Back in 2021, I (Yixin) wrote down some thoughts on how the emerging "age of artificial intelligence" can change legal work and make lawyers more efficient. AI development has grown by leaps and bounds since then, and some of the systems are being "field tested" by patent lawyers. Dr. Van de Wiele, the co-author on this article, is a co-founder of an AI development company called ClaimWise.
1. Introduction
In the ever-evolving field of patent litigation, efficiency and accuracy are paramount. The process of analyzing patent infringement or invalidity can be labor-intensive and prone to human error, requiring meticulous examination of claim elements, product features, and prior art. As technology advances, Generative Artificial Intelligence (GenAI) emerges as a promising tool to revolutionize this process. By automating the time-consuming aspects of patent analysis, GenAI offers the potential to accelerate legal workflows, enhance precision, and uncover more legal arguments. However, the adoption of GenAI in patent litigation is not without its challenges. This article explores how GenAI can be effectively utilized in patent infringement and invalidity analyses, addressing the critical concerns raised by the patent litigation community to ensure its successful integration into the legal landscape.
2. Using AI in Patent Infringement Analyses
Any "human-independent" AI-powered solution has its challenges when applied to patent law. Let's suppose GenAI understands the technical aspects of a product and can search for all patents and patent applications that might cover this product to generate a claim chart. A first hurdle is that words in a patent do not necessarily mean what they appear to mean. Under U.S. patent law, patent claims are construed by the courts. Because patent coverage is often different from the "plain and ordinary meaning" of the claim language, it can only be elucidated through the process of claim construction. As AI technology advances, a large language model (LLM) might consider all case law, all potential interpretations of words by a person of ordinary skill in the art (PHOSITA or POSA), and also intrinsic evidence like patent disclosure and the prosecution history. (An LLM is a deep learning algorithm that can recognize, summarize, translate, predict and generate text and other forms of content based on knowledge gained from massive datasets.) Yet, at the end of the day, GenAI would not be allowed to render legal opinions without human intervention. A human attorney must step in at some point to verify the results.
For a human attorney to check AI-generated work, so to speak, to sign off on a patent infringement analysis, the attorney must at least retrace AI's analysis to ensure that the parsing of the prosecution history, the disclosures in the patent specification, and the court opinions are all correct. No hallucinations are allowed. The attorney must ensure that the AI did not miss any relevant information that would have been found by a human attorney expending a reasonable effort. GenAI should always facilitate this process by providing accurate and transparent references to the patent disclosures it uses to generate claim constructions.
3. Challenges with using AI in Patent Invalidity Analyses
The caveat outlined above (a requirement for human intervention in patent analysis) also exists in any human-independent patent invalidity analysis. The universe of information is vastly expanded in an invalidity analysis. All publicly available information, and all background information in the relevant discipline or disciplines, must be considered. For a scientific or engineering publication that has its intended audience other practitioners in a particular field of art, there might be very little background information in the paper itself. The background information may derive from "experience" – things that are known to practitioners in a field of art but not written down and published. The publicly available information also includes publications that are behind paywalls, or textbooks that are not easily obtainable in electronic forms. Currently, GenAI cannot gather all the relevant and publicly available information for a patent invalidity analysis. It is important to acknowledge, however, that experts in the field already use AI-driven search engines to gather prior art, regardless of their coverage. Most in- house IP departments at innovative companies tend to have access to databases that utilize GenAI to retrieve relevant prior art based on semantic relevance. This powerful feature accelerates the manual aspects of prior art searches (reading, skimming,) and increases the relevance of the prior art found. Although the data coverage may not always include analog documents or highly specific publications, GenAI can eliminate the time-consuming task of scanning hundreds of documents by instantly extracting the most relevant paragraphs.
Another issue for using GenAI in patent analysis is potential
over-inclusion of prior art references. A POSA in patent law is
presumed to know all the prior art in the relevant field. With the
explosion of on-line publishing, self-publishing including
blogging, and posting in on-line forums, AI can easily gather,
collate, and summarize such electronically published information.
Currently, the test for whether to include such information as
prior art is human-centric, i.e., how available such information
was to a POSA, and whether POSA would regularly access such
information. Under such a standard, unless and until AI is
regularly used by POSA, the information may fall within a gray
zone: it can be found by AI, but it still might not be prior art.
Furthermore, even if a particular AI tool was available at the
relevant time, the constantly evolving ability of any type of AI
makes it easy to argue that the POSA would not be able to find such
information from a myriad of on-line sources. Notably, the USPTO
has requested
comments to better understand how widespread the use of AI tools is
and whether a POSA should be assumed to have access to these tools.
While this may vary by industry, the availability of free LLMs like
ChatGPT or Gemini have certainly rendered AI readily accessible to
anyone. This suggests that the definition of the POSA may need to
be revisited to acknowledge the
already widespread use of AI in searching for prior art.
