Many businesses expect AI adoption to improve efficiency, expand operations and deliver other benefits. Yet AI often fails to meet those expectations. In risk management or decision-making, incorrect advice from AI can even put a business at risk.
Why does this happen? A common cause is hallucination, a limitation of AI. When AI uses contaminated or unnecessary data, or interprets, reasons about or applies data incorrectly, it may produce conclusions far removed from reality. The industry has therefore focused on technologies that reduce hallucinations and produce accurate conclusions.
Youngkuk Hamn, CEO of startup Worflogy, found a possible answer in ontology, a longstanding branch of philosophy. His approach also makes active use of semantic data, which is essential to building ontology-based AI.
Ontology and semantic data help AI understand the meaning of words and sentences and reason in a more human-like way. An ontology is a knowledge map or encyclopedia that organizes real-world concepts and relationships in a form AI can understand. Semantic data, created using an ontology, assigns meaning and relationships to data to support reading, understanding and reasoning.
AI equipped with these technologies can understand and reason about real-world concepts and relationships, reducing the likelihood of hallucination. Consider the well-known fabricated account of King Sejong throwing a MacBook Pro. Semantic data associates King Sejong with a 15th-century Joseon king and the MacBook Pro with a computer introduced in the 21st century. The incompatible time periods would allow AI to identify the statement as erroneous.
King Sejong
[Concept: 15th-century Joseon king]
➔
Mismatch detected (error)
[15th century vs. 21st century]
➔
MacBook Pro
[Concept: 21st-century computer]
Causal verification in the semantic layer: Clearly defining conceptual attributes and domain rules allows AI to check temporal and spatial causal relationships and prevent hallucinations.
Ontology and semantic data can also help trace reasoning backward to improve the overall process. Even without new data, existing concepts and relationships can support further reasoning. By focusing on accurate data, the approach can also reduce computing costs and resource use.
These advantages have already given ontology and semantic data a significant role in the global AI market. Big data company Palantir is one example: it applies ontology after collecting data to guide AI toward suitable decisions under a clearly defined objective.
Worflogy focuses on this same point: making ontology and semantic data work toward the user's explicit goals. Hamn says using all available data for AI computation is practically impossible and inefficient. He envisioned an ontology tool that first derives context and a clear objective from data, then uses only data relevant to them.
Large volumes of raw data
Enterprise-wide unstructured data
➔
Context and goal filter
Ontology schema mapping
➔
Data optimized for computation
Efficient decision reasoning
Goal alignment: Filter semantic knowledge nodes that match business objectives to reduce costs and significantly improve reasoning accuracy, rather than processing all data indiscriminately.
Hamn has an unusual background. After studying Japanese regional studies and working in web development, he moved to the United Kingdom to study corporate strategy, risk management and security theory. He also researched semantic data, design-based risk simulation, corporate decision-making cases and methods, project risk, and semantic data design and decision theory informed by standardized execution processes.
After returning to South Korea, he worked in commercialization planning, feasibility analysis and risk management at an aerospace company. He observed businesses neglecting risk management and failing to collect and manage internally generated data, undermining decisions and the growth of both companies and employees. He developed an ontology tool to address these problems using the ontology and semantic data technologies he had studied.
Worflogy first introduced an ontology tool suited to businesses and research institutes. It connects employees' competencies, careers and work-generated data in a knowledge graph. Using this structure, it analyzes matters such as recruitment, revenue growth, overseas expansion and process improvement, defines starting points and destinations, and proposes the steps to get there. The aim is to help people use their capabilities and improve work efficiency.
Member profiles and career history
Work output nodes
➔
Knowledge graph connections
Competency-to-project mapping
➔
Suitable role recommendations and scenarios
Automatically suggested growth paths
Knowledge graph-based talent matching: Combine actual work outputs with ontology schemas, going beyond isolated career descriptions to derive roles aligned with business objectives.
The tool also supports corporate decision-making and risk management. Sound decisions require clear data and causal analysis rather than experience or speculation alone. Ontology defines employee behavior and performance, organizational data and market data in detail, then organizes and interprets their relationships under the business's objectives. The same principle can model potential crises and identify logical gaps in the business structure.
Worflogy's ontology tool can also serve research institutes. It structures researchers' backgrounds, research processes and results, data and ideas as nodes and links. It then preserves, organizes and interprets them under the institute's explicit objectives.
Many AI tools analyze company and employee information to improve efficiency. Hamn emphasizes that Worflogy's operating principles and structure differ. General-purpose AI architectures interpret data, while Worflogy integrates operational data into an ontology to create a knowledge graph, then applies domain-specific knowledge to use that graph. Its customized integration of real-world data is presented as its principal strength and technological moat.
1.
Identify issues and generate ontology queries: Search relevant knowledge graphs when a business issue arises
2.
RAG-based context retrieval: Combine external regulations, internal work assets and ontology rules to establish context
3.
Dialectical debate among agents: AI agents examine feasibility and logical gaps from different perspectives
4.
Final decision proposal: Report verifiable scenarios designed to minimize risk to human decision-makers
Dialectical knowledge verification workflow: Beyond summarizing information, the workflow is designed for AI to formulate and debate hypotheses using rules defined in the ontology schema, aiming for decisions with very few flaws.
Worflogy offers its ontology tool in two forms: on-premises, operated on the customer's own servers, and SaaS, accessed through the cloud or online. This allows use in ordinary businesses and research institutes as well as environments requiring high security, such as defense and space science.
Hamn completed patent registrations related to semantic data and ontology before building the tool. Although still at an early stage, he is planning validation projects with South Korean defense companies and research institutes. Worflogy plans to establish technological barriers, complete validation by August, and then officially launch its enterprise ontology tool.
Worflogy is working to improve and broaden adoption of the tool. It aims to improve the UI and UX so that users can configure semantic data and build ontologies easily, lowering entry barriers for a wider range of businesses and research institutes. It is also raising investment. Building on its registered patents, it plans to obtain certification as an innovative technology venture and participate in national R&D projects to demonstrate ontology's value.
Hamn said, "Semantic data and ontology can connect a company's knowledge and data in coherent contexts, turning them into irreplaceable knowledge assets. We aim to implement these technologies accurately, apply them usefully, and help South Korean businesses and research institutes grow."
Cha Joo-kyung, IT Donga (racingcar@itdonga.com)
Copyright IT Donga. Unauthorized reproduction and redistribution prohibited.
This archive preserves a full backup of the article for the corporate website redesign and to prevent loss. The original is accessible through the source links on the company timeline page.