“In the context of LLMs and AI, ontology refers to the formal, structured representation of knowledge within a specific domain, defining entities, their properties, and relationships.” – Ontology
In the context of large language models (LLMs) and artificial intelligence (AI), an ontology serves as a formal, structured representation of knowledge within a specific domain, explicitly defining entities, their properties, and the relationships between them. This creates a shared vocabulary and logical framework that enables both humans and machines to communicate effectively, reason about data, and draw inferences beyond explicit programming.1,2,3
Core Components and Functionality
An ontology typically comprises three key elements: classes (or concepts, such as ‘person’ or ‘organisation’), attributes (properties like ‘name’ or ‘role’), and relationships (connections, e.g., ‘works for’ or ‘co-presents with’). Unlike a simple taxonomy, which organises items hierarchically, an ontology captures complex interconnections, allowing AI systems to infer new knowledge-for instance, deducing that two co-presenters at a conference are both speakers.2,4
In LLMs and AI applications, ontologies underpin knowledge bases, acting as a ‘single source of truth’ for semantic understanding. They facilitate knowledge sharing, enhance machine readability, and support advanced features like personalised recommendations or conversational AI by contextualising data through defined rules and relations.1,5
Applications in AI and LLMs
- Semantic Web and Knowledge Graphs: Ontologies power graph-based systems, such as those used by Palantir, enabling the mapping of entities and relationships for intelligence analysis and decision-making.3[tags]
- Enterprise AI: They provide structured memory for LLMs, ensuring business-aligned reasoning, explainability, and scalability across teams and tools.5
- Ontology Engineering: Involves designing ontologies that remain current, comprehensive, and adaptable, often using languages like OWL (Web Ontology Language) built on RDF standards.3
Distinctions and Common Misconceptions
Ontologies differ from glossaries (mere term lists) or taxonomies (hierarchical categorisations) by incorporating relational logic for reasoning. They evolve with domains, addressing challenges like maintaining specificity and supporting use cases in dynamic environments.3,4
Key Theorist: Tom Gruber
The most influential strategist and theorist associated with ontologies in AI is Tom Gruber, whose seminal definition has shaped the field. Gruber, an American computer scientist and entrepreneur born in 1959, coined the widely adopted definition: ‘An ontology is a formal, explicit specification of a shared conceptualisation.’ This emphasises ontologies as agreements on domain representations, bridging human intuition and machine processing.3,7
Gruber’s backstory intertwines philosophy, AI research, and enterprise innovation. Holding a PhD in Computer Science from Stanford University (1989), he pioneered work in knowledge acquisition and sharing during the 1990s AI ‘knowledge representation’ era. At Stanford, he contributed to ontology engineering tools and co-developed early frameworks for collaborative knowledge systems. His philosophical roots-drawing from ontology’s classical study of being-influenced his pivot to computational semantics, arguing that ontologies enable ‘shared understanding’ among agents.7
Professionally, Gruber founded?? companies, including Siri Inc. (acquired by Apple in 2010), where he served as Chief Technology Officer. There, he applied ontologies to natural language understanding, structuring voice queries into entity-relationship models-directly precursor to modern LLM knowledge integration. Post-Siri, he consulted on AI ethics and semantic technologies, authoring over 200 publications. His work underscores ontologies’ role in scalable AI, influencing tools like Protégé at Stanford and OWL standards.3,7
Gruber’s legacy positions ontology as indispensable for agentic AI systems, where structured knowledge graphs (as in Palantir’s platforms) enable reasoning over vast, interconnected data.[tags]
References
1. https://www.jorie.ai/post/what-is-an-ontology
2. https://www.earley.com/insights/role-ontology-and-information-architecture-ai
3. https://en.wikipedia.org/wiki/Ontology_(information_science)
4. https://www.decidr.ai/blog/what-is-ontology-and-how-it-powers-intelligence
6. https://www.geeksforgeeks.org/machine-learning/introduction-to-ontologies/
7. https://protege.stanford.edu/publications/ontology_development/ontology101-noy-mcguinness.html
8. https://www.youtube.com/watch?v=UW57RW-4kWs
