Inventing is one of my passions and as an IBM Master Inventor I’ve been fortunate enough to work on several interesting patents. IBM has been the United States patenting leader for the past 25 years. As I near 50 patent issuances I would like to highlight a few of my favorite patents I invented that have been issued.
Corpus quality analysis (US9754207B2)
AI systems live and die based on the quality of their data. Given an AI system using some amount of data we want to analyze a new data source to see if it is worth (in money and/or time) adding that new data to our AI system. The analysis in this patent includes a search for positive and negative features in the new data source, since we want to both verify we are adding high-quality documents and that we are NOT adding low-quality or worse yet “misleading” documents.
Cognitive security for voice phishing activity (US9966078B2)
This patent describes an AI system that protects you from phishing attacks by analyzing conversation patterns and detecting suspicious patterns. The system sits between you and the caller, intercepting suspicious questions and prompting the phisher for information that only a legitimate caller could know. For instance if the caller asks for a password to “check an account”, the AI system asks the caller to provide secret information such as the callee’s account number or mother’s maiden name.
Linguistic based determination of text location origin (US9514125B1)
Linguistic based determination of text creation date (US9436677B1)
This pair of patents helps answer the question “when and where did this text come from?” Language and cultural norms evolve quickly, memes come and go, and with knowledge of these patterns we can quickly suggest when text may have been written. Having good metadata around text is useful in an AI system and methods like these patents to recover that text metadata are valuable.
Cognitive digital security assistant utilizing security statements to control personal data access (US9600687B2)
Every day we interact with digital systems that ask us for more and more data. Laws such as GDPR have been introduced to make explicit why these systems are asking us for our data and what they will do with it, however it is still cumbersome to go through all of these requests and decide which ones to allow and which to deny. This patent describes an AI system that parses data access request and the accompanying benefit statements, compares them to a set of personal preferences, and automatically approves the ones matching your preferences and denies the ones that don’t.
Dynamically gain and produce insight from a table using NLP (US9286290B2)
When a table structure is known and well-formed, machines can quickly extract all of the information in the table, but not understand which parts of the table are most significant. Contrast this to when humans read a table, they scan the table and quickly extract insights such as table cells with maximal/minimal values or those with very high (or low) variance. This patent helps a machine extract just these most valuable insight statements for narration and storage, thus maximizing the use of storage and index space for table data, which is especially helpful on very large tables.