Managing meaning: designing scalable semantic systems for humans & AI

In his keynote at KMWorld Europe & Taxonomy Boot Camp London, Noz Urbina, Founder & Principal Consultant, Urbina Consulting, addressed the issue of people asking for what they want, but not what they need.

In mythology, the story of King Midas asking that everything he touches turn into gold backfired momentously because he didn’t understand the ramifications of what he asked for as opposed to what he actually intended. Most misinterpretations of requests, particularly when AI is involved, do not have such dire consequences. However, a study from Anthropic suggests that desperation causes AI to cheat, blackmail, and cut corners.

Within organizations, Urbina pointed out that every worker knows what is needed to optimally and compliantly do their job for the greatest measurable output. The common denominator is context. We must maintain shared contextual understanding amongst people and machines to gain meaning. Even something that seems factual, such as the boiling point of water, is contextual. Is it at high altitude? Does it differ if the use case is cooking pasta or sterilizing bandages?

Capturing knowledge should be all about getting knowledge out of one head and getting it into another. We can use it to create more knowledge, or get it into a machine where it can be found and reused.

Ontologies act as mindmaps for relationships that structure information. Taxonomies are lists of actual data such as menus, search, recommendations, product comparison tools, data dashboards, or data dictionaries. Knowledge graphs create databases not of tables and roles but of meaningful relationships providing a flexible path through data. Urbina reminded us that Google’s search experience has been powered by graphs for many years.

The trend is toward semantic structured content, which returns an answer rather than a page. Modular semantically tagged and structure building blocks (think Lego) provide component content. The idea is to turn data into knowledge and knowledge into data. Next up is semantic metadata layers that provide the right content and facts to an AI, controlling logic and maximizing context and focus to improve accuracy, explainability, and performance. Knowledge graphs are morphing into context graphs and context engineering is replacing prompt engineering.

In this new AI-powered KM environment, think value not volume. Many of the older metrics measured output but not impact. Silos probably will not disappear but running pipes between them can enhance knowledge sharing. AI technologies benefit from the human in the loop, but Urbino suggested that humans are the loop.

The KMWorld Europe and Taxonomy Boot Camp London held 14-15 April 2026 in London UK, brings KMWorld to London for the first time and returns the in-person Taxonomy Boot Camp to the city.

The KMWorld conference returns to the JW Marriott in Washington D.C. on November 17-20, 2026  https://www.kmworld.com/Conference/2026.

KMWorld 2026 is a part of a unique program of five co-located conferences, which also includes Enterprise Search & Discovery, Enterprise AI World, Taxonomy Boot Camp, and Text Analytics Forum.