Author: Patent Consultant Maria Laini.
We are now well into the AI revolution, which is disrupting the day-to-day work of all kinds of professionals. From AI-agents for business and project management to image and video generation tools, there seems to be infinite opportunities for AI to boost our productivity. Patent attorneys are not exempted from this, and we already have a vast choice of AI-based solutions that can be used to draft and even prosecute patent applications.
At this point, there are two big questions being debated: are these tools useful? And even more importantly: will, in the future, AI agents replace patent attorneys?
I feel that, as a junior patent consultant, it is important for me to consider these questions. Whilst I do not have all the answers, I thought it would be useful to put down in writing some of my first impressions working with one of the main products on the market: the Patent Drafting AI Copilot designed by Solve Intelligence.
Reading suggestion if you want more knowledge on patent drafting: Mind the Gap (Between Words and Meaning) when patent drafting
What is an AI Copilot?
An AI patent drafting Copilot is an assistant that is built in an existing or customised text editor. One of the interesting features of this type of system is the presence of a chat-based interface that is meant to be used by the attorney to interact with the AI assistant to draft the patent application.
You can think of this as a virtual agent ready to write the patent application according to the attorney’s instructions (or prompts), with the additional advantage that it can also provide context-aware outputs. This means that the agent will not just blindly execute individual instructions but will provide responses that reflect the context of the overall drafting project.
AI Copilots are only one kind of AI drafting tools. AI-native patent drafting tools are another alternative. An interesting introduction to each of these tools, including their pros and cons is provided here.
Legal style with no legal judgment
It is worth starting with how you should not work with an AI Copilot.
One may think that you could just feed the AI assistant some technical documents and let it draft the whole application whilst you sip a cup of coffee and plan your next holiday trip. However, this is not how you should approach these tools.
To put it simply, an AI drafting tool is trained to speak and write like a patent attorney but cannot think like one. In order to get the content right, legal supervision is therefore essential.
I have run a little experiment myself. I have tried to see what happens, when you let the machine do it all. I uploaded a technical/commercial document describing a particular mobile application I found on the Internet, and I instructed my AI assistant to write the whole draft without me overseeing the process.
The draft produced was, at a quick scan, very convincing. The claim language was there, so as the style. It seemed written by an attorney, with the additional benefit that it didn’t take several hours to be put on my desk.
However, when reading the draft, I quickly spotted many fundamental problems with the content. The claims were either irrelevant or overly specific, often focusing on features that were not important and missing the essence of the invention. In a similar fashion, the description was off-track, highlighting advantages of irrelevant features. Finally, the background section was shamelessly pointing to the solution, by reciting some of the features of the independent claim as “notably needed” (for the non-technical readers: this is the exact opposite of what you should be doing, when writing the background section of a patent application).
What made things worse was that the document created was very long, and it was impossible to know where to start with the (many) corrections needed, as it was now a finished product.
I obtained a similar result with the figure generator function of the Copilot. Whilst you can get nice, patent-friendly technical drawings by simply uploading figures in the system, without supervision and without a good set of claims, the AI assistant will get things wrong and create completely irrelevant and wrong labels for various features automatically identified in the uploaded drawings. This will then reflect in the description, if the description is written after generating the drawings.
So, what is the right approach to draft with AI?
After experimenting with this product, and after listening to podcasts and reading articles from experts in the field, I still do not have all the answers, but I have learnt some good lessons, that I am applying when using AI for my drafting.
Fast typing and nice writing skills
The first and most obvious advantage of AI tools for drafting is that they are incredibly skilled writers. It is very convenient to use them to find nice and/or alternative ways to express your thoughts when describing the invention.
Another great benefit is coherence. Can you relate to my frustration of finding that, in the draft you spent hours writing, the claimed “authorisation message response” turns into an “authorisation response” or “authentication response message” in various paragraphs of the description? AI tools are good at avoiding these mistakes.
