Artificial intelligence (AI) has the potential to significantly enhance the patenting process in the legal world. AI models, specifically large language models (LLMs), utilize deep neural networks to generate text that closely resembles human language. These models have been trained […]
Artificial intelligence (AI) has the potential to significantly enhance the patenting process in the legal world. AI models, specifically large language models (LLMs), utilize deep neural networks to generate text that closely resembles human language. These models have been trained to predict accurate responses based on a vast number of documents that teach them about human language.
By using LLMs as tools, patent attorneys can benefit from their use in content generation and technical research. When initially introduced nearly a year ago, the use of LLMs in the patenting process faced skepticism regarding accuracy, the legal status of their outputs, and privacy concerns when communicating with third parties. However, with the right approach, professional standards, and ethical guidelines, incorporating LLMs as part of experts’ toolkit can bring various advantages.
LLMs can assist patent attorneys in researching fundamental concepts related to invention. This ensures that attorneys are well-versed in the technological domain before starting the patenting process. It may involve delving into research topics to gain a complete understanding of the underlying technology and asking questions that facilitate communication with inventors.
Additionally, LLMs can aid in creating portions of patent application templates or introductory materials. Patent attorneys can obtain these benefits without the risk of disclosing confidential information and receive quick feedback and analysis that can improve patent quality by enhancing their technical understanding.
While LLMs offer rapid knowledge augmentation, practitioners should approach these tools with caution. It is important to note that generated LLM responses cannot be automatically considered accurate. Patent attorneys must exercise careful scrutiny to verify the veracity of generated responses—both independently and within the context of the entire document. Errors, sometimes known as “hallucinations,” can occur. Furthermore, some LLMs are retrained based on conversation logs, and operators have discretion over reviewing these sessions.
When utilizing this technology, care should be exercised when using public LLMs or LLMs that are not confined to secure workspaces to avoid disclosing confidential or other non-public information, including the risk of incorporating such information into the training of the model.
Apart from these general concerns, there is the issue of prior art and authorship that is specific to the use of LLMs in the patenting process. First, LLMs are trained based on existing data and can be regarded as well-informed but lacking inventiveness. In this context, the output of LLMs can be seen as a collection of potential prior patents, such as training data documents. This text could be fully or partially identical (derivative) to previous training documents. If patent attorneys include such output in a patent application, it can pose problems in terms of incorporating prior art into the application.
An even more complex problem arises with the reverse standpoint: that language AI models can process existing datasets and come up with potentially new and non-obvious answers to user queries.
For example, consider a discovery where a system utilizes a machine learning (ML) algorithm as one step in a larger inventive process. If specific examples of that ML algorithm are requested from a language AI model, it might provide feasible and inventive options that inventors have not considered. If one or more of these alternatives are included in the patent, the question arises as to who “invented” the use of this specific ML algorithm.
This raises questions about authorship; however, many jurisdictions consider AI cannot be an inventor. If the LLM contributes an inventive element, it may be impossible to pinpoint the “inventors” of the combination.
While LLMs can play a significant role in supporting patent attorneys, the impact of their use on the patenting process should be carefully considered.