Making AI Smarter by Eliminating Flawed Data Jim Barnebee, GetSwift’s VP for AI and Infrastructure, was granted a US Patent for his pioneering work with knowledge graphs.
Artificial intelligence algorithms require gobs of data. With each introduction of new and varied data sets, machine learning applications mature. The better and more comprehensive the data, the better the outcome.
But what happens when the training set that informs AI is corrupted with flawed data?
Not great things, as it turns out. That’s why Jim Barnebee, GetSwift’s Vice President of Artificial Intelligence, and his former colleagues from IBM’s Watson Group invented a system to allow machine-learning applications to interpret the quality of new data through a process called veracity. The US Patent and Trademark Office granted Jim and his three colleagues from IBM a patent this past September.
The patent solved several of the most persistent problems in the world of knowledge graphs, massive data and Ai, like the need to continually retrain machine-learning systems to distinguish between unique stores of data.
Not all data is equal because of its location or time of creation, and for systems that consume data like rocket fuel, that’s problematic. That’s why Jim and the team created the concept of veracity to allow a machine learning system to vet the truthfulness of datasets and, in turn, determine when knowledge graphs should use those datasets to get updated with more reliable data.
For businesses, the applications of veracity are enormously important.
Consider: if your company sells shoes around the globe, you have massive amounts of data about those sales and the supply chain that feeds them in every location.
Aside from the language and cultural barriers, you have strong and weak markets and varying levels of data quality. Yet you desperately want your data to tell you which models of shoes are flying off the shelves, where are data anomalies, and where there’s possible theft or fraud. AI can be trained to search for these and other anomalies once the data is made available and truth versus non-truth can be defined and replicated.
And to make this pattern completely auditable, the patent also includes the use of blockchain to create a transparent and accountable record. Blockchain is a better audit path than human memory.
This is much like the way humans learn, absorbing information as we grow and experience the world, replacing earlier constructs with better ones that form our understanding going forward. This process—that is, constantly improving knowledge graphs—has been eluding AI researchers for quite a long time.
You can view the full patent here.
Jim’s training in both ontology (which, in computer science, is essentially the representation of the knowledge from a set of concepts) and blockchain helped the team technically solve the twin problems of veracity and memory. A former DARPA developer with groundbreaking work on Java as a language and encryption tool for the US Navy, Jim worked on ontologies and knowledge graphs for years before joining IBM Watson. Previously he served as the ontology evangelist at Unisys, led custom ontology work at Orbis, and also founded and moderates one of LinkedIn’s largest ontology special interest groups.
Jim continues to use his expertise in AI as a VP at GetSwift, alongside a uniquely talented (and growing) team assembled by CTO Dennis Noto. Jim and the team are using AI to transform cloud-based, real-time delivery management software into the next generation of computing. To stay up to date with the latest developments at GetSwift and apply for future opportunities, follow GetSwift on LinkedIn.