AI is not only for menial tasks
The other lesson I learnt is that AI Copilots are not only great to relieve attorneys of the most menial tasks, such as renumbering dependent claims, create reference numerals and fix typos throughout the description.
Whilst these are useful benefits, they only represent a fraction of what AI can do for improving our patent drafts. In my opinion, these are not even the best and most useful applications of AI for drafting.
A brainstorming tool
I found that the best way to approach AI drafting is to start from defining the invention and use the Copilot as a “second mind” to brainstorm with (although you should proceed with caution – see paragraph below).
The job of a patent attorney is quite lonely, as we are used to draft alone. However, I know for sure that all the attorneys I have spoken with always dream of having a second opinion on their work, especially at the initial stage of drafting, when you need to really understand the invention and define the essential elements of it, to put in the independent claims.
An AI Copilot can help with this. It can provide ways to reframe the invention, see the problem from a different perspective and challenge you when you get stuck in your ways. I think that this is a very powerful aid for attorneys.
How can this work in practice? After uploading the invention disclosure documents (that you should, of course read and understand), you could ask the AI to summarise the invention, and iteratively make a list of essential features and important aspects to introduce in the independent claims. You can then instruct the Copilot to write the first independent claim and iteratively refine it.
At this stage, it is especially important to understand that you are in control and need to be critical of everything that is written. The AI Copilot is very good at writing, and, if appropriately trained, can be very good at writing like a patent attorney, but it cannot think like one. The thinking is still very much our job.
Be careful to the dangers of AI people pleasing
An important risk to be aware of is the pitfall of “people pleasing” in AI tools. I made some experiments with the Copilot, asking various types of questions, and checking the answers.
It turns out that my virtual assistant tends to answer in a way that would please me, even if I am wrong. For example, if I ask, “please find the errors and inconsistencies in this claim”, the chances of being told “there is no inconsistency in the claim” is probably very low, regardless of how the claim is written. If I ask: “do you think this sentence should be rephrased”? I will be provided some alternative to phrase the sentence in an allegedly better way, even if there was no need for that.
I have been experimenting with asking questions neutrally, and by adding small instructions such as “please, provide me with an unbiased response”. My impression is that, with appropriate rephrasing of questions, the quality of the answers generally improves.
Another trick I tested was to ask to provide reasoning behind a suggestion. For example, I may ask if the word “housing” is the most appropriate to identify a certain feature of the invention. I could ask the Copilot to present me with alternatives, highlighting pros and cons of each suggestion, and presenting the reasoning behind the recommendations. This approach helps greatly with minimising the risk of being overly influenced by the machine.
The Solve Intelligence AI Copilot
The Solve Intelligence AI Patent Copilot comes with a text editor in which a patent application draft skeleton is created based on the jurisdiction and the type of invention.
As an example, the assistant can, in principle, differentiate between US and European drafting styles, but can also adapt the drafting style dependently on the field of the invention (e.g., engineering, chemistry or software). So far, I only tested the engineering and software drafting styles, but I have not noticed any difference. However, I have not explored these features in depth, so I do not have a strong opinion, now.
The text editor can be sided with a chat used to prompt the AI agent, and to insert and/or correct text in the text editor. Text can be refined multiple times, and changes can be made in tracked form in the text editor. I find this to be a very nice feature, because it allows you to quickly view, at each iteration, where changes have been made in the text.
AI instruction tools and prompts
The chat is complemented by an “AI instructions” section. In this section, you can select prompts per each section of the patent application (title, background, field of invention, summary etc). These prompts can either be selected from a library of pre-packaged ones or created by you from scratch.
It is also possible to build a plan of AI structures for the detailed description. For example, let’s say that you have four figures and you want to organise the detailed description to describe each figure in sequence. You can build a plan, where you divide the detailed description in four sections. Each section can be linked to the relevant figure to be described, and you can also indicate which features of the figures and/or claims you want to describe in detail.