The over-inclusion problem might be more pronounced in analyzing "motivation to modify or combine prior art teachings," and "reasonable expectation of success." Such "TSM" (teaching, suggestion, motivation) information is rarely found verbatim. The prior art reference would not say "modify A to include B, to make X," with X being the exact claimed invention in a later litigation." Oftentimes one would see something like, "there is a problem with A, and perhaps adding something like B, C, D, E, ..., N, O, P could help," and then another reference would discuss some useful properties which B has. In this example, if a technical expert then uses the AI-found and AI-analyzed TSM to opine on the obviousness of a patented technology, the opposing litigant could probably make a convincing case that such TSM was not in fact available to POSA, because it was buried in too much other information, and was too obscure, that the human expert could not even find it without resorting to the help of AI. In this scenario, once a technical expert has seen AI-generated TSM information, he or she cannot unsee it, and the expert's effectiveness in the lawsuit could be undermined. We have not seen a test case on this point, and the USPTO is gathering comments on how the availability of AI tools should affect the legal definition of POSA in a particular technical field of art. The law is unclear, and this may limit litigants' willingness to use AI to search for and analyze prior art.
4. How GenAI revolutionizes the process
Despite the concerns raised above, a patent litigator using GenAI can accelerate an infringement analysis significantly and identify more legal arguments. By providing an LLM with contextual information such as claim infringement charts, models are already able to replicate most of the steps inherent to an infringement analysis, learning from the prior work of patent professionals.
First, GenAI significantly reduces the manual work inherent to an infringement analysis by analyzing the patent. With GenAI, patent litigators no longer need to spend hours reading dozens of pages, scanning patent descriptions for disclosures, extracting product features from evidence, or summarizing disclosures to identify interpretations of claims. Unlike patent professionals, who might take 5-6 hours for this type of manual work, the model can immediately grasp the semantic context of the claims and complete these tasks in seconds.
Second, GenAI decreases the likelihood of human error inherent
in text analysis and interpretation. Litigation disputes often
hinge on the interpretation of specific words in a claim element. A
correct interpretation of the word could be outcome dispositive.
Human execution of this interpretation process involves risk and
creativity, leading to potential errors in dividing claim elements,
misinterpreting disclosures, or overlooking product features in
infringement evidence. GenAI, leveraging the power of neural
networks, can identify more relationships between words than humans
can in a fraction of the time. Its ability to identify
textual
relationships and generate various claim constructions makes GenAI
an excellent tool for analyzing claims while simultaneously
minimizing the risk of errors and missed information. To address
the concern about hallucinations, it is important to distinguish
generative capabilities from augmented retrieval capabilities. By
providing an LLM with more context (e.g. prior examples), the
information it retrieves will also become more relevant and
diminish the possibility for hallucinations. Solutions can provide
an initial interpretation of a claim construction based on the
information specifically provided by the patent text and
prosecution history. While the end decision will indeed lie with
the litigator, the litigator will have spent their time more
usefully on formulating legal arguments and interpreting claims
rather than searching for, reading or studying patents.
Third, reducing manual work and limiting human error with GenAI frees up time that can be devoted to more legally consequential issues. For instance, using GenAI in an infringement analysis allows litigators to focus on developing legal arguments. By automating the task of fetching disclosures or product features, more attention can be given to studying particular textual relationships for infringement arguments.
In summary, if the results generated by GenAI and humans turn out to be substantially the same, the legal profession may accept AI-generated analysis. Patent litigators should implement GenAI to accelerate legal processes and retrieve transparent, accurate information. Identifying information through semantic overlap is often the most time-consuming step in an infringement analysis and should not be overshadowed by the more consequential but less time- consuming task of identifying legal arguments. However, to get to that point, enough lawyers must be willing to: (1) both run AI and conduct human analysis; (2) compare results; and (3) disclose such results publicly so that the general "reasonableness" of an AI-driven analysis can be established.
5. Conclusion
In conclusion, while GenAI holds the potential to significantly
accelerate and refine patent infringement and invalidity analyses,
it is essential to address the concerns raised by the patent
litigation community. GenAI can reduce the manual workload,
minimize human error, and enhance the efficiency of infringement or
invalidity analyses by leveraging its capabilities in semantic
analysis and information retrieval. However, the integration of
GenAI must be approached rationally, ensuring human oversight to
validate AI-generated results and mitigate the risks of
under-inclusion and over-inclusion of prior art. As AI technology
continues to
evolve, its role in the patent litigation process will likely
expand, necessitating ongoing dialogue and adaptation within the
legal framework to fully realize its benefits while maintaining the
integrity and reliability of patent analysis.
The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.