This is a follow up on our previous Circular Letter regarding the Draft Guidelines for the Examination of AI-related patent applications issued by BPTO, currently under public consultation.
Through extensive discussion across various forums, relevant issues that deserve special attention were identified, to ensure the rights of IP owners are adequately protected.
1. Limitations on the Protection of Training Methods and Improvements on AI Models and Techniques The draft guidelines suggest that methods of training AI—including neural networks, genetic algorithms, support vector machines, and regression methods—when not applied to technical field, should be regarded as mathematical methods and therefore ineligible for patent protection.
The draft guidelines also determine that claims in categories directed exclusively to the abovementioned subject matter will not be accepted due to involving ineligible subject matter. Claims in these categories that contain technical features using AI for solving a technical problem should be redrafted for claiming the concrete technical application of the AI technique or model instead.
Our comments: The Brazilian IP Law does not consider technical application and/or determination of a technical field as patentability requirements for an invention.
The claim wording required by these guidelines imposes restrictions that could significantly limit Protection rights for AI models and techniques, including training methods, neural networks, genetic algorithms, Support vector machines, and regression techniques.
It is important to demonstrate to the BPTO, through concrete examples, that these inventions should be eligible to patent protection, provided that they involve a technical solution to a technical problem and envolve a new technical effect, in line with international standards.
2. Excessive Requirements for achieving Sufficient Disclosure For inventions referring to improvements in AI models and techniques, these guidelines require detailed description of all elements of the AI models and techniques, such as the functions and parameters used, the training method adopted, the hardware architecture providing support for the AI system.
For AI based inventions, the guidelines cite some examples including raw data processing, training techniques, validation, and testing procedures, as elements that may be required for sufficient disclosure.
Our comments: The text of the guidelines is vague regarding when such extensive information is truly necessary, giving space for objections requesting excessive details for meeting sufficient disclosure and clarity.
The Guidelines should more expressly establish that disclosures should only encompass what is necessary to enable a person skilled in the art to reproduce the invention without undue experimentation, in accordance with our IP Law and the general Examination Guidelines.
3. Definition of the Person Skilled in the Art The guidelines define the person skilled in the art as a person or group of people who have combined knowledge of both AI methods and techniques and the technical field in which AI is applied.
The guidelines further assume that the person skilled in the art has the ability to apply and experiment with common routines related to data processing, model selection, and the use of AI techniques, as well as machine learning training strategies.
Our comments: This standard appears excessively high and inconsistent with the definition of a skilled person in the General Examination Guidelines. Such a rigorous standard could unduly raise the inventive step threshold for AI inventions, potentially restricting patentability.
Furthermore, the proposed selection of individuals to represent the skilled person might introduce hindsight bias; choosing experts after knowing the invention does not realistically reflect an inventor’s actual inventive process.
We believe the correct approach is to align the definition with that in the General Guidelines: a person with average knowledge of the technique in question (technical and scientific expertise, and/or with practical operational knowledge) at the time of filing the application.
The Guidelines should also more clearly specify that a skilled person’s ability to use AI should consider the AI tools available at the time of filing the application.
FINAL REMARKS We are preparing detailed arguments and proposing amendments to the current draft to address the issues outlined above, aiming to prevent unnecessary barriers to patentability and to better reflect the realities of AI innovation.
Any additional input supporting our position or highlighting other relevant concerns would be highly appreciated to strengthen our efforts.
