No matter the industry you work in, everyone has been talking about AI (Artificial Intelligence, of course, though I prefer to use the term LLM – large language model). To hear some tell it, a massive, world changing technology that will lead to a new era of peace and prosperity by eliminating drudge work from our lives. Get on board, or risk getting left behind like the luddites of old, with only a 10% chance that the world will be destroyed (note the presence of both carrot and stick). Then there are the critics who find AI to be merely not very impressive at best, and a tool of popular disenfranchisement, surveillance, and martial destruction at worst.
Is it appropriate to use an LLM to draft a patent application? Well, what is involved with drafting a patent application in the first place? Knowledge of the law, certainly. An understanding of the technology involved. Expertise in the complex intersection of these fields that, for most people, are held apart. Long periods of sustained concentration as an Attorney carefully crafts a document that accurately describes an invention while also carving out legal protections. At first blush it seems like a task that is prime candidate for a machine that contains all existing human knowledge. For inventors without legal expertise, using an LLM to write your patent application can seem like a thrifty way to avoid the fees of an Attorney, literally cutting out the middleman. For an Attorney, it can be tempting to ask an LLM to perhaps provide a first draft, or to provide editorial suggestions (all under expert supervision, of course), allowing you to focus on client relationships and the interesting, high-level work of strategy and prosecution. However, as you may discern from the title of this article, there are potential pitfalls to consider.
Pitfall 1: Novelty destroying Disclosure, Privilege, and Confidentiality
First of all, there is the matter of privilege and disclosure. My colleague Peter Matalanis recently wrote about this so I will direct you to his insights rather than repeat them. In brief, there is a distinct possibility that anything uploaded to an LLM could be considered a disclosure of the invention, which could impact novelty considerations. This clearly extends to the process of drafting, because the LLM must have the working principles of the invention as input to produce any written material. Such a disclosure may not be unearthed during examination, but should the invention become involved in litigation then LLM records may become available to a determined opposing party. LLMs are very impressive, but can they write a specification for an invention that they do not have information about? Safeguards are possible but they should not be assumed.
Pitfall 2: False Data and Hallucinations
The ability for an LLM to create a reference from whole cloth is well know at this point. The issue is well documented in both academic writing and in submitted legal documents. LLMs basically work by generating the next ‘packet’ of text (or image or video, etc) based on what it has seen previously. This can occasionally result in some hybridisation of ideas or texts that are problematic. The reason that something incorrect can slip past an expert supervisor like an academic or lawyer is that the output produced by an LLM is very detailed and confidently presented, even when incorrect. Judging from the number of occurrences, it seems all too easy to be lulled into accepting something that looks real and true. Quite clearly, the presence of false or misleading data and citations is something that should be avoided in a patent specification, where unsupported claims and impossible data can quickly cause an application to unravel under scrutiny. The best antidote against this when using an LLM is close scrutiny of any output, no matter what pressure you may be under. Double, triple check any statements of fact or conclusions proposed. Even better, know before giving the LLM an input what the outcome should look like and be very sceptical of anything that deviates from your expectations.
Pitfall 3: Professional Ethics and Obligations
The above pitfalls lead us to this third. Patent Attorneys, like most professional service providers, must uphold a standard of professional ethics. In Australia this is governed by the Code of Conduct for Trans-Tasman Patent and Trade Marks Attorneys 2018, as well as the IPTA Code of Ethics. The main relevant obligation from the Code of Conduct is 12: Responsibility. As an Attorney, one is responsible for one’s own work, acts and defaults, as well as those of associated persons. If part of your work is outsourced to an LLM, and it makes an error of the kind described above (or otherwise), you as the Attorney are still responsible. This may seem obvious but the convincing and frankly obsequious nature of the LLM output can make error checking quite difficult. This leads to 14: Competency, in particular provision 14(2): an Attorney must carry out work with due skill and care. To what extent does merely checking the output of an LLM as opposed to drafting a specification oneself meet this standard? What is the nature of the skill of drafting for an Attorney: is it as a writer or as a content supervisor? On the other hand, it may be possible to argue that if one is not making use of the most powerful available tools, then that is not showing sufficient skill and care. These are difficult questions without clear answers.
Pitfall 4: Personal Ethics
The professional ethics and obligations as mentioned above are a minimum threshold but not the only considerations. Every person is guided by their own morality and ethics, and it is fair to say that the rise of LLMs is challenging to many. It may be that the first two pitfalls are merely technical challenges that LLM engineers will eventually overcome. Other considerations, such as the environmental or political, for example, are stickier. One must consider what kind of person and what kind of professional one aspires to be, and to act accordingly.
Pitfall 5: Other Considerations
Researchers collaborating across a number of world-renowned universities recently released results of a study showing that after just 10 minutes of AI use a person’s ability and, crucially, willingness to problem solve are reduced. Similarly, it was remarked by author David McCullough that “Writing is thinking. To write well is to think clearly. That’s why it’s so hard.” Drafting a patent specification is an act of extended problem solving carried out through the written word. The act of writing is, in my experience, inextricable from the act of problem solving in this way. Indeed, it was through the writing of this article that I learned many of my own opinions on this topic and discovered a way to express them that I had not previously had access to. It has been common in drafting patents to finalise a comprehensive set of claims, only to get half way through writing the description and find that they are not as good as they initially seemed. This is linked to a similar phenomenon I encountered when working as a university tutor many years ago: true understanding of a topic only comes when you have to explain it to someone else, in writing or otherwise. What insights and crucial detail may be lost when we draft an application using an LLM, because we did not fully engage in the act of writing? The brightest minds are often the ones best able to delude themselves of the thoroughness of their own knowledge within the comfort of their own cranium. Thoughts must be released, and clearly expressed, to have an impact. Does supervising the output of an LLM give results in the same way?
Conclusion
To draft a patent specification is to will into existence a document both legal and technical that must stand up to often intense scrutiny. The first examination doesn’t happen at the patent office, it happens between the mind and fingertips of the drafting Attorney. It may be possible for an LLM to act as a useful assistant in this process. However, LLMs are vast, complex, powerful machines designed to make you want to use them, and high levels of care must be taken.
Have any of the above pitfalls made you reconsider how you approach and use AI? Do you have any questions about your innovations and IP protection? Reach out to one of our patent attorneys via mail@mbip.com.au for professional expertise and advice.
