There has been a lot of discussion in recent years around how the patent system can be applied to, and indeed may need to be adapted in light of, artificial intelligence and related technologies. Indeed, our previous blogs have covered everything from the basics of AI patentability to whether AI can be designated as an inventor. There are also a number of reports and ongoing reviews into the subject, with most of the attention focused on how the patent system can help AI. However, a report from the UK Intellectual Property Office (UKIPO) has turned that question around, and asked how AI can help the patent system.
Back in April, the UKIPO published a study investigating the feasibility of AI-assisted patent prior art searching. As anyone who has performed a patent search, whether a prior art search, freedom to operate search, or indeed a search for any other reason will confirm, finding relevant patent documents is perhaps not as easy a task as it may initially appear to be. Legalese, different terminology used in different fields of technology, and documents in different languages all conspire against the humble patent searcher. And whilst a number of tools exist to aid patent examiners and other searchers, most notably the extensive patent classification systems (the CPC has over 250,000 categories!), any further help will no doubt be welcomed.
As well as easing the burden of those actually doing the searching, increasing the ability of searchers to find relevant documents will also help both patentees and actors in the marketplace without patents of their own. A more robust search during the patent application process will mean that a patentee can be more assured that the resulting patent is valid and valuable, whilst improved freedom to operate searches can give greater confidence to a manufacturer or importer before they enter a market that they are not infringing any third party rights.
The study into the feasibility of using AI to assist patent searches was carried out by Cardiff University and commissioned by the UKIPO. The objectives of the study were to evaluate the viability of different AI technologies for patent prior art searching, test different approaches to identify the most effective algorithms, and to fully evaluate an optimal solution. Areas of AI technology that were reviewed included natural language processing, supervised and unsupervised machine learning, and semantic knowledge, and a number of key challenges were identified, such as legal wording, long sentences and the technical nature of patent claims.
While overall the study concludes that it is not currently feasible to provide a fully automated patent searching solution, it nevertheless identified areas where the use of AI could aid a human when performing a patent search. As the report puts it, in the proof-of-concept approach developed, “the user keeps the role of the key decision maker, whereas the AI provides intelligent decision support”. Such areas where the use of AI was found to be of assistance included ranking documents returned by a search by relevance and in classifying patent documents. On the other hand, no effective AI technique was found to be effective at processing, or “reading”, a patent and generating a search query. Ultimately, the specialist expertise and experience of a human patent examiner was still required to generate a search term to initiate the search process, even if AI could help interpret the results.
No doubt the patent examiners at the UKIPO will be breathing a sigh of relief – the machines aren’t coming for their jobs (yet). But in the future we can likely expect more and more integration of AI technologies into the patent process. And who knows, maybe one day we’ll have AI patent examiners performing searches for AI patents for inventions invented by AI.
This article is for general information only. Its content is not a statement of the law on any subject and does not constitute advice. Please contact Reddie & Grose LLP for advice before taking any action in reliance on it.