I find this feature extremely useful to write a first description of the basic features of systems and/or flow diagrams as illustrated in the figures. The Copilot removes the need manually type long paragraphs describing basic features illustrated in the figures, such as “system 100 comprises a processor 120, a communication interface 130…”). This is, in my opinion, a rather mundane part of the drafting process, which is also highly susceptible to typos and errors, so it is nice to have it done automatically. Then, I proceed to manually insert additional information, which I think can be useful, for example, to provide further support and clarification for particular features.
Of course, you can also ask the Copilot to introduce further, more detailed paragraphs. For example, you could ask it to write “a paragraph outlining examples of short-range communication protocols”, and the Copilot may come up with Bluetooth, NFC etc.
Prompt AI-enhancement function
A nice feature is the “prompt enhancement tool”, which can be used to refine a drafted prompt that you have written. Once the prompt has been enhanced, you can amend it and enhance it again, in an iterative process. This feature is very useful, because you can start with a very quickly drafted prompt and have it re-written in a much more specific and accurate way in a matter of seconds.
Patent drawings with Solve Intelligence
To make the most of the AI Copilot, you need to create the drawings with the dedicated figure generation AI functions. The program can then map the drawing’s labels into the description, and create ad hoc sections, describing what is illustrated.
I generally have a habit of creating a first set of drawings after the claims and before writing the detailed description, and I have maintained this order of steps whilst drafting with the Copilot. It is clear to me that the drawings help the Copilot to learn more about the context of the invention and can be used to refine the claims.
There are different ways of creating drawings in the Solve Intelligence AI Copilot.
The most basic approach is to create figures separately (e.g., in Power Point) and then upload them in the AI editor. You can either introduce labels yourself externally or do it within the AI Figure Editor (manually). There is a function for creating automatic labels with AI but, at least in my experience, it can hallucinate pointless or incorrect labels.
Another way to make figures is to create a patent-style technical drawing from an uploaded image. I tested this function, and I think that it can work very well for simple mechanical drawings, but the AI can misunderstand the nature of some features illustrated in the uploaded image and create incorrect reference labels. For example, a drawn hole may be interpreted as a button or the top view of a screw. I have also not found a way to interactively change details of an image created through this function. If you are not satisfied with the image created you must start again from scratch and hope that the AI gets it right the second time. Positively, three alternatives are given per each round, so that you can select the image or images that you prefer.
Another tool is the “Generate with AI” function, which allows the creation, from scratch, of drawings based on your own prompts (which can be enhanced by the Copilot itself). There are different types of drawings you can select: flowcharts, sequence diagrams, node diagrams, system diagrams. I think that this function works particularly well for computer-implemented inventions, and to create flow chart and sequence diagrams.
So, does AI save you time?
One of the main benefits of AI tools is that they reduce the time you spend on trivial and mundane tasks. For example, I spend virtually no time at all now for choosing reference numerals for my claims, and if I want to spot any typo or inconsistency, I can easily ask the AI assistant to look out for mistakes on my behalf. Whilst the application is still checked by me, the efficiency increase is notable.
However, I can say that these benefits do not necessarily translate into a net reduction of the hours spent on the draft. That can happen with simpler inventions, but it is not always guaranteed, at least for me.
One thing I am positive about is that the time I would have spent fixing typos, choosing numerals and so on can now be spent on more fundamental tasks, such as claim construction and overall drafting strategy. There is therefore a drastic redistribution of the time spent on different tasks, with more time and energy available for the most creative and important work. This is, in my view, the greatest benefit of AI-assisted drafting.
Will AI replace patent attorneys?
After practising with AI-assisted patent drafting for a while, I can certainly say that, whilst AI tools are excellent aids in the attorney’s job, they cannot, at least for now, work alone, and they are only effective thanks to the instructions of the attorney. This means that, at least for now, we can confidently say that our work as patent attorneys is not over.
Bibliography
(1) https://www.patentext.com/blog-posts/patent-drafting-Copilots-vs-ai-native-platforms
(2) https://www.buzzsprout.com/2317109/episodes/17539882-ai-in-patent-practice-a-reality-check
