Series Overview
Part 1 | Why Does AI Get Things Wrong So Convincingly?
Part 2 | Why the OXMV Loop and Evidence Ledger Are Necessary
Part 3 | Mapping Knowledge for AI with Ontology
By Manripungchwi Jung Yeon-tae
More Knowledge Does Not Necessarily Mean Better Understanding
Imagine a library containing millions of books but no classification system, shelf numbers, or catalogue. The library holds an enormous amount of knowledge, yet finding the right information and applying it to a real-world problem would be extremely difficult. Generative AI faces a similar challenge. It can answer questions about law, construction, finance, and medicine, but possessing large amounts of information is not the same as understanding reality accurately.
Experts do not make judgments from isolated words. They examine the relationships among people, organizations, documents, actions, time periods, conditions, evidence, and responsibility. Reality is not simply a collection of words. It is a structure of relationships.
To handle professional knowledge responsibly, AI needs more than a warehouse of information. It needs a map showing where knowledge belongs and how its elements are connected. That map is ontology.
Entities, Attributes, Relationships, and Rules
Ontology is a structured blueprint describing what exists within a field, what characteristics each element possesses, how the elements are connected, and under what conditions a conclusion may be reached. In construction, clients, general contractors, subcontractors, designers, and supervisors are entities. Contract dates, project values, construction periods, and document numbers are attributes. Contracts between parties, relationships between design changes and schedule extensions, and connections between increased costs and supporting records are relationships. Requirements that must be satisfied before additional costs can be claimed are rules.
The same word may carry different meanings across professional fields. “Responsibility” may refer to civil, criminal, or administrative liability. On a construction site, it may mean responsibility for design, execution, supervision, or safety management. Likewise, ordinary income, accounting revenue, and taxable income are not identical concepts.
Ontology distinguishes concepts that appear similar and connects information that appears separate. Professional expertise is not merely knowing many terms. It is the ability to recognize important differences and understand meaningful relationships.
Connecting Law, Construction, Finance, and Medicine
Loading an AI system with statutes and court decisions does not automatically turn it into a legal expert. To analyze a contract dispute, the system must connect contractual duties, alleged breaches, responsibility, damages, causation, notice, limitation periods, and evidence. The existence of a related precedent does not determine the outcome. The facts, legal issues, applicable rules, and relevant period must also be comparable.
Construction disputes likewise cannot be resolved through a single document. Contracts, design documents, change orders, schedules, meeting minutes, work logs, and cost records must be connected through time and causation. Suppose a design change is followed by a 60-day extension. The sequence alone does not prove that the design change caused the entire delay. Material shortages, weather, labor problems, or the contractor’s own performance may also have contributed.
In finance, the movement of money between accounts does not reveal the complete meaning of a transaction. The relationship between sender and recipient, source and purpose of the funds, frequency, fragmented transfers, and final destination must be considered together. AI may identify suspicious patterns, but it should not label a customer a criminal based on limited data.
In medicine, chest pain alone does not establish heart disease. The location and nature of the pain, its onset, the patient’s age, underlying conditions, accompanying symptoms, and test results must be connected. AI’s role is not to confirm a diagnosis but to identify urgent warning signs and guide the user toward appropriate medical care.
Ontology, Knowledge Graphs, and the Evidence Ledger
If ontology is the blueprint of knowledge, a knowledge graph is a map that places real people, events, documents, and relationships onto that blueprint. Within a knowledge graph, contracts, official letters, and meeting minutes are no longer isolated files. The system can examine who had authority to issue an instruction, which document supports a particular claim, and whether dates and amounts are consistent.
The Evidence Ledger adds another layer. It records which evidence supports or contradicts each claim and identifies what remains unverified. Consider the claim that a design change caused a 60-day delay. A design-change instruction alone is not sufficient. A revised schedule, extension approval, site records, and evidence of the actual cause of delay may also be required.
Proving that a design change occurred is different from proving that it caused all 60 days of delay. Ontology defines relationships. The Evidence Ledger records the evidence supporting or challenging those relationships. The OXMV Loop repeatedly checks whether claims, facts, rules, evidence, and conclusions remain consistent. Together, they transform a fluent AI answer into a traceable professional-knowledge structure.
Not Every Connection Is Causation
One of the greatest risks in knowledge systems is confusing relationships with causation. If a company adopts AI and its sales subsequently increase, AI cannot automatically be identified as the sole cause. Market growth, a major contract, or the withdrawal of a competitor may also explain the change. If symptoms improve after medication, natural recovery or other treatment may have contributed.
A good ontology does more than connect information. It distinguishes contractual, temporal, evidentiary, and causal relationships. An unproven causal connection must remain a hypothesis requiring verification rather than being presented as fact.
Ontology does not guarantee truth by itself. If concepts are defined incorrectly or outdated laws and standards are applied, the structure may produce an incorrect conclusion. When new evidence or rules emerge, the ontology and all connected judgments must be updated. Ontology is not a finished temple of knowledge. It is a living map that must change as reality changes.
AI Analyzes; Humans Decide
AI can detect suspicious financial transactions, classify medical risks, and find gaps between contractual claims and evidence. It should not, however, declare someone a criminal, diagnose a disease from limited information, or determine personal intent and responsibility from circumstantial evidence alone. Classification is a tool for understanding and review. It must not become a verdict that brands a human being.
Law, construction, finance, and medicine use different concepts and rules, but they share fundamental questions: What is known? Who acted? What evidence supports the claim? Which rules and exceptions apply? What should AI analyze, and what must humans decide?
Ontology connects different areas of expertise through the common language of claims, evidence, relationships, and rules. At the same time, it reveals where professional fields meet and where they must remain distinct. Giving AI a map of knowledge also requires us to reconsider our own systems of judgment.
What do we call a fact? Are we confusing possibility with causation? Which people or conditions have been omitted from our analysis?
A map is not the destination, but we cannot travel safely over a long distance without one. Ontology can give AI a map. Human beings must still give that map direction and meaning—and remain responsible for the final decision.
The next installment, “Can Humans Understand AI’s Judgments? — Why XAI Is Necessary,” will examine how people can understand and verify the reasons behind AI-generated conclusions.
About the Author
Manripungchwi Jeong Yeon-tae is a professional-knowledge AI planner, designer, and GPT development specialist. He currently serves as President of the Construction Industry Information Institute and develops professional GPTs across construction, law, taxation, finance, management, medicine, science, history, and philosophy. His research focuses on verified professional-knowledge AI combining ontology, XAI, the OXMV Loop, and the Evidence Ledger.
