Top insights from Knowledge Graph Insights Larry Swanson, Content architect, podcast host, community event organiser London 2026 #Designing4Meaning #wiad26

Top 10 Knowledge Graph Insights Take-home lessons from 45 podcast interviews and 6 years of immersion in the semantic technology community Larry Swanson, WIAD London, 10 March 2026

1

  1. At first glance, knowledge graphs and LLMs seem totally at odds with one another.

  1. But if you think about how human’s process information, it becomes clear that you need at least two kinds of thinking.

  1. So in the same way that contrasting fundamental properties contribute to a holistic human mindset and approach…

  1. …hybrid AI architectures get closer to the innate cognitive awesomeness that drives human intelligence.

  1. KGs and LLMs complement each other AI systems benefit from both learning systems and knowledge systems, and from both general and domain-specific information. Machine Learning Knowledge Representation

2

  1. When you first discover a new technology, it’s tempting to use it everywhere…

  1. …especially if you’ve devoted a good part of your life to understanding and applying it.

  1. But your CEO doesn’t care about your cool craft or your carefully developed practice expertise…

  1. …they just want to see business outcomes that support their need to grow the bottom line and keep their board happy.

  1. Always focus on a business problem Ground your work in a clearly understood problem space that addresses a real and measurable business problem.

3

  1. Many engineers and business people regard the “semantic web” as a hippie dream that failed.

  1. In fact, semantic web technologies appear in more than half of the web pages on the internet…

  1. …and virtually every major technology company uses semantic web technology behind the scenes…

  1. …and traditional industries like media, financial services, pharma, energy, and others use knowledge graphs extensively.

  1. The Semantic Web is alive and well The Semantic Web is “the most catastrophically successful thing which people have called a failure.”

4

  1. Enterprise systems, including AI, benefit from semantically meaningful, diligently stewarded data and content…

  1. …but engineers too often work with hastily assembled databases full of vaguely labeled tables and messy data…

  1. …resulting in messy systems that address only narrow concerns and obscure business meaning.

  1. AI needs clean data and meaningful metadata If data is “the new oil,” then you need light sweet crude, not messy sludge, and it needs to be semantically understandable.

5

  1. Ontologists, especially enthusiastic newcomers, are sometimes tempted to overbuild knowledge graph models…

  1. …but semantic technology spans a whole spectrum of ways to add meaning to data.

  1. Know where you are on the semantic spectrum You don’t always need a full-blown, reasoning-capable ontology for your knowledge graph.

6

  1. People have always struggled to align with each other on fundamental questions like what things are…

  1. …and we’re notoriously bad at communicating what we mean to say and how we understand each other.

  1. Even the main word we use to describe these attempts has been demoted to a trite, rhetorical truism.

  1. Getting people to agree on meaning is hard The biggest bottleneck in knowledge work is not technology, but rather the need for clear communication and human agreement.

7

  1. Business activities tend to happen in silos, resulting in data practices that hinder enterprise understanding.

  1. When marketing, sales, accounting, HR, and operations each use a different word to describe a “customer,” meaning is lost…

  1. …and opportunities are squandered as those language differences are captured in enterprise systems…

  1. …when with a more broadly informed and conceptually connected view they could see that they’re talking about the same thing. Semantic Layer

  1. Silos hide connections — KGs reveal them Understanding data independent of its isolated use in any one silo can can unlock hidden business treasures.

8

  1. It’s tempting to think of knowledge graphs as just another technology…

  1. …but the practices that make up ontology design and knowledge engineering trace their origins to well before the computer age.

  1. The ontology principles that capture shared human understanding of a domain come from the metaphysics branch of philosophy.

  1. And the logic that undergirds the powerful reasoning capabilities of knowledge graphs also comes directly from philosophy.

  1. The organisation and labelling schemes that facilitate discovery of knowledge artefacts trace their origins to library science.

  1. And the field of symbolic AI to which knowledge graphs belong traces its roots back to linguistics concepts like the semiotic triangle. Thought or Reference s e z i l o b m y s Symbol re Semiotic Triangle stands for fe rs to Referent

  1. KGs are grounded in the humanities While they are expressed in technical systems, knowledge graphs are built on a foundation of centuries-old human wisdom.

9

  1. Historically, it has been difficult to build ontologies and taxonomies and to populate them with instances and examples.

  1. The arrival of LLMs accelerates once-burdensome tasks like entity extraction, the process of identifying entities in documents…

  1. …and taxonomy construction, the practice of organising terms and concepts into hieararchical classification schemes.

  1. The extracted terms and draft taxonomies still need human oversight, but tasks that once took weeks now take just hours.

  1. LLMs facilitate knowledge graph building Many tedious ontology design and semantic engineering tasks are now assisted and accelerated with LLMs.

10

  1. As the Silicon Valley hype machine puts up billboards touting human-free workplaces… SV “replace people” ad

  1. …humans remain as relevant as ever, providing the leadership that computer systems need.

  1. Put humans in control, not just in the loop As Yaakov Belch said in our conversation, forget about “human in the loop” - let’s put “humans in control.”

Top 10 Knowledge Graph Insights 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. KGs and LLMs complement each other Always focus on a business problem The Semantic Web is alive and well AI needs clean data and meaningful metadata Know where you are on the semantic spectrum Getting people to agree on meaning is hard Silos hide connections — KGs reveal them KGs are grounded in the humanities LLMs facilitate knowledge graph building Put humans in control, not just in the loop

Thank you! /LarrySwanson LarrySwanson.com