Adopting AI requires a thorough, organized, prepared approach for it to best deliver on its plethora of lofty promises. Through the power of knowledge graphs, enterprises can fuel AI strategies with accurate, trustworthy, and scalable information that grounds and contextualizes AI models, driving successful outcomes.
Jim Morris, solution engineer, Progress, and Stephen Reed, senior account manager, Progress, joined DBTA’s webinar, Building Knowledge Graphs to Power Your AI Initiatives, to examine how knowledge graphs can effectively prop up AI initiatives to be smarter, more accurate, and less prone to hallucinations and bias.
“What’s the real difference between an AI initiative at company A versus company B? Because if you think about the tools that are out there—the ChatGPTs, the Copilots, the Groks, all those—they’re kind of accessible to anyone. It’s kind of like a level playing field from that perspective,” examined Reed. The difference, he continued, is how enterprises integrate their private data with their AI solutions.
With that being said, why is it so hard to implement AI projects with private data? Akin to interviewing and onboarding a new employee, generative AI (GenAI) does not immediately know what data your enterprise maintains. Reed highlighted the following as top barriers:
- Inability to make sense of structured and unstructured data
- Different data repositories using different nomenclatures
- Critical/sensitive data being difficult to identify or locate
- Too many disparate data systems and data siloes
- Off-the-shelf GenAI is not trained on any private data
Many of the top concerns relating to AI and large language model (LLM) implementation—such as hallucinations, data biases, missing information, and a lack of explainability—can be mitigated or eliminated by the proper usage of private data, said Reed.
One way to successfully integrate private data and AI is through knowledge graphs, a method of connecting isolated and independent data with semantic associations, eliminating siloes. These relationships are described by a taxonomy or knowledge model, which is “a way to group information so that it’s in a logical category. It can be easily referenced, searched, and potentially even distributed.”
Reed offered this metaphor: When entering a grocery store, you enter a taxonomy. If a grocery store grouped its items alphabetically, it wouldn’t make much sense to the shopper. Instead, it groups items based on categories—produce, meat, and dry goods—because that is a logical way of organizing information.
Morris then walked webinar viewers through a live demonstration of a knowledge graph and how to build one, asserting how knowledge graphs “are essential to making AI trustworthy and how knowledge models or knowledge modeling is essential to making knowledge graphs trustworthy.”
These demonstrations, illustrated through Progress’ platform, demonstrated:
- Why RAG solutions without knowledge graphs fail
- The importance of subject matter experts and knowledge workers
- Semaphore model building and enrichment tools
- The role of databases, spreadsheets, and websites
- How to put it all together
This is only a snippet of the full Building Knowledge Graphs to Power Your AI Initiatives webinar. For the full webinar, featuring Morris’ expansive demonstrations, a Q&A, and more, you can view an archived version of the webinar here.