Originally posted on IPKat.
Furthermore, whilst the focus remains on integrating systems from third-party providers for automating simple tasks, the transformative power of actual AI is being overlooked by many patent attorneys. The capabilities of today’s large language models (LLMs) are now so good that “putting the attorney in the driving seat” should mean nothing less than attorney-led sophisticated prompt engineering of the foundational LLM.
The patent profession emerges from the technological dark ages
Patent firms are notoriously slow to adopt new technology. Despite being literally at the forefront of innovation in terms of the clients we serve, the patent profession is remarkably slow on the uptake with respect to technological solutions for itself. There are numerous repetitive tasks in patent drafting, prosecution and portfolio management that could have been automated decades ago, without the need for AI. Simple, automatable tasks such as updating claim numbering and dependencies, figure labelling and citation management are just a few examples. Bizarrely, many patent attorneys still don’t even use reference managers to organise citations in their documents, even when many documents are being cited.
Within the biotech field, the way in which the patent industry manages the generation of sequence listings and sequence data is particularly egregious. We can remember well her incredulity when she entered the profession and saw how sequence listings were prepared, which involved copying and pasting individual sequences from the specification or data into a retro and highly temperamental JavaScript program. WIPO Sequence is a step up, but a small one, and many sequence listings are still prepared by copy and paste. However, sequence listings are highly structured XML files that can be easily generated from input sequences using simple code and FASTA files.
It is therefore clear that, before we even consider AI, the low-hanging fruit for the industry may be simple automation, especially for large firms and in-house departments handling high volumes of applications (many large in-house departments are way ahead of firms in this respect, having already built and adopted their own systems). It is therefore unsurprising that many AI patent software providers are now selling automation, albeit under an AI umbrella.
Automation versus AI
Whilst automation is welcome, it is critical to distinguish between automation and AI, and for firms to be clear on what they are actually buying. Automation follows a deterministic set of rules to produce a predictable output, such as renumbering a list. By contrast, LLMs are probabilistic engines designed for natural language generation and machine learning. Therefore, automation of routine tasks and the use of sophisticated LLMs can and must serve different purposes and solve different problems for patent attorneys.
That is not to say there is no value in automation. Sequence listings are a particularly good example of where automation is a far better solution than LLMs. LLMs are extraordinarily bad at dealing with sequence data, given that they are probabilistic based and so can make mistakes when copying sequences of arbitrary-looking characters. Additionally, the system prompts behind many of the foundational models generally instruct the models to simplify and summarise complex text.
Ask Gemini what its system prompt tries to optimise, and it may tell you that instead of handing you a dense “wall of text”, it optimises linguistic simplicity and structural clarity. Linguistic simplicity and structural clarity are not things that sequence listings have ever been accused of being. The sequences in a sequence listing must be identical to the sequence data, with a single nucleotide or amino acid substitution potentially rendering the entire patent useless because it no longer covers the product (Evolve insights). The tendency of LLMs to hallucinate or smooth over data could therefore lead to the introduction of silent, catastrophic errors into sequence listings. Unlike a dedicated automation tool that maps data precisely, an LLM might perceive a sequence as a string of text to be completed or balanced, potentially inventing residues to satisfy its probabilistic training.
So, you shouldn’t use LLMs to prepare sequence listings. The good news, however, is that you don’t need to. You just need code to automate the process. Code that could have been written two decades ago. AI patent software providers know this and are now promoting extensive functionality that clearly falls squarely in the category of automation, not AI. This includes sequence listing solutions, case management software, transferable templates, figure labelling, and automatic claim dependency and claim numbering updates, and claim trees. Anything that doesn’t require sophisticated understanding of language and context is probably an automated code solution, not AI, or at best an automated solution with a sprinkling of RAG and LLMs. Even many of the solutions that do use AI, lean heavily on code and automation to assist the user experience and integration within the software solution platform.
Why are firms buying automation now when these kinds of solutions have been available for decades from various providers? The answer is probably something to do with AI hype, FOMO and client pressure. It seems that automation and case management may be the easiest way to get on the “AI” bandwagon these days.
Are patent attorneys missing the true value of AI?
It is of course a good thing that patent attorneys are finally entering the 21st century, albeit 20 or so years after many other professions. We are fully in favour of automation for easy routine tasks. However, you don’t need sophisticated LLMs to do this. Furthermore, the focus of the AI software providers on tasks that are really automation masks a fundamental truth about the offerings of these companies, namely that you don’t need them to harness the incredible power of the current foundational LLMs. Worse, third-party solutions put an unnecessary barrier in the way of patent attorneys learning how to harness the true power of these models.
The foundational LLMs such as ChatGPT, Claude and Gemini are now so powerful that there is little to no value in placing a third-party provider between you and the underlying model in terms of the quality of the work product. If your focus is on increasing value of sophisticated patent work, the additional layers that third-party providers place between the patent attorney and the foundational LLM are not only unnecessary but also highly restrictive. The benefit that AI patent software solutions provide in terms of ease-of-use comes at the (significant) cost of preventing customisation. They also prevent patent attorneys from learning key skills for adapting to the new AI world we find ourselves in, particularly sophisticated and effective prompt engineering. Additionally, reliance on these pre-packaged solutions is inherently constraining, as the patent firm becomes effectively locked into the developer’s interpretation of a patent workflow, creating a black box environment where the professional is subservient to the software’s limitations rather than being empowered by its capabilities.
Analysis
As a recent report from Anthropic highlighted, “[w]hile Claude is able to respond in a highly sophisticated manner, it tends to do so only when users input sophisticated prompts”, highlighting the importance of skills and suggests that how humans prompt the AI determines how effective it can be. Patent attorneys wishing to use AI to add value therefore need to understand how to interact directly with LLMs and develop sophisticated prompt engineering skills rather than simply learning the interface of a single service provider. The true value from AI lies in incorporating these models into a bespoke workflow, tailored to a particular technical field, the needs of the client and incorporating and evolving with an attorney’s expertise. It is the difference between relying on public transport and learning to drive.
Author: Rose Hughes
Rose is a biotech and pharmaceutical patent specialist with over a decade of experience in intellectual property. Rose is a patent attorney at Evolve, where she leverages our unique fractional in-house model to provide clients with deep patent law expertise combined with the strategic commercial oversight typically associated with senior in-house counsel.
With a PhD in Immunology from UCL, Rose applies her technical background to complex innovations in biologics, cell and gene therapies, and the rapidly emerging field of AI-assisted drug development. Previously, Rose held the role of Director. Patents at AstraZeneca, where she was responsible for global IP portfolios and IP strategy at every stage of the pharmaceutical pipeline, from platform development and on-market commercialization to SPCs and patent term extensions.
A recognized thought leader in the field, Rose has been a regular contributor to IPKat since 2018, offering practical insights into European patent law developments. She is also a frequent speaker on the epi podcast, a guest lecturer for the Brunel University IP law Postgrad Certificate, and a contributing author to published books A User’s Guide to Intellectual Property in Life Sciences (2021) and Developments and Directions in Intellectual Property Law (2023).
