The use of artificial intelligence in creating inventions raises complex questions for patent law, particularly concerning disclosure, inventive step and inventorship. At the International Association for the Protection of Intellectual Property (AIPPI), held at PACIFICO Yokohama, a session titled “Artificial Creativity or Intellectual Reality: Use of Generative AI in Pharma and Biotech and its Impact on the Patenting Process” addressed these emerging challenges, highlighting diverse perspectives and the urgent need for clarity as AI becomes more integrated into the inventive process.

AI’s growing role in innovation

The discussion opened with a compelling study on AI-generated molecules, where an AI system was tasked with creating new molecules. In Phase I clinical trials, AI-designed molecules achieved an 80-90 percent success rate, significantly higher than the historical industry average. In Phase II, the success rate was around 40 percent, comparable to the historic average and based on a limited sample size. While AI excels at improving drug safety and identifying “drug-like” properties, enhancing efficacy remains a key area for improvement.

Given this, how much information about the AI’s involvement must be included in a patent application?

Viewpoints on this issue vary, reflecting the lack of a global consensus. One approach suggests the level of disclosure should depend on how the AI was used. If AI is merely a tool for a task, such as screening a compound, extensive disclosure may not be necessary. However, if the AI model itself is part of the claim or the invention is a novel algorithm, then full disclosure is essential. This could include:

• The AI model’s architecture
• The specific training data used
• The date the training was completed
• The AI’s output used as supporting evidence for the invention’s efficacy

From a Japanese perspective, the current patent law has no specific rules for AI-related inventions. The primary requirement is to describe the invention in the specification clearly enough for a person skilled in the art to carry it out. This practical approach aims to satisfy the requirements of most jurisdictions, even without explicit AI guidelines.

In the United States, the situation is different. The U.S. Patent and Trademark Office issued guidelines in April 2021, reinforcing that the duty to disclose material information is a core part of patent practice. This places the onus on patent practitioners to be transparent about the role AI played in the invention, as they are ultimately responsible for the accuracy of their submissions.

A major concern raised was reproducibility. A primary worry is that data generated by a sophisticated AI might not be reproducible by a patent examiner. If an examiner cannot replicate the process or verify the results, it becomes very difficult to assess patentability requirements such as inventive step and sufficiency of disclosure. This challenge strikes at the heart of the patent system: a limited monopoly in exchange for a full public disclosure of the invention.

Strategic considerations for practitioners

So, when is the right time to disclose AI’s involvement? There is no one-size-fits-all answer; the decision is situational.

However, a key principle for practitioners is the responsibility for accurate submissions. Maintaining credibility with patent offices is crucial. One practical tip is to frame discussions about AI’s future potential in the future tense to avoid making unsubstantiated claims.

Ultimately, the most important directive for patent attorneys and agents is to be proactive. It is no longer enough to ask inventors what they invented. Practitioners must now start questioning inventors about their use of AI in the inventive process. Understanding how AI was used is the first and most critical step in navigating these new and complex disclosure requirements.

- Excel V. Dyquiangco, reporting from Yokohama