Inventors' Use of Artificial Intelligence Large Language Model Chatbots

I'm not really known for my hot takes. What I'm about to say isn't especially spicy, but it's as close as I will get to a controversial, timely post, so please bear that in mind.

While I am a registered patent agent, nothing that I'm writing here is advice for any client. I'm not a lawyer, and nothing in this post is legal advice. I'm providing this for informational purposes only. It should go without saying, but I'll say it anyway. Reading my blog post does not make you my client. The only way to become my client is to sign an engagement letter with me. If you do need help with a patent application, though, please don't hesitate to reach out.

Now with all of that in mind, I have definitely noticed inventors providing materials that come from artificial intelligence (AI) large language model (LLM) chatbots. These are products like OpenAI's ChatGPT, or Anthropic's Claude, or Google's Gemini, and so on.

My chief concern is that inventors who share information about their work with an LLM may prevent themselves from ever being able to get a patent to protect their invention. An inventor who shares information with one of these LLMs is sharing information with a party other than any co-inventors, or any patent professional they've retained to help with the process of prosecuting their patent application through the United States Patent and Trademark Office. In that sense, the company that made and owns the LLM is a third party. Sharing details of an invention with a third party may make it harder to get a patent that fully protects that invention - or even get a patent for that invention at all. Without going into all of the details, once an inventor has publicly disclosed their invention, there's a one year timer that starts running. Anything about that invention that doesn't make it into a patent application before that one year timer runs out might not be patentable anymore. It isn't yet totally clear if chatting with an LLM counts as disclosing an invention. (And, again, I'm not a lawyer.) But in United States v. Heppner, Judge Rakoff of the Southern District of New York ruled in February that chatting with an LLM was not protected by attorney-client privilege. In January, the US District Court for the Northern District of California dismissed a trade secrets violation claim. The plaintiff accused OpenAI of stealing her trade secrets. But Open AI pointed out that the plaintiff had shared those secrets with ChatGPT, OpenAI's LLM. Neither of these cases is directly about invention disclosure and patents, but I think inventors would be wise to keep details of their patents away from LLMs.

A secondary concern surrounds the costs for an inventor to get a patent for their work. Inventors need to keep in mind that if they retain a patent professional to help with patent prosecution, that patent agent or attorney is ethically obligated to fully assess all materials they receive from the inventor. I can help inventors save on costs, because my rates are far below market rates. I also am not early in my career, so I have a lot of skill with technical writing. I can read, understand, develop arguments, and write quickly. I also have two little kids, so I need to get through my work quickly so that I can get to the next soccer game. All of that helps me keep costs for clients under control. But if a client sends me a 20 page, bullet-pointed document with descriptions of their invention, I am obligated to read all of it. I have to consider all the information in that document, even if it's not meaningful. I have to write out my thoughts and share those with my client. And all of that takes time, which means increased costs for my client. So clients who are trying to avoid raising the cost of the patent application process might also want to avoid using LLMs.

It is really cool that LLMs can make a color-coded spreadsheet with multiple worksheets explaining different aspects of an invention, all based on scraping text from a website and a couple of emails that an inventor feeds the model as input. But the LLM does not understand your invention. The LLM does not understand the patenting process. It does not know what aspects of your invention are novel or non-obvious. It does not know how to write a claim, let alone a "strong claim". (Yes, I have really seen LLM output purporting to present "strong claims" about an invention. Spoiler alert: they were not.) All of that is pretty cool. But for me as a patent agent, I'm going to have to read everything, take it seriously, and then spend time explaining to my client, in writing, why the LLM is wrong about the patentability of aspects of an invention. I can do that work, but it will end up costing my client money.

Speaking only for myself, I would rather receive just a short description of the invention directly from the inventor in the inventor's own words. Even if it's only a paragraph long! That is more helpful to me than 20 pages from an LLM. I can use that brief information from an inventor to get through the patenting process faster and at lower cost than if the inventor were to inundate me with LLM output.

Maybe inventors are thinking about their limited time and trying to save some of it by using a LLM to put together some writing quickly. It's true that writing can be time-consuming, and generating writing quickly is a major strength of a LLM. And maybe the inventor's time is so valuable that they're willing to save a couple of their hours, and then pay for several of my hours parsing LLM output. That might be a sensible tradeoff for the inventor. I hope that any inventors who are making that tradeoff are doing so consciously, well-informed of the time and costs involved. I suspect that inventors are not yet being so deliberate about assessing and choosing those tradeoffs. So, inventors, I urge you to please be aware that it will take more time from your patent professional if you share LLM output with them. That time will have a real, calculable cost to the inventor.

Another possibility is that inventors might think they can't do a good job describing their work to a patent professional. Perhaps they're worried about missing details or emphasizing the wrong details. This reminds of a common situation I encountered back when I was still working in academia. Graduate students have to defend their work publicly, in front of an entire department of professors who are world experts in their field. That can be a daunting prospect. Naturally, some grad students would feel really nervous heading into their public defense. I would remind students that, sure, their audience might know more biology/geography/whatever, but no one in the room could possibly know more about their own work. No one knows more than the grad student about how samples were collected or prepared, how the data were analyzed, or why they made those choices. This is true for inventors, too. No LLM is going to be able to tell an inventor something about their invention that they don't already know. So, inventors, once you've retained a patent professional, I encourage you to just use your own words. Even if you can't pump out pages of writing, your own language will be better than a Large Language Model.