Limitations of AI in Patent Searching

On November 30, 2022 ChatGPT was released to the public.  Immediately, the world went into a frenzy and many AI tools started popping up in industries that no one would have imagined.  Yet, there is at least one industry that’s been on the AI scene for years, and has been slowly but surely making advancements that have already significantly impacted the profession.  If you’re reading this, and you looked at the title of the article, then you know what we’re referring to – patent searching.

We’ve seen the development of many types of AI search tools in different shapes and forms.  For instance, some traditional Boolean search engines have added semantic / natural language processing search fields to their search engine.  A paragraph (or even an entire document) describing the invention can be input and the AI algorithm will analyze the important terms and phrases, and spit out a list of, what it believes to be the most relevant patent documents.  Typically, the results are, in fact, not always the most relevant, so Boolean operators are then used to filter and improve the results.

Stand-alone AI search tools are being developed as well, where the tool accepts the invention description and provides the algorithm’s most relevant results.  Here too, the tool will generally provide means for improving the results, such as giving a thumbs up or thumbs down to the hits depending on how relevant or non-relevant the document is.  Following this feedback a second iteration of the AI search is run, with hopefully better results.  This can repeat itself and the results will continue to improve along the way.

The one thing these different search tools all have in common is that the initial results are generally not the best.  While some results on the list might be appropriate, a large portion will likely be irrelevant.  Human input is necessary to actually make the tool useful.  Such tools would be most beneficial in providing an initial thrust to a search by finding some relevant documents to start with, or as a closer of the search to see if, by chance, any relevant documents missed in the traditional search would be found by the AI tool.

Additionally, stand-alone search tools often only have access to patent data, and do not search through non-patent literature.  For Patentability and invalidity searches, this can pose a serious risk of missing relevant prior art.

But all of this makes searching inefficient for the professional search provider.  The professional searcher is typically familiar with one or more traditional search engines, its features and best search practices.  By adding another step to the process, whether at the beginning or end, it means more time must be spent on the search.  In an industry where quick turnarounds matter, using the extra tool would be disadvantageous.

Instead, these types of AI search tools could be useful for the non-professional searcher, such as an IP lawyer, R&D manager or inventor.  In some cases, these individuals might want to perform a patent search without having to spend the time creating a list of keywords or classification codes, or to figure out the best Boolean operator combinations.   They might have multiple searches to perform and also see this option as an inexpensive way to handle their search needs.

Of course, there are certainly searches that AI can help perform, even without any human input.  The Landscape search typically aggregates information based on categories including industry, competition and geography.  The data is then analyzed for the purpose of competitive intelligence, R&D, company due diligence, and more.  Unlike Patentability, FTO or Invalidity searches, which get into the nitty gritty inventive aspects of a product or invention, Landscape searches are meant to view the results from a high level of detail.  They show technology trends, white spaces, competitor portfolio, etc. so the requirement for human input to refine and filter the results is reduced.  As such, the AI search tool is more effective in providing meaningful results quicker and more efficiently for this type of search.

At the end of the day, AI capabilities are improving at the speed of light.  Even though perhaps the technology is not yet there to replace a human searcher for certain searches, we can be sure that it won’t be too long until it’s reached that point.