Patent prosecution data and the unfolding AI-ML possibilities

There used to be a time when inventors or their attorneys had to take a working model of their invention and handmade drawings to a patent office to convince a patent examiner that his/her invention met the best mode or enablement requirements! In the United States, the practice of submitting working models and drawings remained until 1880. Then came a time when attorneys started using typewriting machines to prepare patent applications and attaching handmade drawings. When typesetting became an integral part of computing and McIntosh (Steve Jobs) introduced the concept of fonts in personal computing machines, and when inkjet and then laser printers arrived; patent lawyers switched to computer printouts to prepare and submit patent applications. Patent drawing preparation became hassle-free with the emergence of CAD. The rest, as they say, is history.

From the early times in patent practice, the type of documents changed from Word Perfect to MS Word, then to PDF, and XML. More recently computer vision and OCR techniques enabled significant reduction in document-size, made transmission easy, reduced storage space, and made conversion of documents from editable to readable and back possible with least impact on a document’s format.

These developments in technologies enabled national IP offices to switch to on-line patent filings and build prosecution history data into online searchable databases which eliminated the need for paper copies of prosecution file-histories. A step ahead, the top 5 global patent offices (the so-called P5: USPTO, EPO, JPO, CNIPO, KIPO) started collaborating on prosecution history data that led to the emergence of Global Dossier (GD). Today GD comprises of millions of patent documents, combining the file histories of patent applications in all the P5 countries. Enroute, USPTO came up with the Patent Application Information Retrieval System (PAIR), with private PAIR for attorney or their delegate’s access and public PAIR for general public’s access. This system has now evolved into Patent Center and a more advanced PEDS (Patent Examination Data System).

This evolution of patent data is widely known in the patent community. An abbreviated restatement of this evolution was needed to place the topic of this note in the right context. That topic is about the possibilities when global patent prosecution data gets curated into AI-ML trained libraries that can power point-solutions for patent prosecution and portfolio development, and even aspects of patent monetization.

Let’s get deeper into this. Global Dossier now has 35-40 million documents. This comprises the file histories of all patent applications pending before P5 patent offices. There are approximately 200 types of documents that get generated when a patent application gets prosecuted at the USPTO, EPO, JPO, and Chinese and Korean patent offices. The file histories include examination reports, response to examination reports, PCT international search reports, PCT examination reports, as applicable, US IDS documents and several other documents that get exchanged between the applicant and the patent office. Naturally, the file histories contain the details of all the references cited by the Examiners in these jurisdictions.

Patent applications are classified for examination and publication purposes. Take CPC (Cooperative Patent Classification) Class H04M relating to telephonic communications, as an example. Curate this data and build a library of file histories of all the patent applications pending in this Class. Train the dataset, using algorithms from a Python software stack. Now phrase questions: What are the top 10 references cited across all patent applications filed/published in the past 10 years in this Class. How many references are forward cited across patent applications for 3G, 4G and 5G? How many patent applications have received first office action allowances? These are just examples of questions that can be framed to generate useful insights from a patent prosecution strategy perspective. Engineer the data to render the answers using algorithms from a MERN software stack. The system will answer this question in a few clicks! More importantly, if more data can be curated and AI-trained and more questions can be framed, over time the system will begin to learn and begin to power the solution to answer questions more accurately.

The key point here is about today’s patent data availability and its AI trainability and what it means to the future of patent prosecution.

Thinking back, it all began when the Statute of Monopolies was promulgated in England in 1623 and the Crown started issuing literae patentes, meaning open letters to let the public know about the useful information disclosed by an inventor who was awarded a monopoly, such that the published information can be put to use to advance innovation. The concept of public notice function thus began and remains central to the legal and institutional framework concerning the obligation of national IP offices to make patent information available for public consumption. Patent Offices across the world have been making progress in ensuring patent information dissemination progresses from supplying paper copies of file histories to searchable interfaces to advanced search systems to now API connectors, bulk downloads, and analytics. The wider the dissemination, the better the public good.

AI/ML/NLP technologies and the availability of large amounts of good quality patent data offer faster and more efficient ways to meet many known needs in the industry and the IP profession. While AI in IP is the beginning of a new era, building real AI solutions takes time and needs significant investments. It is but a matter of time to see real players with deep technologies to emerge in this area.