When I worked as a patent attorney, I often pondered the sustainability of our industry – not in terms of the environment, but in terms of the scalability of our practices and the quality of outcomes. Multiple indicators suggested a concerning trajectory: a surge in what could be termed as "crappy patents," especially in Asia, accompanied by lengthening patent specifications and prosecution times.
While there have been some positive trends in the last years, the core concerns still exist. An example of this fact is that some European corporations founded an Industry Patent Quality Charter in 2023, to support and “ensure that patent quality in Europe and through the EPO is the highest in the world”.
Given this background, I've welcomed the emergence of patent drafting tools, particularly those incorporating generative AI, with mixed feelings. While still in an early stage of adoption, these tools hold immense potential to revolutionise the landscape of intellectual property protection and utilisation in business.
However, this advancement is not without its concerns, both for individual cases and, notably, the future of the entire patent system. Now is the time to discuss the consequences and chart a sustainable course forward. At a minimum, practitioners must remain vigilant about their methodologies, tools, and the potential implications they carry.
In this blog post I aim to contribute my part to this discourse by delving deeper into the topic, with a particular focus on patent drafting and examination – recognising them as pivotal aspects of the discussion.
AI Assisted Patent Drafting – a SWOT Analysis
Let's begin on a positive note. Artificial intelligence boasts two significant strengths that have already demonstrated their value in the patent space.
Firstly, it enables more efficient prior art searching, utilising e.g. claim drafts as queries. Computers are increasingly capable of understanding both technology and the patenting logic. Secondly, AI exhibits the capability to generate reasonably high-quality patent text following given instructions.
I have elaborated on these aspects, along with the interplay between specialist AI (Graph AI) and generalist AI (LLMs), in a previous blog post.
The strengths open up new opportunities. Both information retrieval, or search, and text generation capabilities are of utmost importance in next-level patent drafting.
Seamlessly integrating search into the drafting process offers a distinct advantage. Understanding prior art in real-time as the draft evolves enables maximised value for clients. The era of "drafting in the dark", where patentability barriers are envisioned without factual basis, becomes obsolete.
Once the correct scope of claims is established, whether with or without AI assistance, patent drafting efficiency increases. Attention shifts from editorial tasks to defining the subject matter, and fewer iterations are needed in putting the specification and drawings together.
Moreover, every patent attorney will sooner or later encounter a stage where drafting becomes routine, prompting the question of automation feasibility. Increasingly, the answer leans towards "yes." Even out-of-the-box LLMs, while lacking the finesse of a seasoned professional, can undertake substantial groundwork.
An opportunity worth mentioning is also that with advancing tools and declining costs, protecting IP with patents becomes more accessible even for smaller companies. However, the landscape remains largely dominated by enterprises with substantial resources.
On the weaknesses front, it's important to recognise the imperfections of AI. No search model can capture all the most relevant prior art in real time. However, it often begs the question: is it preferable to have some relevant prior art available than none at all?
Furthermore, concerning Generative AI, the risk for errors and hallucinations is real. Hence, users must exercise discretion and familiarise themselves with their tools.
These weaknesses give rise to significant threats that demand our attention. Firstly, the ease of generating lengthy patent applications with minimal substantive content poses a risk of flooding the prior art space over time. This could ultimately undermine the functioning of the patent system, making it increasingly challenging to distinguish genuine intellectual contributions from computer-generated noise, potentially leading to longer prosecution times.
Secondly, there's a risk of over-reliance on AI outputs, leading to incorrect judgments and the potential for wrongly defined subject matter, irreparable errors in applications, and loss of rights. Balancing the benefits of AI with judicious use is essential. Moreover, reckless use, such as making R&D or business data too openly available for AI, could result in the leakage of sensitive business secrets.
Thirdly, the proliferation of low-quality patents could devalue patents as a trusted source of technical information. This erosion of trust would undermine the patent system's core purpose of incentivising and advancing technical progress for the betterment of humankind.
AI-Assisted Patent Examination
Ultimately, the responsibility for maintaining the quality of granted patents lies with examiners at the patent and intellectual property offices, who serve as gatekeepers in the process.
In terms of opportunities and threats, the examination process appears more straightforward. Drawing from our experiences with various Patent and Trademark Offices (PTOs), integrating AI into the prior art search toolkit as a "citation candidate provider" notably enhances examination quality. Benefits include capturing documents inaccessible through Boolean searches and gaining insights to refine searches using new classification codes and synonyms.
However, risks arise when excessive reliance is placed on AI. Blindly accepting AI recommendations without fact-checking can be detrimental. Thus, it's crucial for examiners to retain control over complex tasks such as approving patentable subject matters.
Nevertheless, there are less critical areas where AI can soon achieve sufficient accuracy for relatively autonomous decision-making, with IPC/CPC classification being a notable example.
Ideas for a Sustainable Patent System
To wrap up, I would like to share a few ideas and suggestions, spanning from the practical to the unconventional. How can we optimise the advantages and address the risks that I outlined above?
Patent applicants and attorneys:
- Leverage AI to craft higher quality patent applications. Strive for concise, well-defined applications that accurately address prior art, avoiding lengthy, low-content submissions.
- Utilise AI to ensure system quality and operational legitimacy. Embrace new convenient opportunities in competitive intelligence and active tools to prevent low-quality patents from entering the pipeline, through methods like third-party observations and oppositions.
Intellectual Property Offices:
- Consider implementing hard limits or additional fees for lengthy patent applications to prevent the flood of AI-generated nonsense. Simplification is key; if an invention can't be adequately described in around 20 pages, the background work may be insufficient. Clear and concise patents benefit everyone.
- Incentivise applicants to engage more deeply with the process. Explore the concept of an "Office Action 0," automatically generated by AI, which prompts applicants to address prior art or amend applications before proceeding. The novelty-destroying prior art recall level of AI is now at a level where this starts to make sense. This would have a positive impact on the queues and prosecution times.
- Use AI in searches! This bullet point needs no explanation.
Webinar on Patent Quality – Join Us with LOT Network
Watch our on demand webinar hosted by our colleagues at LOT Network discussing the quality of patents in the AI era, further exploring the SWOT analysis outlined above. Watch the webinar here.
.webp)
.webp)
