impact on POSITA: As more AI-generated prior art emerges, the knowledge level of a person having ordinary skills in the art (POSITA) may increase, raising the bar for non-obviousness and making it more difficult for human-made inventions to be considered inventive.
- AI Inventorship: Current laws require inventors to be human beings, presenting a challenge for inventions created solely by AI.
Huang suggests that a separate prior art database and an independent non-obviousness standard for AI-related inventions might be a future trend.
Conclusion
Patenting AI is a complex and evolving field. While AI is transforming industries and driving innovation, the legal framework surrounding its protection is still catching up. Understanding the nuances of patent eligibility, inventive step, and sufficient disclosure in different jurisdictions is crucial for innovators seeking to protect their AI-based inventions. As AI continues to advance, ongoing dialogue and adaptation of patent laws will be necessary to foster innovation while ensuring fairness and clarity in the patent system.
