Knowledge integration is the process of synthesizing multiple knowledge models or representations into a common model (representation). Unlike information integration, which merges data with different schemas, knowledge integration emphasizes combining understanding from various perspectives on a given subject. For example, a set of student grades can be interpreted in multiple ways; an integrated view might use a common model like a student performance index to unify these interpretations.
This concept has been studied in fields such as education, cognitive science, and artificial intelligence. In education, the Web-based Inquiry Science Environment (WISE), developed at the University of California at Berkeley, applies knowledge integration theory to help students connect scientific ideas. In AI, knowledge integration is crucial for building systems that learn from new information and update existing knowledge bases.
Theoretical Foundations
Knowledge integration involves incorporating new information into existing knowledge through an interdisciplinary approach. This process requires determining how new and existing knowledge interact, how existing knowledge should be modified to accommodate new information, and how new information should be adapted in light of prior understanding. A learning agent that actively investigates the consequences of new information can detect and exploit learning opportunities, such as resolving knowledge conflicts or filling gaps, enabling it to learn beyond the explicit content of the new information.
Applications in Artificial Intelligence
In Artificial intelligence, knowledge integration is used to combine outputs from multiple models or sources. For instance, Large language models may integrate knowledge from different training corpora or user interactions. Techniques like semantic matching help align representations, while minimal mappings reduce effort in validation and visualization. The machine learning program KI, developed by Murray and Porter at the University of Texas at Austin, was created to study automated and semi-automated knowledge integration to assist knowledge engineers in constructing large knowledge bases.
Educational Perspectives
The knowledge integration perspective in learning emphasizes helping students build coherent understanding by connecting ideas across contexts. WISE supports this by providing inquiry-based activities that prompt students to articulate and refine their ideas. Research by Linn and others has shown that this approach improves students' ability to reason about complex scientific phenomena. The University of Waterloo offers a Bachelor of Knowledge Integration program, started in 2008, which teaches students to synthesize knowledge across disciplines.
Techniques and Challenges
A key challenge is mapping between different knowledge representations. Semantic matching identifies correspondences between concepts, while minimal mappings provide high-quality mappings from which all others can be derived in linear time. These techniques help automate the integration process, but human oversight is often needed to ensure accuracy. In Machine learning, knowledge integration can involve combining predictions from multiple models, such as in ensemble methods, or updating a model with new data without catastrophic forgetting.
Future Directions
As AI systems become more complex, knowledge integration will play a vital role in enabling them to learn continuously and reason across domains. Research in Deep learning and Neural networks explores ways to integrate knowledge from different modalities or tasks. The development of Transformer (architecture)-based models has also highlighted the need for integrating knowledge from diverse sources to improve performance and reliability.