“We anticipate that even future LLMs will have the same drawbacks as new employees when it comes to drafting claims, but will not benefit from training new patent practitioners.”
Like most patent attorneys, I receive multiple emails each month about artificial intelligence tools that purport to help patent attorneys draft patent applications. I've been demoing it, and I have no doubt that in five years' time almost every patent drafting practitioner will be using these generative AI tools in some way. However, I also believe that these tools are not particularly useful for drafting claims.
I'm not anti-technology when it comes to patent drafting.
I'm not afraid of losing my job to technology, so please don't take this article as a backlash against generative AI in the trash. In fact, I look forward to a time when all the tedious work of patent drafting and prosecution will be done by helpful technology. I already use some AI tools for certain aspects of patent drafting and proofreading. If these tools were taken away from me, I would feel like I was reversing out of the driveway without a backup camera. Not too long ago that was the norm, but now it feels irresponsible. It took a bit of a learning curve to figure out how to best use these, but I now feel that working without such tools is inefficient and error-prone.
AI can be an inexperienced employee, but can and should it serve more than that?
I'm not saying that generative AI tools can't create claims that pass the sniff test. There is no doubt that AI can output well-structured claims that include the elements of the invention. Like almost every patent attorney I've worked with throughout my career, he had drafted usable claims within a year of drafting a patent application.
These claims describe inventions provided by the customer. These claims covered preferred embodiments. Those claims were grammatically correct. However, these claims were always heavily revised under the supervision of lawyers and took years for me to fully understand.
Every term used in a claim and its position within the text is important. Patent infringement lawsuits can be won or lost by relatively trivial wording issues. To get the best value, your claim wording needs to be accurate in the right areas.
Some parts of the claim need to be set out in a broader way than others. If the claims are unduly broad in terms of novelty, at best the examination will be lengthy and at worst patentability may be impossible. If a claim is too narrow in scope in terms of novelty, it is worthless because it can be easily designed. If the claims are too broad in describing conventional aspects, patentability may become more difficult as the examiner may be searching into areas of technology outside the client's anticipated scope. If your claims are too narrow when describing traditional aspects, your design may also be too simple. These aspects are typically overlooked by new patent attorneys, but are learned through years of experience and evolving preferences.
Let's be honest: Inexperienced employees often don't save the time that experienced patent practitioners have in drafting patent claims. The act of an experienced practitioner rewriting an inexperienced practitioner's argument can be more inefficient than starting from scratch. Experienced practitioners find a balance between deciding which languages to keep, which languages to change, and which languages to discard (even if they are stylistically different from the language preferred by experienced practitioners). try to find. It can be a complex task due to the intersection of technical and legal terminology. The wording must be as precise as technically possible, taking into account the legal rules regarding claim interpretation.
From my experience demoing AI tools, they seem to fall into the same output bucket. It seems unlikely that experienced patent drafters will generate claims that provide sufficient value. Perhaps I'm underestimating future advances in this field, or I might take a different view if I were actually using these tools for my clients, but even for a future LLM, I'm not sure if I'd be able to get the billing right. We expect them to have the same drawbacks as new employees when it comes to drafting patent practitioners, but without the benefit of training new patent practitioners.
In contrast to new employees, we do not support AI professional development. You can also train the model, but the results are not very satisfying. I would rather have an experienced colleague who can teach me the art of writing claims language. I would rather help someone build a career, albeit one that is aided by evolving technology. Rather, I want to help people who can immerse themselves in the craftsmanship of thinking deeply about how to solve difficult problems that have no single correct answer. I love drafting claims because it is impossible to draft perfect language and it requires professional evolution to reach mastery. Without sounding too cheesy, I like to share this experience with others.
Humans' sophisticated way of making claims is difficult to imitate.
Drafting claims should always balance patentability/invalidity with infringement. Almost all of patent law is on a spectrum, with few things clearly black or white. Whether the claim language is sufficient to obtain a patent may depend on the ability of the examiner and patent attorney. Additionally, there are very few patents that are 100% valid or 100% invalid. It depends on the budgets of the parties, their attorneys, expert witnesses, judge, and jury. Almost everything depends on many factors that are not simple. The patent attorney's job is to make the best possible arrangements for the client, but perfection cannot be achieved.
We are trying to thread a tough needle in this regard, but which direction we lean in may depend on the client's preferences and the patent practitioner's past experience. As a theoretical example, an attorney may draft an independent claim that has a 90% chance of being patentable and valid, but only a 10% chance of actually being infringed; A lawyer may draft a claim that has a 10% chance of being patentable and valid. Patentable, but 100% likely to be infringed.
These differences may be the result of input from the client, but most often they are based on practitioner experience and questions asked by the practitioner. These are often based on years of reviewing competitors' patents and using the United States Patent and Trademark Office's (USPTO) “broadest reasonable interpretation” rubric (i.e., “reasonable” is not too broad. It is based on an intuition gained from understanding how the world (with various interpretations) understands the scope of the real world. examiner's scope) applies to patent applications. In some cases, it's based on subtle comments the client has made over the years, which I can't imagine being able to replicate with AI.
Two terms you often hear in the patent profession are “litigation-quality” patents and “low-quality” patents. This requirement gained further attention with the passage of the America Invents Act (AIA) and the creation of the Patent Trial and Appeal Board (PTAB), which invalidates patents. However, as noted above, a great deal of uncertainty exists in patent law due to the limitations of the language and the subjectivity of interpreting claim language in light of prior art. It is natural for two people to disagree about whether an invention is obvious or whether it is directed to an abstract idea (whatever that means). If you want certainty, stay away from patents and patent law.
Why am I making this point? Because I expect generative AI to be most useful for tasks with a higher degree of certainty. These tasks include drafting parts of the specification that relate to aspects other than novelty, such as a well-defined technology background description.
Certainly, you can adjust your AI model to account for uncertainty. But while this is deep work, more certainty is shallow work. Deep work involves balancing a wide range of inputs to achieve the client's desired level of patentability/invalidity/infringement while also considering the client's business goals. (Read on if you're interested in a breakdown of this topic) Deep Work: Rules for Staying Focused and Successful in a World of DistractionsNewport, California, January 5, 2016).
You can enter a transcript of an inventor or businessman, but it lacks emphasis, ignores body language and conversational tone, makes assumptions about coined words, and makes important business and marketing mistakes. Revealed coffee conversations will be ignored. need. It also ignores what level of patentability/invalidity/infringement ratio is desired.
Achieving this balance can be very difficult. That's why, as an inexperienced patent drafter, I couldn't understand it, and my mentor at the time couldn't explain it in enough detail. It requires human experience.
Leverage AI to avoid being replaced by AI
And frankly, this is what makes this job so rewarding. Even inexperienced patent attorneys can obtain patents. However, this experience is necessary to obtain the patent that comes closest to achieving the balance desired by the client, especially if this desire is not clearly telemarked.
I'm not discounting the potential for generative AI tools to help with invoicing. Tools could be developed to compare claims and invention disclosures and ask whether different wording would be helpful. Alternatively, the tool can perform a prior art search using the claim language and suggest differences to the user. These tools are more interactive and help you go deeper into refining your claim language rather than creating an initial draft of it.
Therefore, I would like to see AI further developed to remove shallow tasks and help with deep tasks. Deep work is what makes it fun. We, the humans who draft patents, will need to continue to improve through deep work to fend off generative AI. If you don't put special effort into asking the right questions, thinking about your business needs, anticipating claim interpretation, or drafting strategic independent and dependent claims, generative AI is likely to be replaced by
Image Source: Deposit Photo
Author: Albando
Image ID: 589653746