Learning knowledge graph makes AI a better decision support tool

More data is being generated in companies and organizations than ever before, but the sheer volume of information does not automatically lead to better decisions. Anna Teern, a doctoral researcher at the University of Oulu, proposes an evolvable knowledge graph as a solution that learns from machine-learning analyses and human interaction.

AI-based assistants have rapidly become more common in fields such as industry, transportation, and property management. They can analyze large volumes of data and identify phenomena that would be difficult for humans to detect. In practice, however, their usefulness depends on whether they can provide relevant and appropriate information for a given operating environment. Humans use their own judgment and apply the suggestions they receive to the situation at hand.

“Data alone is not sufficient to support decision-making in complex situations. Human experience, expertise, and an understanding of what information is relevant at a given moment are also needed,” says doctoral researcher Anna Teern from the University of Oulu.

Collaboration between humans and AI

The research concerns hybrid intelligence, which combines the strengths of humans and artificial intelligence. AI can process vast amounts of data quickly, while humans can evaluate situations, goals, and their significance within a broader context.

In many current systems, however, these strengths remain separate. Data moves through systems, but experts’ tacit knowledge, experience, and interpretations do not change the knowledge within AI or the way instructions are given to humans. As a result, intelligent assistants may provide users with plenty of information but little real support for decision-making.

The learning knowledge graph serves as a shared knowledge base for both humans and machines. It represents information and relationships between entities in a form that can be utilised by both experts and AI systems.

“The key innovation of my research is a knowledge graph that continuously learns. It is not merely a static database; it can be enriched with new data, machine-learning results, and expert feedback. When the system encounters new situations, the knowledge graph can be refined and developed further. As a result, the assistant that utilizes it also becomes better at supporting users over time,” summarizes doctoral researcher Anna Teern.

The solution is particularly needed in environments where decisions must be based on both data and human interpretation. In companies, information is often scattered across information systems, documents, and employees’ experience. A learning knowledge graph helps bring this information together to support decision-making.

Applications From Maintenance to Autonomous Vehicles

The doctoral research was conducted as design science research, in which solutions were developed and evaluated through practical application cases. The study first examined the design principles, development process, and learning mechanisms of learning knowledge graphs. The solution was then applied in connection with an intelligent assistant to evaluate its functionality in practice.

The empirical research settings focused on industrial maintenance and decision-making in autonomous driving. However, the results are not limited to these fields but can be widely utilized in the development of future intelligent assistants.

“The main conclusion of the study is that the value of intelligent assistants arises from their ability to learn both from their operating environment and from human expertise. When the knowledge base evolves continuously, AI does not remain an isolated analytical tool but can become genuinely useful support for decision-making. At the same time, the expert remains at the centre of the process,” summarises professor Tero Päivärinta, who supervised the dissertation.

A learning knowledge graph does not make decisions on behalf of humans. Its purpose is to provide the right information at the right time and to make experts’ knowledge more visible and easier to utilize.

As part of her doctoral research, Teern worked on the 6G Visible research project conducted by the University of Oulu and the Finnish Meteorological Institute. The project developed new solutions that combined 6G technology, artificial intelligence, and distributed computing into a service for autonomous mobility in future 6G networks.

The project developed a service to support route planning, particularly in challenging weather conditions. The solution moved beyond relying solely on vehicle sensors and static maps toward a broader situational awareness that also accounted for weather, traffic, and human experience.

“The key innovation was the utilization of so-called ‘invisible information’: the system combined data from multiple sources in the cloud, such as weather observations, radar forecasts, road information, and traffic data, into a single continuously updated situational picture. This information was managed through an evolvable knowledge graph whose goal is to learn over time from both humans and machine learning methods, thereby improving decision-making,” says Anna Teern.

The hybrid-intelligent autonomous driving system (HI-ADS) developed in the solution combined the computational and predictive capabilities of an information system with human judgment and contextual understanding. This made it possible for route planning to take into account not only weather and traffic information but also driver preferences, experience, and real-time feedback.

Master of Arts Anna Teern will defend her doctoral dissertation at the University of Oulu on Thursday, 1 October. The title of the dissertation in Information Processing Science is Design of Learning Knowledge Graphs. The opponent will be Professor Polyxeni Vassilakopoulou (University of Agder), and the custos will be Professor Tero Päivärinta (University of Oulu).