Will AI Replace Patent Specialists?

Are you an inventor looking to patent your invention? Do you know how to draft a patent claim?

The patent claim defines the scope of legal protection granted to an invention – that is, what ultimately must be protected from unauthorized use. That’s why the claim is the most important part of the entire patent application.

The rise of artificial intelligence (AI) in recent years has inevitably impacted the field of patent application drafting. After all, if programmers ask AI to write code, why shouldn’t inventors ask AI to draft patent claims?

Recently, researchers from Taiwan conducted a study on drafting claims in patent applications for biomedical inventions – and it turns out things are not so simple [Evaluating application of large language models to biomedical patent claim generation - ScienceDirect].

As the authors point out, there are already a number of specialized tools that help inventors describe their inventions in words. One such tool is the Patent Generative Transformer (PGT) – a language model trained to assist with drafting patent applications. This model can generate parts of the patent application text based on existing sections, create titles and abstracts, improve text, and evaluate how well different parts of the application are related. However, the model completely ignores dependent claims.

There are also commercial tools that make life easier for potential applicants, such as ClaimMaster, which detects and corrects textual errors, adjusts figure numbering, generates descriptions and images based on claims, and more .

Nevertheless, existing models struggle when it comes to drafting claims, especially dependent claims, which expand upon or add features to the invention described in the independent claim. Dependencies can be complex – for example, when one claim depends on several others. This logical structure has proven too difficult for these tools without additional specialized training.

Furthermore, using AI tools raises concerns about confidentiality. Submitting confidential invention details to cloud-based services may be unsafe. Using a local text generator would require the applicant to train and fine-tune the language model themselves, which can be an overwhelming task when dealing with large models.

In short, it’s still difficult to do without a specialist. Although the authors of the above-mentioned study were able to fine-tune tools to generate biomedical patent claims with proper style, structure, and technical content, they concluded that experienced patent engineers remain indispensable for such a complex task.

That’s why we recommend turning to professionals to prepare your patent application. They’ll not only help you understand what exactly you’ve invented, but also advise how best to describe it to protect your interests.