Episode #530: The Hidden Architecture: Why Your Startup Needs an Ontology (Before It's Too Late)

In this episode of the Crazy Wisdom Podcast, host Stewart Alsop sits down with Larry Swanson, a knowledge architect, community builder, and host of the Knowledge Graph Insights podcast. They explore the relationship between knowledge graphs and ontologies, why these technologies matter in the age of AI, and how symbolic AI complements the current wave of large language models. The conversation traces the history of neuro-symbolic AI from its origins at Dartmouth in 1956 through the semantic web vision of Tim Berners-Lee, examining why knowledge architecture remains underappreciated despite being deployed at major enterprises like Netflix, Amazon, and LinkedIn. Swanson explains how RDF (Resource Description Framework) enables both machines and humans to work with structured knowledge in ways that relational databases can't, while Alsop shares his journey from knowledge management director to understanding the practical necessity of ontologies for business operations. They discuss the philosophical roots of the field, the separation between knowledge management practitioners and knowledge engineers, and why startups often overlook these approaches until scale demands them. You can find Larry's podcast at KGI.fm or search for Knowledge Graph Insights on Spotify and YouTube.

Timestamps

00:00 Introduction to Knowledge Graphs and Ontologies
01:09 The Importance of Ontologies in AI
04:14 Philosophy's Role in Knowledge Management
10:20 Debating the Relevance of RDF
15:41 The Distinction Between Knowledge Management and Knowledge Engineering
21:07 The Human Element in AI and Knowledge Architecture
25:07 Startups vs. Enterprises: The Knowledge Gap
29:57 Deterministic vs. Probabilistic AI
32:18 The Marketing of AI: A Historical Perspective
33:57 The Role of Knowledge Architecture in AI
39:00 Understanding RDF and Its Importance
44:47 The Intersection of AI and Human Intelligence
50:50 Future Visions: AI, Ontologies, and Human Behavior

Key Insights

1. Knowledge Graphs Combine Structure and Instances Through Ontological Design. A knowledge graph is built using an ontology that describes a specific domain you want to understand or work with. It includes both an ontological description of the terrain—defining what things exist and how they relate to one another—and instances of those things mapped to real-world data. This combination of abstract structure and concrete examples is what makes knowledge graphs powerful for discovery, question-answering, and enabling agentic AI systems. Not everyone agrees on the precise definition, but this understanding represents the practical approach most knowledge architects use when building these systems.
2. Ontology Engineering Has Deep Philosophical Roots That Inform Modern Practice. The field draws heavily from classical philosophy, particularly ontology (the nature of what you know), epistemology (how you know what you know), and logic. These thousands-year-old philosophical frameworks provide the rigorous foundation for modern knowledge representation. Living in Heidelberg surrounded by philosophers, Swanson has discovered how much of knowledge graph work connects upstream to these philosophical roots. This philosophical grounding becomes especially important during times when institutional structures are collapsing, as we need to create new epistemological frameworks for civilization—knowledge management and ontology become critical tools for restructuring how we understand and organize information.
3. The Semantic Web Vision Aimed to Transform the Internet Into a Distributed Database. Twenty-five years ago, Tim Berners-Lee, Jim Hendler, and Ora Lassila published a landmark article in Scientific American proposing the semantic web. While Berners-Lee had already connected documents across the web through HTML and HTTP, the semantic web aimed to connect all the data—essentially turning the internet into a giant database. This vision led to the development of RDF (Resource Description Framework), which emerged from DARPA research and provides the technical foundation for building knowledge graphs and ontologies. The origin story involved solving simple but important problems, like disambiguating whether "Cook" referred to a verb, noun, or a person's name at an academic conference.
4. Symbolic AI and Neural Networks Represent Complementary Approaches Like Fast and Slow Thinking. Drawing on Kahneman's "thinking fast and slow" framework, LLMs represent the "fast brain"—learning monsters that can process enormous amounts of information and recognize patterns through natural language interfaces. Symbolic AI and knowledge graphs represent the "slow brain"—capturing actual knowledge and facts that can counter hallucinations and provide deterministic, explainable reasoning. This complementarity is driving the re-emergence of neuro-symbolic AI, which combines both approaches. The fundamental distinction is that symbolic AI systems are deterministic and can be fully explained, while LLMs are probabilistic and stochastic, making them unsuitable for applications requiring absolute reliability, such as industrial robotics or pharmaceutical research.
5. Knowledge Architecture Remains Underappreciated Despite Powering Major Enterprises. While machine learning engineers currently receive most of the attention and budget, knowledge graphs actually power systems at Netflix (the economic graph), Amazon (the product graph), LinkedIn, Meta, and most major enterprises. The technology has been described as "the most astoundingly successful failure in the history of technology"—the semantic web vision seemed to fail, yet more than half of web pages now contain RDF-formatted semantic markup through schema.org, and every major enterprise uses knowledge graph technology in the background. Knowledge architects remain underappreciated partly because the work is cognitively difficult, requires talking to people (which engineers often avoid), and most advanced practitioners have PhDs in computer science, logic, or philosophy.
6. RDF's Simple Subject-Predicate-Object Structure Enables Meaning and Data Linking. Unlike relational databases that store data in tables with rows and columns, RDF uses the simplest linguistic structure: subject-predicate-object (like "Larry knows Stuart"). Each element has a unique URI identifier, which permits precise meaning and enables linked data across systems. This graph structure makes it much easier to connect data after the fact compared to navigating tabular structures in relational databases. On top of RDF sits an entire stack of technologies including schema languages, query languages, ontological languages, and constraints languages—everything needed to turn data into actionable knowledge. The goal is inferring or articulating knowledge from RDF-structured data.
7. The Future Requires Decoupled Modular Architectures Combining Multiple AI Approaches. The vision for the future involves separation of concerns through microservices-like architectures where different systems handle what they do best. LLMs excel at discovering possibilities and generating lists, while knowledge graphs excel at articulating human-vetted, deterministic versions of that information that systems can reliably use. Every one of Swanson's 300 podcast interviews over ten years ultimately concludes that regardless of technology, success comes down to human beings, their behavior, and the cultural changes needed to implement systems. The assumption that we can simply eliminate people from processes misses that huma...