Context Graph and Process Knowledge, Jessica Talisman, Contextually LLC

Jessica argues that the industry’s current obsession with “context graphs” and “decision traces” is a misdirection if treated as a purely technical or data engineering problem. While decision traces—logs of how an AI arrived at a conclusion—are valuable, they are essentially meaningless in isolation. The speaker asserts that the true challenge is a knowledge management problem, requiring a formal infrastructure to define the relationships, roles, and policies that give a decision its context. Without a pre-defined knowledge model (or ontology), capturing raw data is just “collecting streams” and hoping for insights that will never materialize.

To build resilient AI systems, the speaker advocates for a shift toward procedural knowledge, a discipline that transforms tacit human expertise into formal computational representations. This process involves four critical steps: acknowledging the human-centric nature of knowledge, investing in “elicitation infrastructure” to extract expertise from people’s heads, building a formal knowledge model before the data persistence layer, and designing systems where knowledge capture is a primary objective. By standing on the “shoulders of giants” in existing procedural research, enterprises can move beyond simple operational logs to create systems that truly understand and manage the complex “negative space” between data points.

Would you like me to break down the “Procedural Knowledge Ontology” (PKO) layers mentioned in the talk, such as Strategic, Tactical, and Operational?