tl;dr— AI-assisted patent drafting isn’t about replacing human intellect with machines; it’s about augmenting our capabilities as legal professionals, freeing us to focus more on the complex and creative aspects of patent law. AI tools, especially Large Language Models (LLMs), have the potential to revolutionize the way we approach patent drafting, significantly enhancing the quality of our work.
I recently had the privilege of speaking at a symposium hosted by the European Patent Office to celebrate the 50th anniversary of the European Patent Convention. I shared my thoughts on a topic that’s close to my heart:
The impact of Artificial Intelligence (AI) on the patent attorney profession.
This article is an extension of that discussion, highlighting how AI-assisted patent drafting is not just about efficiency but about enriching the quality and creativity of our work.
Join me on my journey towards the AI-supercharged patent attorney
My journey with AI began not just as a professional necessity but as a personal fascination. Being a computer nerd, I’ve always been intrigued by the evolving landscape of digital technology and its potential impacts on our profession.
AI, especially in the form of Large Language Models (LLMs), is redefining the boundaries of what we can achieve in patent drafting.
A word on confidentiality
Before we explore the transformative potential of AI-assisted patent drafting, let’s address the elephant in the room: confidentiality.
Sure enough, you shouldn’t put confidential invention details into ChatGPT and let it draft patent claims for you because you can’t control where your data will end up.
But the online-hosted ChatGPT is only the tip of the LLM iceberg.
There are numerous LLMs out there that run completely offline, without sending any data anywhere. While none of them is quite as good as GPT-4 right now, these models are catching up quickly, a hot candidate of mine being Llama 2.
These models offer a promising solution to maintaining confidentiality while leveraging AI’s capabilities.
Once the confidentiality problem is solved, I believe the ear of AI-assisted patent drafting will unfold in three stages:
The three stages of AI-assisted patent drafting
Stage 1: Automation and Language
The first stage of AI-assisted patent drafting has a focus on automation and linguistic improvement.
Example tasks that can easily automated include:
- Generating a description shell from the claims
- Generating standard figures from the claims like flowcharts and structural overviews
- drafting standard boilerplate language
Some of these rather programmatic automations can even be done with standard non-AI tools in visual basic or python, as one of the commenters on this LinkedIn post rightly points out.
In terms of linguistic improvement, this is where LLMs play their strength, helping patent attorneys improve their writing to ensure clarity and conciseness in patent applications.
Stage 2: Technical Depth and Domain Expertise
At the second stage, AI’s role extends to adding technical depth.
For instance, an LLM equipped with comprehensive knowledge bases like Wikipedia can suggest definitions of key claim terms.
It can also propose technical effects for claim features to be put into the summary section, thereby enriching the patent application with detailed technical insights.
Stage 3: Legal Expertise and Creative Guidance
In the third stage, AI trained on patent-specific information will take it to the next level. An example of this would be an LLM suggesting a compelling narrative why certain claim features match with the legal framework for CIIs at the EPO, using the specific language from case law. This level of integration signifies AI’s evolution from a drafting assistant to a creative legal tool.
Will AI replace patent attorneys?
Here’s a great comment on a related LinkedIn post:
Patent drafting (if done well) is an exercise in applied judgment. This is where AI-based tools are at a built-in disadvantage compared to human patent attorneys and patent agents. I don’t see how you can train an AI tool to cater to a client’s preferences and draft a document around uncertain and indeterminate law.
Stuart Shapley
Case in point – how on god’s green earth would you train an GPT tool to draft claims around the shifting requirements for subject matter eligibility in the U.S.? Neither the Federal Circuit nor the Patent Office can agree on where the boundaries for eligibility should be.
I completely agree.
My vision of AI-assisted patent drafting is that these tools are a patent attorney’s wingman freeing the human attorney from the simpler, repetitive tasks, and helping the human attorney produce more compelling text.
Writing great claims, on the other hand, requires a very strategic balance between conflicting goals like claim scope vs. patentability, which can only be achieved based on an understanding of the strategic goals of the client.
Putting it all together
An LLM that integrates all three stages would be a powerhouse tool for patent attorneys. Such a model would not only handle routine tasks and improve writing quality but also add technical and legal depth to patent applications.
As far as I can see, the technical foundations for such domain-specific models are already available. For instance, the recent introduction of custom GPTs allows for the tailoring of AI models to specific tasks and industries.
In the context of patent law, custom GPTs could be trained to specialize in various technical fields, understand specific legal jurisdictions, and adapt to the unique styles and requirements of individual patent attorneys.
This offers an unprecedented level of precision and relevance in AI-assisted patent drafting.