The AI impact on patent searches and analyses: my views after Search Matters 2019
Above: EPO headquarters near River Isar (Image copyright: user "Bbb" on Wikmedia Commons under CC-BY-SA-1.0).
Besides very interesting news and tutorials, the EPO Search Matters event 2019 highlighted new opportunities and challenges in patent analyses and searches. More specifically, it highlighted the impact of AI, which was the topic discussed in the first plenary presentation and in the round table at the end of the first days.
Since this topic attracts fears and expectations as well, I summarize here my views about this debated topic, in terms of why, how, when.
Why AI is required in patent informatics. There is an ever and ever increasing gap today between the increasing number of patent and non-patent documents to analyze and the increasing demand of recall, to avoid missing critical documents. This gap cannot be reasonably filled by the overhead of search professionals, but has to be filled by the use of intelligent programs (hence artificial intelligence) in supporting patent search professionals.
How AI has to be used in Patent Informatics. AI includes a lot of technologies and approaches, from those more automatic, to those more under the direct control of the user. In Patent Informatics, the most suitable approaches will be those under the users control, having the possibility to import best practices from patent professionals.
When AI will have an impact on patent informatics. I think that this will be very soon, but we have not to overlook that some technical and business challenges have still to be faced. These challenges are summarized here below, structured by layers:
| Layer | Challenge |
|---|---|
| Application layer | The system should be able to include or import the best practices developed by patent professionals (controllability) and to justify the reasons of results provided (observability). |
| Technology layer | Suitable technologies, also specific for patents, have to be developed and validated for extracting info from patent metadata, texts and drawings (technologies must be there). |
| Data layer | Patent and non-patent data used have to be available in a suitable form (as linked data) and for a fair price (data must be there). |
An extended paper will detail key points summarized her and will be available soon through IntelliSemantic newsletter, whose registration form is available here.
published: by Alberto Ciaramella @intellisemantic
Copyright © 2006-2023 by IntelliSemantic S.r.l.
Società registrata al registro delle Imprese di Torino con Numero REA TO - 1026789
Sede legale: Corso Susa 299/A -10098, Rivoli (To) - Italy
Capitale sociale 30.000 Euro
Partita IVA 09124770018