*This article is composed of selected excerpts from actual assets used by HYDRAFT® in the course of client consultations.
Recently, in the AI field, Ontology AI has been attracting more attention due to the limitations of RAG (Retrieval-Augmented Generation). Since most of my clients are actively utilizing RAG at their points of business interaction, I have prepared this brief overview to understand the market and technological changes in advance. Although the domestic market has not yet reached a level demanding Ontology AI, companies that aim to lead the market need to proactively understand the concept of Ontology, prepare a structurally sound data infrastructure, and plan long-term roadmaps.
[1. Limitations of RAG]
RAG, which has been in the spotlight in recent years, generates answers by simply retrieving documents. As such, it has limitations when document quality is low, data is insufficient, or structures are incomplete, making it difficult to provide accurate responses. Furthermore, RAG references each document fragment independently, which means it cannot fully understand semantic relationships or structural contexts between different pieces of information. This results in limitations for complex logical reasoning and generating consistent answers.
• Operates based on retrieval and reference rather than pretraining
• Works in sequence: Retrieval → Augmented Generation
• Answer accuracy decreases if document quality, volume, or structure is insufficient
[2. What is Ontology AI?]
Unlike RAG, Ontology AI does not stop at document search or pattern recognition. Instead, it understands and connects semantic relationships and structural contexts between data. By structuring relationships between data in the form of a knowledge graph, it can create a logical and hierarchical knowledge system. Ontology AI enables complex conceptual reasoning, and even with incomplete information, it can interpret context and derive meaningful conclusions. This makes Ontology AI increasingly important in fields such as education and industry, where results based on complex data are required.
• Knowledge-based AI that understands and connects ‘semantic relationships’ and ‘structural context’ between data
• Structures concepts and relationships as a knowledge graph for logical reasoning and consistent interpretation
• Generates meaningful outcomes by understanding an organization’s knowledge structure holistically
[3. Metaphysics and Ontology]
To better understand Ontology AI, it is useful to briefly review the core concepts of an Ontology system: ‘Entity’ and ‘Relationship.’ These concepts are not new; they have been studied for a long time. The first systematic study of these concepts was conducted by Aristotle in his Metaphysics, which I often reference in practice when explaining the essence of a brand.
Ontology, a subfield of Metaphysics, specifically studies entities and their relationships. Therefore, in IT, ‘Ontology’ can be seen as a structural implementation of Aristotelian ontological thought at the data infrastructure level: defining what exists in a given data environment and systematically structuring the relationships among those entities.
[4. Ontology in Philosophy]
Ontology is the study of beings (Ousia; Substance) as a way of being (Being). In other words:
Metaphysics:
• The highest-level philosophy exploring the principles and essence of existence.
• What does it mean to exist?
• Subfields include ontology, causality, time, and space.
Ontology:
• Studies how entities exist (Being).
• Classification, attributes, relationships, categories
Substance Theory:
• Focuses on the most fundamental entities (Substance)
• What remains unchanged amidst change?
• How is essence distinguished from attributes?
Although ontological thinking existed before Aristotle, systematic ontology as we understand it today was established in his Metaphysics. Note that an entity (Ousia; Substance) is the combination of matter and form.
[5. Characteristics of an Ontology System]
Traditional data systems process discrete data points (e.g., customer → order → product) and cannot easily understand the contextual meaning between steps. In contrast, an Ontology system defines the meaning of each data point at the ‘Entity’ and ‘Relationship’ level and connects them, effectively creating a ‘map of data relationships.’ This allows AI not just to retrieve data but to reason coherently and generate new meaning optimized for the domain.
[6. Example of an Ontology System]
To illustrate Ontology features, consider a Hospital Management System (HMS). In an Ontology system, a patient reporting a headache is represented not as text but as a structured semantic relation: <has symptom → headache>. Based on these logical rules of entities and relationships, AI can semantically infer outcomes such as:
Entities:
• Patient: a person with symptoms
• Symptom: headache, fever, dizziness
• Doctor: person providing diagnosis and prescriptions
• Medication: drugs prescribed
Events:
• Diagnosis: interaction between patient and doctor
• Prescription: resulting action from the diagnosis
Relationships:
• Patient has symptom (has symptom)
• Doctor performs diagnosis (performs)
• Diagnosis records symptoms (records)
• Diagnosis produces prescription (produces)
• Prescription includes medication (includes)
User Scenario:
• Patient A may have a prior diagnosis for headache
• The diagnosis likely included headache-related medication
• Medications may cause specific side effects
• Patient A reports dizzines
• The dizziness may be a side effect
• Doctor reviews medication side effects and considers changes
• Automated alerts and follow-up guidance are sent to the patient
[7. Structural Significance of Ontology]
Structurally defining and connecting ‘Entities’ and ‘Relationships’ in an Ontology system reflects real-world structures at the data level. This approach closely aligns with use-case–driven system design in Constructivism or MSA (Microservice Architecture). Whereas RAG answers “What is being asked?”, Ontology explores “Why is it being asked?”—moving beyond simple retrieval to understanding the fundamental structure of knowledge and deriving meaningful conclusions. If humans aim for AI to achieve not just search and summarization but contextual understanding and creation of new meaning, Ontology will inevitably supersede RAG.
[8. Palantir: Practical Implementation of Ontology]
Currently, the company implementing Ontology systems most precisely is Palantir, led by Peter Thiel and Alex Karp. Palantir applies Aristotelian metaphysical, Ontology-based thinking to business operations, naming it the Palantir Ontology. This system goes beyond simple data retrieval, understanding an organization’s knowledge structure and enabling full-scale simulations—effectively an <End-to-End Digital Twin>. By integrating data and processes holistically, Palantir achieves high-level reasoning and simulations unattainable with conventional RAG. In a future discussion, I will share insights into Palantir’s backstory, why it is highly valued in the investment market, and its unique business model. Thank you.
© 2025 Hydraft®
All rights to this content are reserved by Hydraft® and are protected under copyright law. Unauthorized reproduction and redistribution are strictly prohibited.
